fix(pr_merge): send Gitea's merge_title_field (gitea-mcp#77) (#82 )

Co-authored-by: Mathias <mthbqv@gmail.com>
chore(context): restore the root rules in AGENTS.md and system-prompt.txt (#80 ) (#81 )
2026-10-10 08:50:06 +00:00 · 2026-10-09 06:02:43 +00:00 · 2026-10-08 08:34:50 +02:00 · 2026-10-08 06:18:07 +00:00 · 2026-10-08 06:02:34 +00:00 · 2026-10-05 09:33:44 +02:00
24 changed files with 1791 additions and 155 deletions
@@ -37,9 +37,9 @@

 ### Git
 - Conventional commits: `feat:`, `fix:`, `chore:`, `docs:`, `refactor:`
- **Trunk-Based Development:** commit directly to main. One logical change per commit.
+- **Trunk-Based Development:** one logical change per commit, merged to main by PR (`main` is protected, see below).
 - Run `task check` locally before every push. CI is the quality gate, not branch protection.
- No feature branches, no PRs for solo/agent work.
+- Short-lived branches and PRs are how work reaches `main` here (`tbd_ship`/`pr_merge`); no long-lived feature branches.
 - Exception: if a parallel agent session is active on this repo, use a short-lived
  `agent/<description>` branch and merge within the same session.

@@ -79,8 +79,8 @@ When acting as a coding agent on this project:
 3. If unsure about a convention, check `DECISIONS.md` or ask
 4. Never modify files outside the project root without explicit permission
 5. When adding a dependency, explain why in the commit message
-6. Commit directly to main. Run `task check` before every push. Never create
-   feature branches unless a parallel agent is simultaneously active on this repo.
+6. Ship by PR (`main` is protected, see below). Run `task check` before every push. Never create
+   long-lived feature branches.
 7. For client projects: never send code or context to cloud APIs — use local models via LiteLLM

 ## Current state — v0.2.5 (2026-05-17)
@@ -105,3 +105,15 @@ Current main: `e31fd3f`. CI green. Deployed via Flux.

 2. **Issue #19** — end-to-end mirror flow verification.
   `repo_mirror_push` is implemented but the full flow (create repo → add push mirror → verify sync to GitHub) has not been tested manually. Do this before relying on it in production.
+
+## `main` is protected: ship by PR (local-dev#42, decided 2026-10-08)
+
+Gitea refuses direct pushes to `main` in this repo, and a PR merges only after its CI status (`CD / Lint / Test / Vet (pull_request)`) passes.
+**Why:** local-dev rule 5's reviewed lane (risk/medium+ is reviewed by a human) was enforced on the merge tools
+(`tbd_ship`, `pr_merge`: gitea-mcp v0.17.0/v0.18.0, merge only for an issue labelled exactly `risk/low` with green CI)
+but not on a direct commit to `main`, which bypassed it. This overrides rule 5's "risk/low commits straight to main"
+for this repo; control code is where the lane matters.
+**How:** `tbd_ship` for a one-file change, or branch -> PR -> `pr_merge` (name the `risk/low` issue); above `risk/low`
+open the PR and a human merges it in the Gitea UI after review. Tags are not protected. CD jobs push image tags to
+the **infra** repo with deploy keys, never to this repo's `main`. Undo (Mathias only): Settings -> Branches, or
+`PATCH /repos/mathias/gitea-mcp/branch_protections/main {"enable_push": true}`.
@@ -78,6 +78,27 @@ These rules apply to every task across every project, regardless of harness.
   when a human reviewer outside the project is required. Document the reason in
   PROJECT.md.

+   **Exception — a GitOps repo where a branch cannot be reconciled.** In a repo whose
+   deployment watches one branch (`mathias/infra`: Flux watches `main`), a branch is
+   invisible to the thing that applies it. A `risk/medium` change to `alerts.yaml` or
+   `litellm-config.yaml` can be linted on a branch but never *exercised* there, so a
+   PR would buy a review of a state that was never deployed.
+
+   There, `risk/medium`+ work stays on `main` and instead requires a **recorded
+   adversarial review pass before the push** — `skills:pr-gatekeeper` or equivalent,
+   reviewing the staged diff. **The reviewer's verdict goes in the issue, not a PR
+   thread.** No verdict recorded means the lane was not used, and saying "the gate was
+   green" is not the same claim.
+
+   Be honest about what this is: a recorded *independent* pass, not four-eyes in the
+   sense `assessor-loop` §4.1 means for a regulated organisation, because the reviewer
+   is an agent. It is more than the repo had and less than the model it mirrors; do not
+   describe it as the stronger property. (infra#470, decided 2026-09-29.)
+
+   This lane depends on the `risk/` label being right. A label that triggers a heavier
+   lane is load-bearing, not cosmetic — if work that belongs in it is unlabelled, the
+   lane silently does not apply.
+
 6. **Close the loop — every substantive task ends with the same ritual.** Shipping
   the code is not the end of the task; capturing it is. Run this unprompted:
   - **Tag + bump SemVer** on the change (annotated tag; minor for a feature or
@@ -106,6 +127,100 @@ These rules apply to every task across every project, regardless of harness.
   revisit Y". Those are exactly the items that need a locator. If it is worth
   saying in a summary, it is worth writing where it can be found again without you.

+7. **Finish what's closable — only involve me for a genuine call.** If a
+   tracker issue, PR, or checklist item is demonstrably resolved — the fix
+   landed, the state changed, you verified it live — close/merge/check it off
+   in the same turn. Don't leave it open "for the record" or park it waiting
+   on me to confirm something that already happened; that's bookkeeping, not
+   a decision. Reserve asking me (`AskUserQuestion`, a `DECISION NEEDED`
+   comment, or otherwise pausing) for when it's genuinely mine to call:
+   irreversible or prod/host-impacting actions, a real judgment call with no
+   clearly-better option, ambiguous intent, or anything secret-writes/
+   destructive-git already gate. Before claiming something is closable,
+   verify it live (re-check the issue, re-run the command) rather than
+   trusting your own earlier summary (brain: `filed-issue-not-same-as-fixed`).
+
+8. **Control code needs an adversary, not a checklist.** Code whose job is to
+   refuse, deny, gate or contain is not adequately verified by a green gate.
+   Rule 5 makes CI the quality gate and that holds for ordinary work — but a
+   passing suite only asserts what its author already thought of, and for a
+   control the author is precisely the person whose imagination just failed.
+   Measured on `dispatch#75`: five review rounds, five real escapes verified by
+   execution, and **CI was green on every one of them** while `git rm
+   ././.dispatch-allow` deleted the file the control existed to protect. So
+   route this work through a review with authority to block, instructed to
+   verify by execution rather than by reading the implementation. Expect to be
+   blocked more than once; that is the control working, not the reviewer being
+   difficult. (brain: `a-green-gate-is-not-evidence-for-control-code`; the
+   practices live in the `pr-gatekeeper` agent.)
+
+9. **A test's name must not claim more than its assertions prove.** Name a test
+   after the mechanism it observes, not the goal it serves.
+   `DoesNotInheritEnv` is honest; `IsContained` is a claim the test cannot
+   support. This is not pedantry about naming — when name and assertion
+   diverge, the name wins in the reader's head and the gap becomes invisible,
+   which is what lets the defect survive review. Found three times in one PR,
+   twice in tests written to fix the previous instance of it; each passed while
+   the thing it was named for was exploitable. If you cannot name the mechanism,
+   you do not yet know what the test observes.
+
+10. **Short sessions, one rc per repo, edit only in a locked worktree of the repo
+   you are changing** (ADR-0002, accepted 2026-10-06).
+   - One session per issue or task, ending with rule 6's close ritual. Related
+     issues may share a session only when they share a worktree and an owner.
+     State that must outlive the session goes in an issue, not the chat.
+   - Start `claude rc` in the repo you will edit (`infra`, `trucker-ecommerce`,
+     `hyperguild`, ...), not in `~/dev`: the rc worktree is a worktree of the
+     launch directory's repo, so launched from `~/dev` it protects nothing you
+     edit (measured: 73% of edits landed in shared checkouts). Start it only
+     while working there.
+   - Keep at most one thin `~/dev` coordinator session. It files issues and hands
+     out tasks; it reads and edits no sub-repo code.
+   - Split cross-repo work per repo. The contract between the halves lives in an
+     issue on each repo.
+   - Edit in a worktree of that repo, locked at creation:
+     `scripts/wt.sh add <repo> <task>` does it (fetch, `wt/<task>` from origin/main, lock);
+     `scripts/wt.sh done <repo>-<task>` removes it after the push and refuses dirty or
+     unpushed work. By hand: `git worktree add ~/worktrees/<name> -b <branch> origin/main`
+     then `git worktree lock --reason "manual: <purpose>" ~/worktrees/<name>`.
+   - Before working a repo another session may touch, announce your issue list to
+     that session (`SendMessage`) and ask for theirs.
+   - Auto-memory is keyed by cwd, so a repo rc starts without it. Cross-repo
+     working rules live in brain (`wiki/homelab/decisions/2026-10-06-062416-auto.md`, wing homelab, hall
+     decisions) and the section below, not in per-directory memory.
+
+## Standing footguns
+
+Each cost real time once. Detail and evidence: brain `wiki/homelab/decisions/2026-10-06-062416-auto.md` (title standing-working-rules).
+
+- **Gate then commit with `&&`, never `;`.** "Green" means exit code 0 and
+  `grep -c FAIL` is 0, not a filtered grep.
+- **Stage by path, never `git add -A` or `git add .`** (they stage from the repo
+  root and swept in other sessions' work twice). Check `git status --porcelain` first.
+- **Before committing in a shared checkout:** `git branch --show-current`.
+- **Before tagging:** `git fetch origin --tags`, then `git describe --tags --abbrev=0 <sha>`
+  and `git tag --sort=-v:refname | head`; go above both. Never take the last tag
+  from a report.
+- **Run only what the project's gate runs.** No repo-wide formatter unless the gate
+  runs it.
+- **Reproduce a CI red at the cited commit before applying an issue's prescribed fix.**
+- **Read an issue's comments (`issue_list_comments`), not just its body, before working it.**
+- **A comment naming a control ("behind branch protection") is not evidence; query
+  the control.** A working detector can hide a missing preventer.
+- **On koala the MCP servers are `mcp__plugin_skills_{brain,gitea,infra}__*`.** The
+  per-turn auth reminder is not evidence; call the tool or run `claude mcp list`.
+  Brain writes always pass `wing` + `hall`.
+- **`cat /etc/hostname` before assuming which host you are on.**
+- **Never run bare `git remote -v` or `git config --list`** (url.insteadOf expands
+  into a live token); use `git config --get remote.origin.url`.
+- **On koala `sudo` is denied outright; k8s exec, env dumps and flux reconcile need
+  per-action approval; decrypting a SOPS file and printing in one Bash call is
+  hook-blocked.** Do the repo-side work and hand the host-side step over in an issue.
+- **`set -o pipefail` plus an optional `grep | head | awk` dies on no match:** append `|| true`.
+- **Gitea `file_write_branch` takes plain text, not base64.**
+- **`OP_SERVICE_ACCOUNT_TOKEN` in the shell is read-only for `op`.** Operational
+  secrets go in the AgentSecrets vault via `set -a; . ~/.op-agent-write.env; set +a`.
+
 ## CAD sprint patterns

 Two CAD sprint archetypes (defs: brain `wiki/homelab/decisions/continuous-agentic-development-cad-concept-2026-06-16`):
@@ -186,25 +301,20 @@ Three machines on Tailscale:

 All development repos live at `~/dev/` (softlink from `~/Documents/local-dev/`).

-Organized in thematic folders:
+Organized by **role**, because the role decides which rules apply (ADR-0003):

-| Folder | Focus | Count |
-|--------|-------|-------|
-| `GO/` | Go web frameworks, API integrations, learning projects | ~10 |
-| `AI/` | ML research, AI frameworks (FinRL, DSPy, crawl4ai) | ~6 |
-| `AGENTS/` | Autonomous agents, coding agents, MCP servers, infra | ~15 |
-| `QKX/` | Invoice processing, financial automation, payment systems | ~13 |
-| `XT/` | Climate data, sustainability (Klimatkollen, Garbo) | ~2 |
+| Folder | Role | Rules that key off it |
+|--------|------|-----------------------|
+| `platform/` | infra, brain, hyperguild, gitea-mcp, dispatch, shared tooling | control code (rule 8), GitOps review lane |
+| `products/` | tapir, cobalt-dingo, openbanking-api, ... | ordinary trunk-based work |
+| `clients/` | client work | local models only, never cloud APIs (Client work rules) |
+| `spikes/` | throwaway | no ceremony |

-See `~/dev/PROJECT_SUMMARY.md` for detailed descriptions of each project.
-
-### Key active projects
-
- **super-koala** (`AGENTS/`) — multi-component agent stack with LangGraph, DSPy, MCP
- **azure-tiger** (`QKX/`) — invoice extraction → ISO 20022 payment instructions
- **gocrwl** (`AGENTS/`) — Go web crawler with containerized deployment
- **koala-ai-stack** (`AGENTS/`) — local AI server infrastructure management
- **klimatkollen** (`XT/`) — Swedish municipal climate data platform
+New projects are born into a role folder (`task new-project -- <name> <client|personal>
+<role-folder>`); a client tag requires `clients/` and `clients/` requires a client tag.
+Existing repos still sit in the old `AGENTS/` and `AI/` folders until they are moved one by
+one (ADR-0003 step 3). `~/dev/PROJECT_SUMMARY.md` (generated) lists every repo and where it
+is. How a session runs on koala (worktrees, rc per repo, the reaper, memory): `docs/working-on-koala.md`.

 ## Knowledge base — actively use it

@@ -246,10 +356,16 @@ entries that age well are about *why*, *how to avoid*, and *what to do when*.
 |---------|-------|-------|
 | **Claude Code, Claude Desktop** | `brain_query` (BM25), `brain_answer` (LLM-synth + sources) MCP tools | `brain_write` MCP tool |
 | **Crush, Pi, Antigravity, other MCP-capable** | same MCP server: `ingestion-brain` (via the `mcp__*_brain__*` namespace once authenticated) | same |
-| **Anything HTTP-only (curl, scripts)** | `POST https://brain-mcp.d-ma.be/query` with `{"query":"..."}` (auth via `BRAIN_MCP_TOKEN`) | `POST .../write` with `{"content":"...","filename":"..."}` |
+| **Anything HTTP-only (curl, scripts)** | Tailnet only: `POST https://koala.tail30472e.ts.net:8443/query` with `{"query":"..."}`, sending `BRAIN_MCP_TOKEN` (not yet enforced on REST, hyperguild#149). The public `brain-mcp.d-ma.be` serves `/mcp` only (infra#517). | `POST .../write` with `{"content":"...","wing":"...","hall":"..."}` |
 | **Browser / human inspection** | `https://git.d-ma.be/mathias/hyperguild` → `knowledge/` and `wiki/` markdown files |

- **Scoping**: defaults to `public` collection; client projects filter to `{client}` + `public`.
+- **Scoping — the brain holds no confidential client content.** Nothing on the read path
+  filters by client: every token holder can read every wing, and `wing` on `brain_query`
+  is an optional argument, not an access control. Client confidentiality is enforced only
+  at write time (`capture`'s I1 gate refuses internal/confidential). So never `brain_write`
+  client material; generalise it first or keep it in the client's own repo. (hyperguild#144,
+  decided 2026-10-01. This line previously claimed client projects were filtered; they
+  were not.)
 - **Routing**: brain_answer's LLM uses berget.ai as primary, iguana ollama as
  fallback. Both are configurable in the `supervisor/ingestion-deployment.yaml`
  on the koala k3s cluster; don't hardcode local-only model names into the
@@ -319,8 +435,20 @@ Workflow:
   and missing-untracked adapters). A drift fails the check with a
   message telling you to stage the regenerated files.

-Behavior rules in this file and per-project rules in `PROJECT.md` apply
-unconditionally on every host, every harness.
+Behavior rules in this file and per-project rules in `PROJECT.md` apply on every
+host and every harness.
+
+**Where a project rule contradicts a rule here, the project rule governs — and it
+must state its reason.** Both halves matter: the project file wins because it knows
+something this file cannot (infra's Flux constraint is real and not visible from
+here), and it must say why because a bare override is indistinguishable from drift.
+
+This replaces the previous wording, which said rules here and in `PROJECT.md` both
+apply *"unconditionally"*. That was not a tiebreak but a contradiction, and it went
+unnoticed until rule 5 and infra's "never create feature branches" had to be obeyed
+at the same time (infra#470). A standard can hold two individually sensible clauses
+that are jointly unsatisfiable, and survive for as long as everyone reads it
+charitably — the same failure the Inception Sprint Oath's G6 had.

 ## Engineering Skills

@@ -390,9 +518,9 @@ Check a host with `task doctor` in the skills repo; it also runs automatically a

 ### Git
 - Conventional commits: `feat:`, `fix:`, `chore:`, `docs:`, `refactor:`
- **Trunk-Based Development:** commit directly to main. One logical change per commit.
+- **Trunk-Based Development:** one logical change per commit, merged to main by PR (`main` is protected, see below).
 - Run `task check` locally before every push. CI is the quality gate, not branch protection.
- No feature branches, no PRs for solo/agent work.
+- Short-lived branches and PRs are how work reaches `main` here (`tbd_ship`/`pr_merge`); no long-lived feature branches.
 - Exception: if a parallel agent session is active on this repo, use a short-lived
  `agent/<description>` branch and merge within the same session.

@@ -432,8 +560,8 @@ When acting as a coding agent on this project:
 3. If unsure about a convention, check `DECISIONS.md` or ask
 4. Never modify files outside the project root without explicit permission
 5. When adding a dependency, explain why in the commit message
-6. Commit directly to main. Run `task check` before every push. Never create
-   feature branches unless a parallel agent is simultaneously active on this repo.
+6. Ship by PR (`main` is protected, see below). Run `task check` before every push. Never create
+   long-lived feature branches.
 7. For client projects: never send code or context to cloud APIs — use local models via LiteLLM

 ## Current state — v0.2.5 (2026-05-17)
@@ -459,4 +587,16 @@ Current main: `e31fd3f`. CI green. Deployed via Flux.
 2. **Issue #19** — end-to-end mirror flow verification.
   `repo_mirror_push` is implemented but the full flow (create repo → add push mirror → verify sync to GitHub) has not been tested manually. Do this before relying on it in production.

+## `main` is protected: ship by PR (local-dev#42, decided 2026-10-08)
+
+Gitea refuses direct pushes to `main` in this repo, and a PR merges only after its CI status (`CD / Lint / Test / Vet (pull_request)`) passes.
+**Why:** local-dev rule 5's reviewed lane (risk/medium+ is reviewed by a human) was enforced on the merge tools
+(`tbd_ship`, `pr_merge`: gitea-mcp v0.17.0/v0.18.0, merge only for an issue labelled exactly `risk/low` with green CI)
+but not on a direct commit to `main`, which bypassed it. This overrides rule 5's "risk/low commits straight to main"
+for this repo; control code is where the lane matters.
+**How:** `tbd_ship` for a one-file change, or branch -> PR -> `pr_merge` (name the `risk/low` issue); above `risk/low`
+open the PR and a human merges it in the Gitea UI after review. Tags are not protected. CD jobs push image tags to
+the **infra** repo with deploy keys, never to this repo's `main`. Undo (Mathias only): Settings -> Branches, or
+`PATCH /repos/mathias/gitea-mcp/branch_protections/main {"enable_push": true}`.
+
 ---
@@ -8,6 +8,12 @@ name: CD
    # (Only applies to branch pushes; tag pushes always trigger.)
    paths-ignore:
      - 'docs/**'
+  # gitea-mcp#68: without this a PR never got a CI run, so the risk/medium+
+  # PR lane merged on local `task check` alone. Only the check job runs for a
+  # PR -- build and deploy stay push/tag-only (no images, no infra writes,
+  # no secrets from a PR).
+  pull_request:
+    branches: [main]

 env:
  IMAGE: gitea-mcp
@@ -44,6 +50,7 @@ jobs:
  build:
    name: Build & Import
    needs: check
+    if: github.event_name == 'push'
    runs-on: self-hosted
    outputs:
      image-tag: ${{ steps.meta.outputs.sha-tag }}
@@ -52,10 +59,12 @@ jobs:

      - name: Derive image tags
        id: meta
+        env:
+          # signals#47: a ref name can carry $(...); pass it via env, not ${{ }}.
+          REF: ${{ github.ref }}
        run: |
          SHA=$(git rev-parse --short HEAD)
          echo "sha-tag=${SHA}" >> "$GITHUB_OUTPUT"
-          REF="${{ github.ref }}"
          if [[ "$REF" == refs/tags/v* ]]; then
            echo "version-tag=${REF#refs/tags/}" >> "$GITHUB_OUTPUT"
            echo "version=${REF#refs/tags/}" >> "$GITHUB_OUTPUT"
@@ -73,6 +73,27 @@ These rules apply to every task across every project, regardless of harness.
   when a human reviewer outside the project is required. Document the reason in
   PROJECT.md.

+   **Exception — a GitOps repo where a branch cannot be reconciled.** In a repo whose
+   deployment watches one branch (`mathias/infra`: Flux watches `main`), a branch is
+   invisible to the thing that applies it. A `risk/medium` change to `alerts.yaml` or
+   `litellm-config.yaml` can be linted on a branch but never *exercised* there, so a
+   PR would buy a review of a state that was never deployed.
+
+   There, `risk/medium`+ work stays on `main` and instead requires a **recorded
+   adversarial review pass before the push** — `skills:pr-gatekeeper` or equivalent,
+   reviewing the staged diff. **The reviewer's verdict goes in the issue, not a PR
+   thread.** No verdict recorded means the lane was not used, and saying "the gate was
+   green" is not the same claim.
+
+   Be honest about what this is: a recorded *independent* pass, not four-eyes in the
+   sense `assessor-loop` §4.1 means for a regulated organisation, because the reviewer
+   is an agent. It is more than the repo had and less than the model it mirrors; do not
+   describe it as the stronger property. (infra#470, decided 2026-09-29.)
+
+   This lane depends on the `risk/` label being right. A label that triggers a heavier
+   lane is load-bearing, not cosmetic — if work that belongs in it is unlabelled, the
+   lane silently does not apply.
+
 6. **Close the loop — every substantive task ends with the same ritual.** Shipping
   the code is not the end of the task; capturing it is. Run this unprompted:
   - **Tag + bump SemVer** on the change (annotated tag; minor for a feature or
@@ -101,6 +122,100 @@ These rules apply to every task across every project, regardless of harness.
   revisit Y". Those are exactly the items that need a locator. If it is worth
   saying in a summary, it is worth writing where it can be found again without you.

+7. **Finish what's closable — only involve me for a genuine call.** If a
+   tracker issue, PR, or checklist item is demonstrably resolved — the fix
+   landed, the state changed, you verified it live — close/merge/check it off
+   in the same turn. Don't leave it open "for the record" or park it waiting
+   on me to confirm something that already happened; that's bookkeeping, not
+   a decision. Reserve asking me (`AskUserQuestion`, a `DECISION NEEDED`
+   comment, or otherwise pausing) for when it's genuinely mine to call:
+   irreversible or prod/host-impacting actions, a real judgment call with no
+   clearly-better option, ambiguous intent, or anything secret-writes/
+   destructive-git already gate. Before claiming something is closable,
+   verify it live (re-check the issue, re-run the command) rather than
+   trusting your own earlier summary (brain: `filed-issue-not-same-as-fixed`).
+
+8. **Control code needs an adversary, not a checklist.** Code whose job is to
+   refuse, deny, gate or contain is not adequately verified by a green gate.
+   Rule 5 makes CI the quality gate and that holds for ordinary work — but a
+   passing suite only asserts what its author already thought of, and for a
+   control the author is precisely the person whose imagination just failed.
+   Measured on `dispatch#75`: five review rounds, five real escapes verified by
+   execution, and **CI was green on every one of them** while `git rm
+   ././.dispatch-allow` deleted the file the control existed to protect. So
+   route this work through a review with authority to block, instructed to
+   verify by execution rather than by reading the implementation. Expect to be
+   blocked more than once; that is the control working, not the reviewer being
+   difficult. (brain: `a-green-gate-is-not-evidence-for-control-code`; the
+   practices live in the `pr-gatekeeper` agent.)
+
+9. **A test's name must not claim more than its assertions prove.** Name a test
+   after the mechanism it observes, not the goal it serves.
+   `DoesNotInheritEnv` is honest; `IsContained` is a claim the test cannot
+   support. This is not pedantry about naming — when name and assertion
+   diverge, the name wins in the reader's head and the gap becomes invisible,
+   which is what lets the defect survive review. Found three times in one PR,
+   twice in tests written to fix the previous instance of it; each passed while
+   the thing it was named for was exploitable. If you cannot name the mechanism,
+   you do not yet know what the test observes.
+
+10. **Short sessions, one rc per repo, edit only in a locked worktree of the repo
+   you are changing** (ADR-0002, accepted 2026-10-06).
+   - One session per issue or task, ending with rule 6's close ritual. Related
+     issues may share a session only when they share a worktree and an owner.
+     State that must outlive the session goes in an issue, not the chat.
+   - Start `claude rc` in the repo you will edit (`infra`, `trucker-ecommerce`,
+     `hyperguild`, ...), not in `~/dev`: the rc worktree is a worktree of the
+     launch directory's repo, so launched from `~/dev` it protects nothing you
+     edit (measured: 73% of edits landed in shared checkouts). Start it only
+     while working there.
+   - Keep at most one thin `~/dev` coordinator session. It files issues and hands
+     out tasks; it reads and edits no sub-repo code.
+   - Split cross-repo work per repo. The contract between the halves lives in an
+     issue on each repo.
+   - Edit in a worktree of that repo, locked at creation:
+     `scripts/wt.sh add <repo> <task>` does it (fetch, `wt/<task>` from origin/main, lock);
+     `scripts/wt.sh done <repo>-<task>` removes it after the push and refuses dirty or
+     unpushed work. By hand: `git worktree add ~/worktrees/<name> -b <branch> origin/main`
+     then `git worktree lock --reason "manual: <purpose>" ~/worktrees/<name>`.
+   - Before working a repo another session may touch, announce your issue list to
+     that session (`SendMessage`) and ask for theirs.
+   - Auto-memory is keyed by cwd, so a repo rc starts without it. Cross-repo
+     working rules live in brain (`wiki/homelab/decisions/2026-10-06-062416-auto.md`, wing homelab, hall
+     decisions) and the section below, not in per-directory memory.
+
+## Standing footguns
+
+Each cost real time once. Detail and evidence: brain `wiki/homelab/decisions/2026-10-06-062416-auto.md` (title standing-working-rules).
+
+- **Gate then commit with `&&`, never `;`.** "Green" means exit code 0 and
+  `grep -c FAIL` is 0, not a filtered grep.
+- **Stage by path, never `git add -A` or `git add .`** (they stage from the repo
+  root and swept in other sessions' work twice). Check `git status --porcelain` first.
+- **Before committing in a shared checkout:** `git branch --show-current`.
+- **Before tagging:** `git fetch origin --tags`, then `git describe --tags --abbrev=0 <sha>`
+  and `git tag --sort=-v:refname | head`; go above both. Never take the last tag
+  from a report.
+- **Run only what the project's gate runs.** No repo-wide formatter unless the gate
+  runs it.
+- **Reproduce a CI red at the cited commit before applying an issue's prescribed fix.**
+- **Read an issue's comments (`issue_list_comments`), not just its body, before working it.**
+- **A comment naming a control ("behind branch protection") is not evidence; query
+  the control.** A working detector can hide a missing preventer.
+- **On koala the MCP servers are `mcp__plugin_skills_{brain,gitea,infra}__*`.** The
+  per-turn auth reminder is not evidence; call the tool or run `claude mcp list`.
+  Brain writes always pass `wing` + `hall`.
+- **`cat /etc/hostname` before assuming which host you are on.**
+- **Never run bare `git remote -v` or `git config --list`** (url.insteadOf expands
+  into a live token); use `git config --get remote.origin.url`.
+- **On koala `sudo` is denied outright; k8s exec, env dumps and flux reconcile need
+  per-action approval; decrypting a SOPS file and printing in one Bash call is
+  hook-blocked.** Do the repo-side work and hand the host-side step over in an issue.
+- **`set -o pipefail` plus an optional `grep | head | awk` dies on no match:** append `|| true`.
+- **Gitea `file_write_branch` takes plain text, not base64.**
+- **`OP_SERVICE_ACCOUNT_TOKEN` in the shell is read-only for `op`.** Operational
+  secrets go in the AgentSecrets vault via `set -a; . ~/.op-agent-write.env; set +a`.
+
 ## CAD sprint patterns

 Two CAD sprint archetypes (defs: brain `wiki/homelab/decisions/continuous-agentic-development-cad-concept-2026-06-16`):
@@ -181,25 +296,20 @@ Three machines on Tailscale:

 All development repos live at `~/dev/` (softlink from `~/Documents/local-dev/`).

-Organized in thematic folders:
+Organized by **role**, because the role decides which rules apply (ADR-0003):

-| Folder | Focus | Count |
-|--------|-------|-------|
-| `GO/` | Go web frameworks, API integrations, learning projects | ~10 |
-| `AI/` | ML research, AI frameworks (FinRL, DSPy, crawl4ai) | ~6 |
-| `AGENTS/` | Autonomous agents, coding agents, MCP servers, infra | ~15 |
-| `QKX/` | Invoice processing, financial automation, payment systems | ~13 |
-| `XT/` | Climate data, sustainability (Klimatkollen, Garbo) | ~2 |
+| Folder | Role | Rules that key off it |
+|--------|------|-----------------------|
+| `platform/` | infra, brain, hyperguild, gitea-mcp, dispatch, shared tooling | control code (rule 8), GitOps review lane |
+| `products/` | tapir, cobalt-dingo, openbanking-api, ... | ordinary trunk-based work |
+| `clients/` | client work | local models only, never cloud APIs (Client work rules) |
+| `spikes/` | throwaway | no ceremony |

-See `~/dev/PROJECT_SUMMARY.md` for detailed descriptions of each project.
-
-### Key active projects
-
- **super-koala** (`AGENTS/`) — multi-component agent stack with LangGraph, DSPy, MCP
- **azure-tiger** (`QKX/`) — invoice extraction → ISO 20022 payment instructions
- **gocrwl** (`AGENTS/`) — Go web crawler with containerized deployment
- **koala-ai-stack** (`AGENTS/`) — local AI server infrastructure management
- **klimatkollen** (`XT/`) — Swedish municipal climate data platform
+New projects are born into a role folder (`task new-project -- <name> <client|personal>
+<role-folder>`); a client tag requires `clients/` and `clients/` requires a client tag.
+Existing repos still sit in the old `AGENTS/` and `AI/` folders until they are moved one by
+one (ADR-0003 step 3). `~/dev/PROJECT_SUMMARY.md` (generated) lists every repo and where it
+is. How a session runs on koala (worktrees, rc per repo, the reaper, memory): `docs/working-on-koala.md`.

 ## Knowledge base — actively use it

@@ -241,10 +351,16 @@ entries that age well are about *why*, *how to avoid*, and *what to do when*.
 |---------|-------|-------|
 | **Claude Code, Claude Desktop** | `brain_query` (BM25), `brain_answer` (LLM-synth + sources) MCP tools | `brain_write` MCP tool |
 | **Crush, Pi, Antigravity, other MCP-capable** | same MCP server: `ingestion-brain` (via the `mcp__*_brain__*` namespace once authenticated) | same |
-| **Anything HTTP-only (curl, scripts)** | `POST https://brain-mcp.d-ma.be/query` with `{"query":"..."}` (auth via `BRAIN_MCP_TOKEN`) | `POST .../write` with `{"content":"...","filename":"..."}` |
+| **Anything HTTP-only (curl, scripts)** | Tailnet only: `POST https://koala.tail30472e.ts.net:8443/query` with `{"query":"..."}`, sending `BRAIN_MCP_TOKEN` (not yet enforced on REST, hyperguild#149). The public `brain-mcp.d-ma.be` serves `/mcp` only (infra#517). | `POST .../write` with `{"content":"...","wing":"...","hall":"..."}` |
 | **Browser / human inspection** | `https://git.d-ma.be/mathias/hyperguild` → `knowledge/` and `wiki/` markdown files |

- **Scoping**: defaults to `public` collection; client projects filter to `{client}` + `public`.
+- **Scoping — the brain holds no confidential client content.** Nothing on the read path
+  filters by client: every token holder can read every wing, and `wing` on `brain_query`
+  is an optional argument, not an access control. Client confidentiality is enforced only
+  at write time (`capture`'s I1 gate refuses internal/confidential). So never `brain_write`
+  client material; generalise it first or keep it in the client's own repo. (hyperguild#144,
+  decided 2026-10-01. This line previously claimed client projects were filtered; they
+  were not.)
 - **Routing**: brain_answer's LLM uses berget.ai as primary, iguana ollama as
  fallback. Both are configurable in the `supervisor/ingestion-deployment.yaml`
  on the koala k3s cluster; don't hardcode local-only model names into the
@@ -314,8 +430,20 @@ Workflow:
   and missing-untracked adapters). A drift fails the check with a
   message telling you to stage the regenerated files.

-Behavior rules in this file and per-project rules in `PROJECT.md` apply
-unconditionally on every host, every harness.
+Behavior rules in this file and per-project rules in `PROJECT.md` apply on every
+host and every harness.
+
+**Where a project rule contradicts a rule here, the project rule governs — and it
+must state its reason.** Both halves matter: the project file wins because it knows
+something this file cannot (infra's Flux constraint is real and not visible from
+here), and it must say why because a bare override is indistinguishable from drift.
+
+This replaces the previous wording, which said rules here and in `PROJECT.md` both
+apply *"unconditionally"*. That was not a tiebreak but a contradiction, and it went
+unnoticed until rule 5 and infra's "never create feature branches" had to be obeyed
+at the same time (infra#470). A standard can hold two individually sensible clauses
+that are jointly unsatisfiable, and survive for as long as everyone reads it
+charitably — the same failure the Inception Sprint Oath's G6 had.

 ## Engineering Skills

@@ -385,9 +513,9 @@ Check a host with `task doctor` in the skills repo; it also runs automatically a

 ### Git
 - Conventional commits: `feat:`, `fix:`, `chore:`, `docs:`, `refactor:`
- **Trunk-Based Development:** commit directly to main. One logical change per commit.
+- **Trunk-Based Development:** one logical change per commit, merged to main by PR (`main` is protected, see below).
 - Run `task check` locally before every push. CI is the quality gate, not branch protection.
- No feature branches, no PRs for solo/agent work.
+- Short-lived branches and PRs are how work reaches `main` here (`tbd_ship`/`pr_merge`); no long-lived feature branches.
 - Exception: if a parallel agent session is active on this repo, use a short-lived
  `agent/<description>` branch and merge within the same session.

@@ -427,8 +555,8 @@ When acting as a coding agent on this project:
 3. If unsure about a convention, check `DECISIONS.md` or ask
 4. Never modify files outside the project root without explicit permission
 5. When adding a dependency, explain why in the commit message
-6. Commit directly to main. Run `task check` before every push. Never create
-   feature branches unless a parallel agent is simultaneously active on this repo.
+6. Ship by PR (`main` is protected, see below). Run `task check` before every push. Never create
+   long-lived feature branches.
 7. For client projects: never send code or context to cloud APIs — use local models via LiteLLM

 ## Current state — v0.2.5 (2026-05-17)
@@ -453,3 +581,15 @@ Current main: `e31fd3f`. CI green. Deployed via Flux.

 2. **Issue #19** — end-to-end mirror flow verification.
   `repo_mirror_push` is implemented but the full flow (create repo → add push mirror → verify sync to GitHub) has not been tested manually. Do this before relying on it in production.
+
+## `main` is protected: ship by PR (local-dev#42, decided 2026-10-08)
+
+Gitea refuses direct pushes to `main` in this repo, and a PR merges only after its CI status (`CD / Lint / Test / Vet (pull_request)`) passes.
+**Why:** local-dev rule 5's reviewed lane (risk/medium+ is reviewed by a human) was enforced on the merge tools
+(`tbd_ship`, `pr_merge`: gitea-mcp v0.17.0/v0.18.0, merge only for an issue labelled exactly `risk/low` with green CI)
+but not on a direct commit to `main`, which bypassed it. This overrides rule 5's "risk/low commits straight to main"
+for this repo; control code is where the lane matters.
+**How:** `tbd_ship` for a one-file change, or branch -> PR -> `pr_merge` (name the `risk/low` issue); above `risk/low`
+open the PR and a human merges it in the Gitea UI after review. Tags are not protected. CD jobs push image tags to
+the **infra** repo with deploy keys, never to this repo's `main`. Undo (Mathias only): Settings -> Branches, or
+`PATCH /repos/mathias/gitea-mcp/branch_protections/main {"enable_push": true}`.
@@ -37,9 +37,9 @@

 ### Git
 - Conventional commits: `feat:`, `fix:`, `chore:`, `docs:`, `refactor:`
- **Trunk-Based Development:** commit directly to main. One logical change per commit.
+- **Trunk-Based Development:** one logical change per commit, merged to main by PR (`main` is protected, see below).
 - Run `task check` locally before every push. CI is the quality gate, not branch protection.
- No feature branches, no PRs for solo/agent work.
+- Short-lived branches and PRs are how work reaches `main` here (`tbd_ship`/`pr_merge`); no long-lived feature branches.
 - Exception: if a parallel agent session is active on this repo, use a short-lived
  `agent/<description>` branch and merge within the same session.

@@ -79,8 +79,8 @@ When acting as a coding agent on this project:
 3. If unsure about a convention, check `DECISIONS.md` or ask
 4. Never modify files outside the project root without explicit permission
 5. When adding a dependency, explain why in the commit message
-6. Commit directly to main. Run `task check` before every push. Never create
-   feature branches unless a parallel agent is simultaneously active on this repo.
+6. Ship by PR (`main` is protected, see below). Run `task check` before every push. Never create
+   long-lived feature branches.
 7. For client projects: never send code or context to cloud APIs — use local models via LiteLLM

 ## Current state — v0.2.5 (2026-05-17)
@@ -105,3 +105,15 @@ Current main: `e31fd3f`. CI green. Deployed via Flux.

 2. **Issue #19** — end-to-end mirror flow verification.
   `repo_mirror_push` is implemented but the full flow (create repo → add push mirror → verify sync to GitHub) has not been tested manually. Do this before relying on it in production.
+
+## `main` is protected: ship by PR (local-dev#42, decided 2026-10-08)
+
+Gitea refuses direct pushes to `main` in this repo, and a PR merges only after its CI status (`CD / Lint / Test / Vet (pull_request)`) passes.
+**Why:** local-dev rule 5's reviewed lane (risk/medium+ is reviewed by a human) was enforced on the merge tools
+(`tbd_ship`, `pr_merge`: gitea-mcp v0.17.0/v0.18.0, merge only for an issue labelled exactly `risk/low` with green CI)
+but not on a direct commit to `main`, which bypassed it. This overrides rule 5's "risk/low commits straight to main"
+for this repo; control code is where the lane matters.
+**How:** `tbd_ship` for a one-file change, or branch -> PR -> `pr_merge` (name the `risk/low` issue); above `risk/low`
+open the PR and a human merges it in the Gitea UI after review. Tags are not protected. CD jobs push image tags to
+the **infra** repo with deploy keys, never to this repo's `main`. Undo (Mathias only): Settings -> Branches, or
+`PATCH /repos/mathias/gitea-mcp/branch_protections/main {"enable_push": true}`.
@@ -125,6 +125,22 @@ by area:
 | Search / Templates | code_search, create_project_from_template |
 | Composite | tbd_ship (trunk-based ship helper) |

+**Merge gate: `tbd_ship` (gitea-mcp#63) and `pr_merge` (gitea-mcp#74).** The two merge tools share one gate
+(`evaluateShipGate`). A PR is merged only when ALL of: the head commit's CI is fully green, the base branch is not
+review-protected, and the issue named in `issue` carries **exactly one** `risk/` label, `risk/low`. No `issue`, no
+`risk/` label, several, an unknown value (`risk/lowish`), `risk/medium`, `risk/high`, or an issue that cannot be read
+all fail closed: nothing is merged, the result is `merged:false` with a `reason`, the `ci_status` and the `risk` tier.
+Unclassified is not low. Above `risk/low` a human reviews and merges in the Gitea UI (which this gate does not touch).
+`pr_merge` additionally fails with an error, merging nothing, when the PR or the base's branch-protection state cannot
+be read. A merge conflict is still an error.
+Both tools bind the merge to the head commit whose CI the gate evaluated (`head_commit_id`), so a push to the PR
+branch after the check makes Gitea refuse the merge instead of merging unchecked code.
+
+Limits: the gate enforces the tier of the issue the caller names; it cannot prove that issue is the one the change
+belongs to, so the link is the caller's claim (local-dev rule 5's convention still applies). Not gated: merging in the
+Gitea UI, and `file_write_branch` committing directly to a branch (`main` included), which is how `risk/low` work
+reaches trunk. Platform-level enforcement for those is branch protection on `main` (local-dev#27 option 2).
+
 Shared tool-layer hygiene (`internal/tools/tool.go`):
 - **Alias normalization** — `repo`↔`name` and `number`↔`index` are reconciled so
  a tool works whichever spelling a caller sends (an explicit canonical wins).
@@ -3,7 +3,7 @@ module git.d-ma.be/mathias/gitea-mcp
 go 1.26.2

 require (
-	git.d-ma.be/mathias/mcp-chassis v0.6.0
+	git.d-ma.be/mathias/mcp-chassis v0.7.1
 	github.com/hashicorp/golang-lru/v2 v2.0.7
 	github.com/stretchr/testify v1.11.1
 )
@@ -2,6 +2,10 @@ git.d-ma.be/mathias/mcp-chassis v0.5.0 h1:0w3dt4t4r8OtZBHIOoZkR6p6L3L6YDiqhMh9bT
 git.d-ma.be/mathias/mcp-chassis v0.5.0/go.mod h1:Ks7EK2UnGAN0H3rJjKUxUagX8/ZBdtLrOlcUbv0RwH8=
 git.d-ma.be/mathias/mcp-chassis v0.6.0 h1:/MuxT9IyzZbN26+COlI2FK0moTyXMLG/qXc/cm0i0Vg=
 git.d-ma.be/mathias/mcp-chassis v0.6.0/go.mod h1:Ks7EK2UnGAN0H3rJjKUxUagX8/ZBdtLrOlcUbv0RwH8=
+git.d-ma.be/mathias/mcp-chassis v0.7.0 h1:ct31v1F1PMhMCva/V0ErEbgMzwnT+27lYKbug2A/quE=
+git.d-ma.be/mathias/mcp-chassis v0.7.0/go.mod h1:Ks7EK2UnGAN0H3rJjKUxUagX8/ZBdtLrOlcUbv0RwH8=
+git.d-ma.be/mathias/mcp-chassis v0.7.1 h1:4h8TOBHTA0L/no4IPcr4tLcQtmKUlhE2/cXyPdNFGmM=
+git.d-ma.be/mathias/mcp-chassis v0.7.1/go.mod h1:Ks7EK2UnGAN0H3rJjKUxUagX8/ZBdtLrOlcUbv0RwH8=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -2,9 +2,12 @@ package auth

 import (
 	"context"
+	"log/slog"
 	"net/http"
 	"strings"

+	chassisauth "git.d-ma.be/mathias/mcp-chassis/auth"
+
 	"git.d-ma.be/mathias/gitea-mcp/internal/gitea"
 )

@@ -25,11 +28,34 @@ func PassthroughMiddleware(validator TokenValidator, onValid, fallback http.Hand
 		token, hasBearer := strings.CutPrefix(authz, "Bearer ")
 		if hasBearer && token != "" {
 			if login, ok := validator.ValidateToken(r.Context(), token); ok {
+				// gitea-mcp#60. This middleware logged nothing at all, so there
+				// was no way to tell a PAT pass-through request from one that
+				// fell through to the shared JWT/static path — only inference
+				// from which failures were absent.
+				//
+				// It matters more here than on the other two paths: this is
+				// where a request acts as a SPECIFIC Gitea user. An audit trail
+				// that cannot say which identity performed a write is not an
+				// audit trail.
+				//
+				// chassisauth.Fingerprint, not a local equivalent: the chassis
+				// emits the same field on its own accepted/rejected lines, and
+				// two different fingerprints for one token cannot be correlated.
+				slog.Info("mcp auth accepted",
+					"auth_path", "passthrough",
+					"caller", login,
+					"token_fp", chassisauth.Fingerprint(token),
+					"path", r.URL.Path,
+				)
 				ctx := withCaller(r.Context(), login)
 				ctx = gitea.WithToken(ctx, token)
 				onValid.ServeHTTP(w, r.WithContext(ctx))
 				return
 			}
+			// A token that is not a valid PAT falls through SILENTLY. The
+			// chassis owns the verdict; logging here too would double-count
+			// every static-token request, since all of them reach this
+			// middleware first.
 		}
 		fallback.ServeHTTP(w, r)
 	})
@@ -0,0 +1,83 @@
+package auth
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+type stubValidator struct{ login string }
+
+func (s stubValidator) ValidateToken(context.Context, string) (string, bool) {
+	return s.login, s.login != ""
+}
+
+// gitea-mcp#60: PassthroughMiddleware logged nothing at all, success or failure.
+//
+// While verifying #59 end to end there was no way to confirm from the server
+// logs whether calls were using the new PAT pass-through (caller's own Gitea
+// PAT, own identity) or falling through to the shared JWT/static path. The only
+// way to reason about it was inference. The chassis names its two paths as of
+// v0.7.0; this is the third.
+//
+// This matters more than the other two: pass-through is the path where the
+// request acts as a SPECIFIC Gitea user. An audit trail that cannot say which
+// identity performed a write is not an audit trail.
+func TestPassthroughMiddleware_LogsTheAcceptedPath(t *testing.T) {
+	var buf bytes.Buffer
+	prev := slog.Default()
+	slog.SetDefault(slog.New(slog.NewJSONHandler(&buf, &slog.HandlerOptions{Level: slog.LevelInfo})))
+	t.Cleanup(func() { slog.SetDefault(prev) })
+
+	reached := false
+	h := PassthroughMiddleware(stubValidator{login: "alice"},
+		http.HandlerFunc(func(http.ResponseWriter, *http.Request) { reached = true }),
+		http.HandlerFunc(func(http.ResponseWriter, *http.Request) { t.Fatal("must not fall back") }))
+
+	req := httptest.NewRequest(http.MethodGet, "/mcp", nil)
+	req.Header.Set("Authorization", "Bearer a-real-gitea-pat")
+	h.ServeHTTP(httptest.NewRecorder(), req)
+	require.True(t, reached)
+
+	var line map[string]any
+	require.NoError(t, json.Unmarshal([]byte(strings.TrimSpace(buf.String())), &line))
+	require.Equal(t, "mcp auth accepted", line["msg"])
+	require.Equal(t, "passthrough", line["auth_path"])
+	require.Equal(t, "alice", line["caller"])
+	require.NotEmpty(t, line["token_fp"], "correlatable without being a credential")
+
+	for k, v := range line {
+		require.NotContains(t, []any{v}, "a-real-gitea-pat",
+			"the PAT must never appear in any field, including %q", k)
+	}
+}
+
+// A token that is not a valid PAT must fall through silently to the chassis,
+// which owns the rejection log. Logging here as well would double-count every
+// static-token request — every one of them reaches this middleware first.
+func TestPassthroughMiddleware_FallsThroughWithoutLogging(t *testing.T) {
+	var buf bytes.Buffer
+	prev := slog.Default()
+	slog.SetDefault(slog.New(slog.NewJSONHandler(&buf, &slog.HandlerOptions{Level: slog.LevelInfo})))
+	t.Cleanup(func() { slog.SetDefault(prev) })
+
+	fellBack := false
+	h := PassthroughMiddleware(stubValidator{},
+		http.HandlerFunc(func(http.ResponseWriter, *http.Request) { t.Fatal("must not accept") }),
+		http.HandlerFunc(func(http.ResponseWriter, *http.Request) { fellBack = true }))
+
+	req := httptest.NewRequest(http.MethodGet, "/mcp", nil)
+	req.Header.Set("Authorization", "Bearer not-a-pat")
+	h.ServeHTTP(httptest.NewRecorder(), req)
+
+	require.True(t, fellBack)
+	require.Empty(t, strings.TrimSpace(buf.String()),
+		"the chassis owns the verdict; logging here too would double-count every static-token call")
+}
@@ -107,8 +107,10 @@ func (c *Client) GetPullRequestDiff(ctx context.Context, owner, repo string, ind

 type MergePRArgs struct {
 	Do    string `json:"Do"`
-	Title string `json:"merge_message_title,omitempty"`
+	Title string `json:"merge_title_field,omitempty"`
 	Body  string `json:"merge_message_field,omitempty"`
+	// HeadCommitID binds the merge to the commit whose CI was checked: if the PR branch moved, Gitea refuses.
+	HeadCommitID string `json:"head_commit_id,omitempty"`
 }

 func (c *Client) MergePullRequest(ctx context.Context, owner, repo string, index int, args MergePRArgs) error {
@@ -0,0 +1,46 @@
+package mcp
+
+import (
+	"context"
+	"encoding/json"
+	"log/slog"
+	"time"
+
+	chassisauth "git.d-ma.be/mathias/mcp-chassis/auth"
+
+	"git.d-ma.be/mathias/gitea-mcp/internal/auth"
+	"git.d-ma.be/mathias/gitea-mcp/internal/gitea"
+)
+
+// auditTargetKeys are the only argument values copied into the audit record:
+// they identify what a call acted on without copying what it wrote. Every
+// other argument (file content, issue bodies) is reduced to args_bytes so the
+// audit trail never becomes a second copy of repo content (gitea-mcp#64).
+var auditTargetKeys = []string{"owner", "repo", "name", "number", "index", "branch", "path", "tag"}
+
+// logToolCall emits one structured record per tools/call (OWASP MCP08).
+func logToolCall(ctx context.Context, name string, args json.RawMessage, start time.Time, err error) {
+	attrs := []any{
+		"tool", name,
+		"caller", auth.Caller(ctx),
+		"args_bytes", len(args),
+		"duration_ms", time.Since(start).Milliseconds(),
+	}
+	if tok, ok := gitea.TokenFromContext(ctx); ok {
+		attrs = append(attrs, "token_fp", chassisauth.Fingerprint(tok))
+	}
+	var parsed map[string]any
+	if json.Unmarshal(args, &parsed) == nil {
+		for _, k := range auditTargetKeys {
+			if v, ok := parsed[k]; ok {
+				attrs = append(attrs, k, v)
+			}
+		}
+	}
+	outcome := "ok"
+	if err != nil {
+		outcome = err.Error()
+	}
+	attrs = append(attrs, "outcome", outcome)
+	slog.InfoContext(ctx, "mcp tool call", attrs...)
+}
@@ -0,0 +1,135 @@
+package mcp_test
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"git.d-ma.be/mathias/gitea-mcp/internal/auth"
+	"git.d-ma.be/mathias/gitea-mcp/internal/mcp"
+	"git.d-ma.be/mathias/gitea-mcp/internal/registry"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+type auditStubTool struct {
+	name string
+	err  error
+}
+
+func (s auditStubTool) Descriptor() registry.ToolDescriptor {
+	return registry.ToolDescriptor{Name: s.name, InputSchema: json.RawMessage(`{"type":"object"}`)}
+}
+
+func (s auditStubTool) Call(context.Context, json.RawMessage) (json.RawMessage, error) {
+	if s.err != nil {
+		return nil, s.err
+	}
+	return json.RawMessage(`{"ok":true}`), nil
+}
+
+// captureAudit routes slog to a buffer and returns the decoded
+// "mcp tool call" records emitted while fn runs.
+func captureAudit(t *testing.T, fn func()) []map[string]any {
+	t.Helper()
+	var buf bytes.Buffer
+	prev := slog.Default()
+	slog.SetDefault(slog.New(slog.NewJSONHandler(&buf, nil)))
+	t.Cleanup(func() { slog.SetDefault(prev) })
+
+	fn()
+
+	var out []map[string]any
+	for _, l := range strings.Split(strings.TrimSpace(buf.String()), "\n") {
+		if l == "" {
+			continue
+		}
+		var rec map[string]any
+		require.NoError(t, json.Unmarshal([]byte(l), &rec))
+		if rec["msg"] == "mcp tool call" {
+			out = append(out, rec)
+		}
+	}
+	return out
+}
+
+func auditServer(tools ...registry.Tool) http.Handler {
+	reg := registry.New()
+	for _, tl := range tools {
+		reg.Register(tl)
+	}
+	srv := mcp.NewServer(mcp.ServerOptions{Registry: reg})
+	return auth.CallerMiddleware(slog.Default(), srv)
+}
+
+func callTool(t *testing.T, h http.Handler, name string, args any) {
+	t.Helper()
+	body := map[string]any{
+		"jsonrpc": "2.0", "id": 1, "method": "tools/call",
+		"params": map[string]any{"name": name, "arguments": args},
+	}
+	b, err := json.Marshal(body)
+	require.NoError(t, err)
+	req := httptest.NewRequest(http.MethodPost, "/", bytes.NewBuffer(b))
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("X-Auth-Request-User", "alice")
+	h.ServeHTTP(httptest.NewRecorder(), req)
+}
+
+// gitea-mcp#64: tools/call dispatched with no record at all.
+func TestToolsCallLogsOneAuditRecordWithToolCallerAndTarget(t *testing.T) {
+	recs := captureAudit(t, func() {
+		callTool(t, auditServer(auditStubTool{name: "repo_delete"}), "repo_delete",
+			map[string]any{"owner": "mathias", "repo": "scratch", "confirm": "scratch"})
+	})
+	require.Len(t, recs, 1)
+	r := recs[0]
+	assert.Equal(t, "repo_delete", r["tool"])
+	assert.Equal(t, "alice", r["caller"])
+	assert.Equal(t, "mathias", r["owner"])
+	assert.Equal(t, "scratch", r["repo"])
+	assert.Equal(t, "ok", r["outcome"])
+	assert.Contains(t, r, "duration_ms")
+}
+
+func TestToolsCallAuditRecordsErrorOutcome(t *testing.T) {
+	recs := captureAudit(t, func() {
+		callTool(t, auditServer(auditStubTool{name: "pr_merge", err: errors.New("boom")}),
+			"pr_merge", map[string]any{"owner": "mathias", "repo": "x", "number": 3})
+	})
+	require.Len(t, recs, 1)
+	assert.Equal(t, "boom", recs[0]["outcome"])
+	assert.EqualValues(t, 3, recs[0]["number"])
+}
+
+// The audit line must not become a second copy of repo content.
+func TestToolsCallAuditOmitsArgumentValuesOtherThanTarget(t *testing.T) {
+	const sentinel = "SENTINEL-FILE-CONTENT-7f3a"
+	recs := captureAudit(t, func() {
+		callTool(t, auditServer(auditStubTool{name: "file_write_branch"}), "file_write_branch",
+			map[string]any{"owner": "mathias", "repo": "x", "path": "a.txt", "content": sentinel})
+	})
+	require.Len(t, recs, 1)
+	raw, err := json.Marshal(recs[0])
+	require.NoError(t, err)
+	assert.NotContains(t, string(raw), sentinel)
+	assert.Contains(t, recs[0], "args_bytes")
+}
+
+// #38 keeps `name` (for repo) and `index` (for number) as legacy aliases; a
+// caller using them must still leave a target in the audit record.
+func TestToolsCallAuditRecordsLegacyAliasTargets(t *testing.T) {
+	recs := captureAudit(t, func() {
+		callTool(t, auditServer(auditStubTool{name: "issue_close"}), "issue_close",
+			map[string]any{"owner": "mathias", "name": "infra", "index": 7})
+	})
+	require.Len(t, recs, 1)
+	assert.Equal(t, "infra", recs[0]["name"])
+	assert.EqualValues(t, 7, recs[0]["index"])
+}
@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"errors"
 	"net/http"
+	"time"

 	"git.d-ma.be/mathias/gitea-mcp/internal/registry"
 )
@@ -90,7 +91,9 @@ func (s *Server) handlePOST(w http.ResponseWriter, r *http.Request) {
 				NewErrorResponse(req.ID, -32602, "invalid params", nil))
 			return
 		}
+		start := time.Now()
 		out, err := s.opts.Registry.Dispatch(r.Context(), p.Name, p.Arguments)
+		logToolCall(r.Context(), p.Name, p.Arguments, start, err)
 		if err != nil {
 			code := -32000
 			if errors.Is(err, registry.ErrToolNotFound) {
@@ -63,6 +63,9 @@ func (t *FileWriteBranch) Call(ctx context.Context, raw json.RawMessage) (json.R
 	if args.Branch == "" {
 		return nil, fmt.Errorf("branch is required: %w", gitea.ErrValidation)
 	}
+	if err := ValidateBranchName(args.Branch); err != nil {
+		return nil, err
+	}
 	if args.Message == "" {
 		return nil, fmt.Errorf("message is required: %w", gitea.ErrValidation)
 	}
@@ -0,0 +1,42 @@
+package tools
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+)
+
+// foldKey must agree with strings.EqualFold, which is how encoding/json matches.
+func TestFoldKeyAgreesWithEqualFold(t *testing.T) {
+	pairs := [][2]string{{"branch", "Branch"}, {"sha", "ſha"}, {"k", "K"}, {"ss", "ß"}, {"a", "b"}, {"Σ", "ς"}}
+	for _, p := range pairs {
+		if got, want := foldKey(p[0]) == foldKey(p[1]), strings.EqualFold(p[0], p[1]); got != want {
+			t.Errorf("%q vs %q: foldKey equal=%v, EqualFold=%v", p[0], p[1], got, want)
+		}
+	}
+}
+
+// The duplicate scan is linear in the number of keys: 90k keys (about the
+// 1 MiB body cap) must scan well under a second.
+func TestRejectAmbiguousKeysScalesLinearly(t *testing.T) {
+	var b strings.Builder
+	b.WriteString("{")
+	for i := 0; i < 90_000; i++ {
+		if i > 0 {
+			b.WriteString(",")
+		}
+		fmt.Fprintf(&b, `"k%d":1`, i)
+	}
+	b.WriteString("}")
+	var dst struct {
+		Owner string `json:"owner"`
+	}
+	start := time.Now()
+	if err := rejectAmbiguousKeys([]byte(b.String()), &dst); err != nil {
+		t.Fatal(err)
+	}
+	if d := time.Since(start); d > 2*time.Second {
+		t.Errorf("scan of 90k keys took %v", d)
+	}
+}
@@ -10,6 +10,9 @@ import (
 	"git.d-ma.be/mathias/gitea-mcp/internal/registry"
 )

+// PRMerge merges an existing PR behind the same gate as tbd_ship (gitea-mcp#74, #63): CI fully green, base not
+// review-protected, and the issue named in `issue` classified exactly risk/low. Above risk/low a human reviews and
+// merges in the Gitea UI; this tool then merges nothing and says why.
 type PRMerge struct {
 	c *gitea.Client
 	a *allowlist.Allowlist
@@ -22,16 +25,18 @@ func NewPRMerge(c *gitea.Client, a *allowlist.Allowlist) *PRMerge {
 func (t *PRMerge) Descriptor() registry.ToolDescriptor {
 	return registry.ToolDescriptor{
 		Name:        "pr_merge",
-		Description: "Merge a pull request. style: merge (default), squash, or rebase.",
+		Description: "Merge a pull request, gated like tbd_ship: it merges ONLY when the head commit's CI is fully green AND the base branch is not review-protected AND the issue named in `issue` carries exactly one risk/ label, risk/low. Otherwise it merges nothing and returns merged:false with a reason and the risk tier (no issue named, no or several risk/ labels, an unknown value, risk/medium, risk/high, an unreadable issue, CI pending/red/absent, a review-protected base). Above risk/low a human reviews and merges in the Gitea UI. style: merge (default), squash, or rebase. Set ci_timeout_seconds to poll CI to completion (default 0 = single snapshot). Returns {merged, pr_number, ci_status, risk, reason}.",
 		InputSchema: json.RawMessage(`{
 			"type":"object",
 			"properties":{
 				"owner":{"type":"string"},
 				"repo":{"type":"string"},
 				"number":{"type":"integer","minimum":1},
+				"issue":{"type":"integer","minimum":1,"description":"Issue number in the same repo whose risk/ label classifies this change. Merge happens only for exactly one label, risk/low; no issue, none, several, an unknown value, risk/medium or risk/high means nothing is merged (unclassified is not low)."},
 				"style":{"type":"string","enum":["merge","squash","rebase"]},
 				"merge_message_title":{"type":"string"},
-				"merge_message_field":{"type":"string"}
+				"merge_message_field":{"type":"string"},
+				"ci_timeout_seconds":{"type":"integer","minimum":0,"maximum":600,"description":"Poll CI up to this long for the head commit's runs to complete. Default 0 (single snapshot)."}
 			},
 			"required":["owner","repo","number"]
 		}`),
@@ -39,12 +44,14 @@ func (t *PRMerge) Descriptor() registry.ToolDescriptor {
 }

 type prMergeArgs struct {
-	Owner  string `json:"owner"`
-	Repo   string `json:"repo"`
-	Number int    `json:"number"`
-	Style  string `json:"style"`
-	Title  string `json:"merge_message_title"`
-	Body   string `json:"merge_message_field"`
+	Owner        string `json:"owner"`
+	Repo         string `json:"repo"`
+	Number       int    `json:"number"`
+	Issue        int    `json:"issue"`
+	Style        string `json:"style"`
+	Title        string `json:"merge_message_title"`
+	Body         string `json:"merge_message_field"`
+	CITimeoutSec int    `json:"ci_timeout_seconds"`
 }

 func (t *PRMerge) Call(ctx context.Context, raw json.RawMessage) (json.RawMessage, error) {
@@ -64,13 +71,47 @@ func (t *PRMerge) Call(ctx context.Context, raw json.RawMessage) (json.RawMessag
 		style = "merge"
 	}

+	// The gate's inputs. An unreadable PR or branch-protection probe is an error: the gate cannot be evaluated, so
+	// nothing is merged. An unreadable issue is a tier (unreadable), not an error: it is reported and not merged.
+	pr, err := t.c.GetPullRequest(ctx, args.Owner, args.Repo, args.Number)
+	if err != nil {
+		return nil, fmt.Errorf("read PR #%d: %w", args.Number, err)
+	}
+	// Gitea refuses closed and merged PRs itself; the gate does not depend on that.
+	if pr.State != "open" || pr.Draft {
+		reason := fmt.Sprintf("the PR is not open (state %q) — nothing to merge", pr.State)
+		if pr.State == "open" {
+			reason = "the PR is a draft — not merged"
+		}
+		return textOK(map[string]any{"merged": false, "pr_number": args.Number, "reason": reason})
+	}
+	bp, err := t.c.GetBranchProtection(ctx, args.Owner, args.Repo, pr.Base.Ref)
+	if err != nil {
+		return nil, fmt.Errorf("branch protection probe: %w", err)
+	}
+	runs := pollRuns(ctx, t.c, args.Owner, args.Repo, pr.Head.Sha, args.CITimeoutSec)
+	risk := riskTierOf(ctx, t.c, args.Owner, args.Repo, args.Issue)
+	gate := evaluateMergeGate(pr.Head.Sha, runs, bp, risk)
+
+	result := map[string]any{
+		"merged":    false,
+		"pr_number": args.Number,
+		"ci_status": gate.ciStatus,
+		"risk":      string(risk),
+	}
+	if !gate.merge {
+		result["reason"] = gate.reason
+		return textOK(result)
+	}
+
 	if err := t.c.MergePullRequest(ctx, args.Owner, args.Repo, args.Number, gitea.MergePRArgs{
-		Do:    style,
-		Title: args.Title,
-		Body:  args.Body,
+		Do:           style,
+		Title:        args.Title,
+		Body:         args.Body,
+		HeadCommitID: pr.Head.Sha,
 	}); err != nil {
 		return nil, err
 	}
-
-	return textOK(map[string]any{"merged": true})
+	result["merged"] = true
+	return textOK(result)
 }
@@ -6,6 +6,8 @@ import (
 	"io"
 	"net/http"
 	"net/http/httptest"
+	"strings"
+	"sync"
 	"testing"

 	"git.d-ma.be/mathias/gitea-mcp/internal/allowlist"
@@ -15,80 +17,396 @@ import (
 	"github.com/stretchr/testify/require"
 )

+// pr_merge is gated exactly like tbd_ship (gitea-mcp#74, #63): it merges only when the head commit's CI is fully
+// green, the base is not review-protected AND the issue named in `issue` carries exactly one risk/ label, risk/low.
+// The fake serves the whole read side the gate needs and records whether (and how) the merge endpoint was hit.
+
+const (
+	greenRuns = `[{"id":1,"status":"completed","conclusion":"success","head_sha":"abc"}]`
+	lowLabels = `[{"id":1,"name":"risk/low"}]`
+)
+
+type prMergeFakeOpts struct {
+	runs              string // workflow_runs array for the head commit
+	issueLabels       string // labels array of issue 42
+	issueStatus       int
+	prStatus          int // GET pulls/7
+	protectionStatus  int // 0 → 404 (unprotected); 200 with required approvals when requiredApprovals>0
+	requiredApprovals int
+	mergeStatus       int
+	emptyHeadSha      bool   // the PR reports head.sha == ""
+	prState           string // default "open"
+	prDraft           bool
+}
+
+type prMergeFake struct {
+	*httptest.Server
+	mu         sync.Mutex
+	mergeCalls int
+	mergeBody  []byte
+	runsQuery  string // raw query of the last workflow-runs request
+}
+
+func (f *prMergeFake) merged() bool { f.mu.Lock(); defer f.mu.Unlock(); return f.mergeCalls > 0 }
+
+func newPRMergeFake(t *testing.T, o prMergeFakeOpts) *prMergeFake {
+	t.Helper()
+	if o.runs == "" {
+		o.runs = greenRuns
+	}
+	if o.issueLabels == "" {
+		o.issueLabels = lowLabels
+	}
+	if o.issueStatus == 0 {
+		o.issueStatus = http.StatusOK
+	}
+	if o.prStatus == 0 {
+		o.prStatus = http.StatusOK
+	}
+	if o.mergeStatus == 0 {
+		o.mergeStatus = http.StatusNoContent
+	}
+	if o.prState == "" {
+		o.prState = "open"
+	}
+	sha := "abc"
+	if o.emptyHeadSha {
+		sha = ""
+	}
+	f := &prMergeFake{}
+	f.Server = httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		p := r.URL.Path
+		w.Header().Set("Content-Type", "application/json")
+		switch {
+		case r.Method == http.MethodGet && p == "/api/v1/repos/owner/repo/pulls/7":
+			w.WriteHeader(o.prStatus)
+			_, _ = w.Write([]byte(`{"number":7,"title":"t","state":"` + o.prState + `","draft":` + boolJSON(o.prDraft) + `,"html_url":"http://x/pulls/7","head":{"ref":"feat","sha":"` + sha + `"},"base":{"ref":"main"}}`))
+		case r.Method == http.MethodGet && strings.Contains(p, "/branch_protections/"):
+			switch {
+			case o.protectionStatus != 0:
+				w.WriteHeader(o.protectionStatus)
+				_, _ = w.Write([]byte(`{"message":"boom"}`))
+			case o.requiredApprovals > 0:
+				_, _ = w.Write([]byte(`{"required_approvals":` + itoa(o.requiredApprovals) + `}`))
+			default:
+				w.WriteHeader(http.StatusNotFound)
+				_, _ = w.Write([]byte(`{"message":"not found"}`))
+			}
+		case r.Method == http.MethodGet && strings.Contains(p, "/actions/runs"):
+			f.mu.Lock()
+			f.runsQuery = r.URL.RawQuery
+			f.mu.Unlock()
+			_, _ = w.Write([]byte(`{"total_count":0,"workflow_runs":` + o.runs + `}`))
+		case r.Method == http.MethodGet && p == "/api/v1/repos/owner/repo/issues/42":
+			w.WriteHeader(o.issueStatus)
+			_, _ = w.Write([]byte(`{"number":42,"title":"t","state":"open","labels":` + o.issueLabels + `}`))
+		case r.Method == http.MethodPost && p == "/api/v1/repos/owner/repo/pulls/7/merge":
+			body, _ := io.ReadAll(r.Body)
+			f.mu.Lock()
+			f.mergeCalls++
+			f.mergeBody = body
+			f.mu.Unlock()
+			w.WriteHeader(o.mergeStatus)
+			_, _ = w.Write([]byte(`{"message":"merge conflict"}`))
+		default:
+			t.Errorf("unexpected request: %s %s", r.Method, p)
+			w.WriteHeader(http.StatusNotFound)
+		}
+	}))
+	t.Cleanup(f.Close)
+	return f
+}
+
+func boolJSON(b bool) string {
+	if b {
+		return "true"
+	}
+	return "false"
+}
+
+func callMerge(t *testing.T, f *prMergeFake, args string) (map[string]any, error) {
+	t.Helper()
+	tool := tools.NewPRMerge(gitea.NewClient(f.URL, "tok"), allowlist.New([]string{"owner"}))
+	out, err := tool.Call(context.Background(), json.RawMessage(args))
+	if err != nil {
+		return nil, err
+	}
+	var res map[string]any
+	require.NoError(t, json.Unmarshal(out, &res))
+	return res, nil
+}
+
+const mergeArgs = `{"owner":"owner","repo":"repo","number":7,"issue":42}`
+
 func TestPRMergeSuccess(t *testing.T) {
-	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		assert.Equal(t, "/api/v1/repos/owner/repo/pulls/7/merge", r.URL.Path)
-		w.WriteHeader(http.StatusNoContent)
-	}))
-	defer srv.Close()
+	f := newPRMergeFake(t, prMergeFakeOpts{})
+
+	res, err := callMerge(t, f, mergeArgs)

-	tool := tools.NewPRMerge(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"owner"}))
-	out, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"owner","name":"repo","index":7}`))
 	require.NoError(t, err)
-
-	var result map[string]any
-	require.NoError(t, json.Unmarshal(out, &result))
-	assert.Equal(t, true, result["merged"])
+	assert.Equal(t, true, res["merged"])
+	assert.Equal(t, "success", res["ci_status"])
+	assert.Equal(t, "low", res["risk"])
+	assert.Equal(t, float64(7), res["pr_number"])
+	assert.True(t, f.merged())
 }

-func TestPRMergeDefaultsToMergeStyle(t *testing.T) {
-	var captured []byte
-	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		var err error
-		captured, err = io.ReadAll(r.Body)
-		require.NoError(t, err)
-		w.WriteHeader(http.StatusNoContent)
-	}))
-	defer srv.Close()
-
-	tool := tools.NewPRMerge(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"owner"}))
-	_, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"owner","name":"repo","index":7}`))
+func TestPRMergeDefaultsToMergeStyleAndPassesStyleThrough(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{})
+	_, err := callMerge(t, f, mergeArgs)
 	require.NoError(t, err)
-
 	var payload map[string]any
-	require.NoError(t, json.Unmarshal(captured, &payload))
+	require.NoError(t, json.Unmarshal(f.mergeBody, &payload))
 	assert.Equal(t, "merge", payload["Do"])
+
+	f2 := newPRMergeFake(t, prMergeFakeOpts{})
+	_, err = callMerge(t, f2, `{"owner":"owner","repo":"repo","number":7,"issue":42,"style":"squash"}`)
+	require.NoError(t, err)
+	require.NoError(t, json.Unmarshal(f2.mergeBody, &payload))
+	assert.Equal(t, "squash", payload["Do"])
 }

+// Gitea 1.27's MergePullRequestOption has merge_title_field and merge_message_field. A field it does not know is
+// ignored without an error, so a custom merge title silently did nothing while this tool sent merge_message_title
+// (gitea-mcp#77). The tool's own argument keeps its name; the wire field must be Gitea's.
+func TestPRMerge_ACustomMergeTitleTravelsAsGiteasMergeTitleField(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{})
+
+	_, err := callMerge(t, f, `{"owner":"owner","repo":"repo","number":7,"issue":42,"merge_message_title":"feat: one line","merge_message_field":"body text"}`)
+
+	require.NoError(t, err)
+	var payload map[string]any
+	require.NoError(t, json.Unmarshal(f.mergeBody, &payload))
+	assert.Equal(t, "feat: one line", payload["merge_title_field"])
+	assert.Equal(t, "body text", payload["merge_message_field"])
+	assert.NotContains(t, payload, "merge_message_title", "not a Gitea field: it is ignored without an error")
+}
+
+// TOCTOU: the gate read CI for ONE head commit. The merge must be bound to that commit, so a push to the PR branch
+// after the check makes Gitea refuse the merge instead of merging code nobody checked.
+func TestPRMerge_TheMergeIsBoundToTheCommitTheGateChecked(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{})
+
+	_, err := callMerge(t, f, mergeArgs)
+
+	require.NoError(t, err)
+	var payload map[string]any
+	require.NoError(t, json.Unmarshal(f.mergeBody, &payload))
+	assert.Equal(t, "abc", payload["head_commit_id"], "the head sha the CI check used")
+}
+
+// A merge conflict is still an error, as before the gate existed.
 func TestPRMergeConflictReturnsError(t *testing.T) {
-	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		w.WriteHeader(http.StatusConflict)
-		_, _ = w.Write([]byte(`{"message":"merge conflict"}`))
-	}))
-	defer srv.Close()
+	f := newPRMergeFake(t, prMergeFakeOpts{mergeStatus: http.StatusConflict})
+
+	_, err := callMerge(t, f, mergeArgs)

-	tool := tools.NewPRMerge(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"owner"}))
-	_, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"owner","name":"repo","index":7}`))
 	require.Error(t, err)
 	assert.ErrorIs(t, err, gitea.ErrConflict)
+	assert.True(t, f.merged(), "the merge was attempted: the gate had passed")
 }

-// #45: pr_merge advertises the canonical `number` (was `index`); `index` stays
-// an accepted alias via the shim.
+// #45: pr_merge advertises the canonical `number` (was `index`); `index` stays an accepted alias via the shim.
 func TestPRMergeNumberCanonical(t *testing.T) {
 	sch := string(tools.NewPRMerge(gitea.NewClient("http://unused", ""), allowlist.New([]string{"owner"})).Descriptor().InputSchema)
 	assert.Contains(t, sch, `"number":`, "pr_merge must advertise number")
 	assert.NotContains(t, sch, `"index":`, "pr_merge must not advertise index")

 	for _, args := range []string{
-		`{"owner":"owner","repo":"repo","number":7}`,
-		`{"owner":"owner","repo":"repo","index":7}`,
+		`{"owner":"owner","repo":"repo","number":7,"issue":42}`,
+		`{"owner":"owner","repo":"repo","index":7,"issue":42}`,
+		`{"owner":"owner","name":"repo","index":7,"issue":42}`,
 	} {
-		var gotPath string
-		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			gotPath = r.URL.Path
-			w.WriteHeader(http.StatusNoContent)
-		}))
-		tool := tools.NewPRMerge(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"owner"}))
-		_, err := tool.Call(context.Background(), json.RawMessage(args))
+		f := newPRMergeFake(t, prMergeFakeOpts{})
+		res, err := callMerge(t, f, args)
 		require.NoError(t, err, args)
-		assert.Equal(t, "/api/v1/repos/owner/repo/pulls/7/merge", gotPath, args)
-		srv.Close()
+		assert.Equal(t, true, res["merged"], args)
 	}
 }

+// gitea-mcp#74: the descriptor tells the caller about the gate.
+func TestPRMergeDescriptorAdvertisesTheGate(t *testing.T) {
+	d := tools.NewPRMerge(gitea.NewClient("http://unused", ""), allowlist.New([]string{"owner"})).Descriptor()
+	assert.Contains(t, string(d.InputSchema), `"issue":`)
+	assert.Contains(t, string(d.InputSchema), `"ci_timeout_seconds":`)
+	assert.Contains(t, d.Description, "risk/low")
+	assert.Contains(t, d.Description, "merged:false")
+}
+
 func TestPRMergeAllowlistRejects(t *testing.T) {
 	tool := tools.NewPRMerge(gitea.NewClient("http://unused", ""), allowlist.New([]string{"allowed"}))
 	_, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"evil","name":"repo","index":1}`))
 	require.Error(t, err)
 }
+
+// ---- the tier: only exactly one risk/low merges ----
+func TestPRMerge_RiskTier_OnlyLowMerges(t *testing.T) {
+	tests := []struct {
+		name         string
+		labels       string
+		wantMerged   bool
+		wantInReason string
+	}{
+		{"risk/low merges", `[{"id":1,"name":"risk/low"}]`, true, ""},
+		{"risk/low among other labels merges", `[{"id":2,"name":"type/fix"},{"id":1,"name":"risk/low"}]`, true, ""},
+		{"risk/medium does not merge", `[{"id":1,"name":"risk/medium"}]`, false, "risk/medium"},
+		{"risk/high does not merge", `[{"id":1,"name":"risk/high"}]`, false, "risk/high"},
+		{"no risk label is unclassified, not low", `[{"id":2,"name":"type/fix"}]`, false, "no risk/"},
+		{"no labels at all", `[]`, false, "no risk/"},
+		{"two risk labels are ambiguous", `[{"id":1,"name":"risk/low"},{"id":2,"name":"risk/high"}]`, false, "more than one"},
+		{"two identical risk/low labels are still ambiguous", `[{"id":1,"name":"risk/low"},{"id":2,"name":"risk/low"}]`, false, "more than one"},
+		{"an unknown value does not merge", `[{"id":1,"name":"risk/lowish"}]`, false, "unknown"},
+		{"the value is case-sensitive: risk/Low", `[{"id":1,"name":"risk/Low"}]`, false, "unknown"},
+		{"trailing space in the value", `[{"id":1,"name":"risk/low "}]`, false, "unknown"},
+		{"RISK/LOW is not a risk/ label", `[{"id":1,"name":"RISK/LOW"}]`, false, "no risk/"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			f := newPRMergeFake(t, prMergeFakeOpts{issueLabels: tc.labels})
+
+			res, err := callMerge(t, f, mergeArgs)
+
+			require.NoError(t, err, "a refusal is a result, not an error")
+			assert.Equal(t, tc.wantMerged, res["merged"])
+			assert.Equal(t, tc.wantMerged, f.merged(), "the merge endpoint is called iff the tier allows it")
+			assert.Equal(t, "success", res["ci_status"], "CI was green: the tier alone decided")
+			if !tc.wantMerged {
+				assert.Contains(t, res["reason"], tc.wantInReason)
+			}
+		})
+	}
+}
+
+func TestPRMerge_NoIssue_Unclassified_NoMerge(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{})
+
+	res, err := callMerge(t, f, `{"owner":"owner","repo":"repo","number":7}`)
+
+	require.NoError(t, err)
+	assert.Equal(t, false, res["merged"])
+	assert.Equal(t, "none", res["risk"])
+	assert.Contains(t, res["reason"], "issue")
+	assert.False(t, f.merged(), "an unclassified change must not merge")
+}
+
+func TestPRMerge_IssueLookupFails_FailsClosed(t *testing.T) {
+	for _, status := range []int{http.StatusInternalServerError, http.StatusNotFound, http.StatusForbidden} {
+		f := newPRMergeFake(t, prMergeFakeOpts{issueStatus: status})
+
+		res, err := callMerge(t, f, mergeArgs)
+
+		require.NoError(t, err, "status %d", status)
+		assert.Equal(t, false, res["merged"], "status %d", status)
+		assert.Equal(t, "unreadable", res["risk"], "status %d", status)
+		assert.False(t, f.merged(), "status %d: an unreadable issue must not merge", status)
+	}
+}
+
+// ---- risk/low is necessary, not sufficient: CI and branch protection still gate ----
+func TestPRMerge_RiskLow_DoesNotOverrideCI(t *testing.T) {
+	for name, runs := range map[string]string{
+		"red":         `[{"id":1,"status":"completed","conclusion":"failure","head_sha":"abc"}]`,
+		"cancelled":   `[{"id":1,"status":"completed","conclusion":"cancelled","head_sha":"abc"}]`,
+		"in_progress": `[{"id":1,"status":"in_progress","conclusion":"","head_sha":"abc"}]`,
+		"no runs":     `[]`,
+		"mixed":       `[{"id":1,"status":"completed","conclusion":"success","head_sha":"abc"},{"id":2,"status":"queued","conclusion":"","head_sha":"abc"}]`,
+	} {
+		t.Run(name, func(t *testing.T) {
+			f := newPRMergeFake(t, prMergeFakeOpts{runs: runs})
+
+			res, err := callMerge(t, f, mergeArgs)
+
+			require.NoError(t, err)
+			assert.Equal(t, false, res["merged"])
+			assert.NotEmpty(t, res["reason"])
+			assert.False(t, f.merged(), "risk/low must not override CI that is not fully green")
+		})
+	}
+}
+
+func TestPRMerge_RiskLow_DoesNotOverrideAReviewProtectedBase(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{requiredApprovals: 1})
+
+	res, err := callMerge(t, f, mergeArgs)
+
+	require.NoError(t, err)
+	assert.Equal(t, false, res["merged"])
+	assert.Contains(t, res["reason"], "review")
+	assert.False(t, f.merged())
+}
+
+// An unreadable PR or an unreadable branch-protection probe is an error, never a merge.
+func TestPRMerge_ReadSideFailures_AreErrorsAndNeverMerge(t *testing.T) {
+	for name, o := range map[string]prMergeFakeOpts{
+		"PR lookup 500":        {prStatus: http.StatusInternalServerError},
+		"PR lookup 404":        {prStatus: http.StatusNotFound},
+		"protection probe 500": {protectionStatus: http.StatusInternalServerError},
+	} {
+		t.Run(name, func(t *testing.T) {
+			f := newPRMergeFake(t, o)
+
+			_, err := callMerge(t, f, mergeArgs)
+
+			require.Error(t, err)
+			assert.False(t, f.merged(), "must not merge when the gate cannot be evaluated")
+		})
+	}
+}
+
+// The classification comes from the issue the caller names, not from `number`: a PR number that is also an issue
+// must not classify itself by accident. (The gate trusts the named issue: documented limit.)
+func TestPRMerge_TheTierIsReadFromTheNamedIssueOnly(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{issueLabels: `[{"id":1,"name":"risk/high"}]`})
+
+	res, err := callMerge(t, f, `{"owner":"owner","repo":"repo","number":7,"issue":42}`)
+
+	require.NoError(t, err)
+	assert.Equal(t, "high", res["risk"])
+	assert.False(t, f.merged())
+}
+
+// ---- review of PR #76 ----
+
+// An empty head sha used to disable BOTH the CI scoping (the runs query lost its head_sha filter, so the repo's most
+// recent runs of any commit were classified) and the merge binding (head_commit_id is omitempty). Fail closed.
+func TestPRMerge_AnEmptyHeadShaFailsClosed(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{emptyHeadSha: true})
+
+	res, err := callMerge(t, f, mergeArgs)
+
+	require.NoError(t, err)
+	assert.Equal(t, false, res["merged"])
+	assert.Contains(t, res["reason"], "head commit")
+	assert.False(t, f.merged(), "an unknown head commit must never merge")
+	assert.Empty(t, f.runsQuery, "no unscoped CI query may be made when the head sha is unknown")
+}
+
+// CI is read for the PR's head commit only: the query names it.
+func TestPRMerge_CIIsScopedToTheHeadCommit(t *testing.T) {
+	f := newPRMergeFake(t, prMergeFakeOpts{})
+
+	_, err := callMerge(t, f, mergeArgs)
+
+	require.NoError(t, err)
+	assert.Contains(t, f.runsQuery, "head_sha=abc")
+}
+
+// Defence in depth: Gitea refuses closed/merged PRs itself, but the gate should not depend on that.
+func TestPRMerge_ClosedOrDraftPRsAreNotMerged(t *testing.T) {
+	for name, o := range map[string]prMergeFakeOpts{
+		"closed": {prState: "closed"},
+		"merged": {prState: "merged"},
+		"draft":  {prDraft: true},
+	} {
+		t.Run(name, func(t *testing.T) {
+			f := newPRMergeFake(t, o)
+
+			res, err := callMerge(t, f, mergeArgs)
+
+			require.NoError(t, err)
+			assert.Equal(t, false, res["merged"])
+			assert.NotEmpty(t, res["reason"])
+			assert.False(t, f.merged())
+		})
+	}
+}
@@ -0,0 +1,116 @@
+package tools_test
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+
+	"git.d-ma.be/mathias/gitea-mcp/internal/allowlist"
+	"git.d-ma.be/mathias/gitea-mcp/internal/gitea"
+	"git.d-ma.be/mathias/gitea-mcp/internal/tools"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// countingServer fails nothing by itself; it records how many requests reached
+// it so a test can assert that a rejected call never touched Gitea.
+func countingServer(t *testing.T) (*httptest.Server, *atomic.Int32) {
+	t.Helper()
+	var n atomic.Int32
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		n.Add(1)
+		w.WriteHeader(http.StatusTeapot)
+	}))
+	t.Cleanup(srv.Close)
+	return srv, &n
+}
+
+// #65 (1): encoding/json matches keys case-insensitively and the last one wins,
+// so a client-side check reading `branch` could see a different value from the
+// one the server acts on. Any two keys equal under strings.EqualFold, and any
+// key that differs from a schema field only in case, are rejected before a
+// request is made.
+func TestParseArgsRejectsCaseVariantKeys(t *testing.T) {
+	cases := map[string]string{
+		"issue repro: branch + Branch": `{"owner":"probe","repo":"t","name":"t","path":"case8.txt","content":"x","message":"m","branch":"feat/y8","Branch":"main"}`,
+		"exact duplicate key":          `{"owner":"probe","repo":"t","path":"p","content":"x","message":"m","branch":"feat/y8","branch":"main"}`,
+		"case-only schema mismatch":    `{"owner":"probe","repo":"t","path":"p","content":"x","message":"m","BRANCH":"main"}`,
+		"nested object duplicate":      `{"owner":"probe","repo":"t","path":"p","content":"x","message":"m","branch":"feat/y8","extra":{"a":1,"A":2}}`,
+		// PR #66 review: a number the token scanner cannot parse as float64 must
+		// not make the scan give up and let the duplicate through.
+		"1e999 decoy before duplicate":  `{"owner":"probe","repo":"t","name":"t","path":"case8.txt","content":"x","message":"m","z":1e999,"branch":"feat/y8","Branch":"main"}`,
+		"-1e309 decoy before exact dup": `{"owner":"probe","repo":"t","path":"p","content":"x","message":"m","z":-1e309,"branch":"feat/y8","branch":"main"}`,
+		// encoding/json folds like strings.EqualFold: long s (U+017F) == s.
+		"nested long-s duplicate": `{"owner":"probe","repo":"t","path":"p","content":"x","message":"m","branch":"feat/y8","extra":{"s":1,"\u017f":2}}`,
+	}
+	for name, raw := range cases {
+		t.Run(name, func(t *testing.T) {
+			srv, n := countingServer(t)
+			tool := tools.NewFileWriteBranch(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"probe"}))
+			_, err := tool.Call(context.Background(), json.RawMessage(raw))
+			require.Error(t, err)
+			assert.ErrorIs(t, err, gitea.ErrValidation)
+			assert.Zero(t, n.Load(), "no request may reach gitea")
+		})
+	}
+}
+
+// #65 (2): `refs/heads/main` as a branch makes Gitea create
+// refs/heads/refs/heads/main, after which every API write to main 500s.
+// Branch names that can create such refs are rejected before any request.
+func TestBranchNameValidationRejectsBeforeRequest(t *testing.T) {
+	bad := []string{
+		"refs/heads/main", "refs/tags/v1", "a..b", "feat x", "feat\tx", "feat/",
+		"/feat", "-feat", ".feat", "feat//x", "feat/.x", "feat.lock", "feat.", "HEAD", "feat@{1}", "",
+	}
+	for _, b := range bad {
+		t.Run("file_write_branch "+b, func(t *testing.T) {
+			srv, n := countingServer(t)
+			tool := tools.NewFileWriteBranch(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"probe"}))
+			args, _ := json.Marshal(map[string]string{
+				"owner": "probe", "repo": "t", "path": "p", "content": "x", "message": "m", "branch": b,
+			})
+			_, err := tool.Call(context.Background(), args)
+			require.Error(t, err)
+			assert.ErrorIs(t, err, gitea.ErrValidation)
+			assert.Zero(t, n.Load(), "no request may reach gitea")
+		})
+		if b == "" {
+			continue // tbd_ship derives a branch when empty
+		}
+		t.Run("tbd_ship "+b, func(t *testing.T) {
+			srv, n := countingServer(t)
+			tool := tools.NewTBDShip(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"probe"}))
+			args, _ := json.Marshal(map[string]string{
+				"owner": "probe", "repo": "t", "path": "p", "content": "x", "message": "m", "branch": b,
+			})
+			_, err := tool.Call(context.Background(), args)
+			require.Error(t, err)
+			assert.ErrorIs(t, err, gitea.ErrValidation)
+			assert.Zero(t, n.Load(), "no request may reach gitea")
+		})
+	}
+}
+
+// tbd_ship's base is the branch-protection probe key: `refs/heads/main` would
+// read as unprotected, so it is held to the same rule as the head branch.
+func TestTBDShipRejectsRefsPrefixedBase(t *testing.T) {
+	srv, n := countingServer(t)
+	tool := tools.NewTBDShip(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"probe"}))
+	_, err := tool.Call(context.Background(), json.RawMessage(
+		`{"owner":"probe","repo":"t","path":"p","content":"x","message":"m","base":"refs/heads/main"}`))
+	require.Error(t, err)
+	assert.ErrorIs(t, err, gitea.ErrValidation)
+	assert.Zero(t, n.Load(), "no request may reach gitea")
+}
+
+func TestBranchNameValidationAcceptsOrdinaryNames(t *testing.T) {
+	for _, b := range []string{"main", "feat/y8", "tbd/fix-abc-1234abcd", "hold/mail-chat-infra-21", "v1.2_x"} {
+		t.Run(b, func(t *testing.T) {
+			assert.NoError(t, tools.ValidateBranchName(b))
+		})
+	}
+}
@@ -37,7 +37,7 @@ const (
 func (t *TBDShip) Descriptor() registry.ToolDescriptor {
 	return registry.ToolDescriptor{
 		Name:        "tbd_ship",
-		Description: "Trunk-based ship of a single-file change: branch from base, write the file, open a PR, and auto-merge (squash) to base ONLY when the head commit's CI is fully green. Fails closed to PR-only (never merges) when CI is pending, red, or absent, when the base branch requires reviews, or when the merge isn't clean — returning the PR for manual handling with a reason. Set ci_timeout_seconds to poll CI to completion (default 0 = snapshot, which returns pending right after opening the PR). Returns {merged, pr_url, pr_number, ci_status, reason, branch}.",
+		Description: "Trunk-based ship of a single-file change: branch from base, write the file, open a PR, and auto-merge (squash) to base ONLY when the head commit's CI is fully green AND the named issue carries exactly one risk/ label, risk/low. Fails closed to PR-only (never merges) when CI is pending, red, or absent, when no issue is named or its risk/ label is missing, ambiguous, unknown, risk/medium or risk/high, when the issue cannot be read, when the base branch requires reviews, or when the merge isn't clean — returning the PR for manual handling with a reason. Set ci_timeout_seconds to poll CI to completion (default 0 = snapshot, which returns pending right after opening the PR). Returns {merged, pr_url, pr_number, ci_status, risk, reason, branch}.",
 		InputSchema: json.RawMessage(`{
 			"type":"object",
 			"properties":{
@@ -50,6 +50,7 @@ func (t *TBDShip) Descriptor() registry.ToolDescriptor {
 				"branch":{"type":"string","description":"Short-lived branch name. Default derived: tbd/<slug>-<hash>."},
 				"pr_title":{"type":"string","description":"PR title. Default: the commit message."},
 				"pr_body":{"type":"string"},
+				"issue":{"type":"integer","minimum":1,"description":"Issue number in the same repo whose risk/ label classifies this change. Auto-merge happens only for exactly one label, risk/low; no issue, none, several, an unknown value, risk/medium or risk/high means PR only, never merged (unclassified is not low)."},
 				"delete_branch":{"type":"boolean","description":"Delete the branch after a successful merge. Default true."},
 				"ci_timeout_seconds":{"type":"integer","minimum":0,"maximum":600,"description":"Poll CI up to this long for the head commit's runs to complete. Default 0 (single snapshot)."}
 			},
@@ -68,10 +69,76 @@ type tbdShipArgs struct {
 	Branch       string `json:"branch"`
 	PRTitle      string `json:"pr_title"`
 	PRBody       string `json:"pr_body"`
+	Issue        int    `json:"issue"`
 	DeleteBranch *bool  `json:"delete_branch"`
 	CITimeoutSec int    `json:"ci_timeout_seconds"`
 }

+// riskTier is the change's classification under the lane model (local-dev .context/AGENT.md rule 5): only a change
+// whose issue carries exactly one risk/ label, risk/low, may merge on green; everything else is reviewed by a human
+// (gitea-mcp#63). The zero value is deliberately NOT riskLow: a forgotten field must never mean "low".
+type riskTier string
+
+const (
+	riskLow        riskTier = "low"
+	riskMedium     riskTier = "medium"
+	riskHigh       riskTier = "high"
+	riskNone       riskTier = "none"       // no issue named, or the issue has no risk/ label: unclassified is not low
+	riskAmbiguous  riskTier = "ambiguous"  // more than one risk/ label
+	riskUnknown    riskTier = "unknown"    // risk/<something else>
+	riskUnreadable riskTier = "unreadable" // the issue lookup failed
+)
+
+// classifyRisk reads the tier from an issue's labels. Exact, case-sensitive prefix "risk/"; exactly one such label
+// is required, so two labels (even identical ones) are ambiguous rather than "the lower one wins".
+func classifyRisk(labels []gitea.Label) riskTier {
+	var found []string
+	for _, l := range labels {
+		if strings.HasPrefix(l.Name, "risk/") {
+			found = append(found, strings.TrimPrefix(l.Name, "risk/"))
+		}
+	}
+	switch len(found) {
+	case 0:
+		return riskNone
+	case 1:
+	default:
+		return riskAmbiguous
+	}
+	switch found[0] {
+	case "low":
+		return riskLow
+	case "medium":
+		return riskMedium
+	case "high":
+		return riskHigh
+	default:
+		return riskUnknown
+	}
+}
+
+const unclassifiedReason = "no risk/ label on the named issue (or no issue named) — unclassified is not low; PR left open, not merged"
+
+// riskBlockReason explains why a tier other than riskLow blocks the merge.
+func riskBlockReason(r riskTier) string {
+	switch r {
+	case riskMedium, riskHigh:
+		return fmt.Sprintf("the issue is risk/%s — above risk/low a human reviews the change; PR left open, not merged", r)
+	case riskAmbiguous:
+		return "the issue carries more than one risk/ label — ambiguous classification; PR left open, not merged"
+	case riskUnknown:
+		return "the issue carries an unknown risk/ value — PR left open, not merged"
+	case riskUnreadable:
+		return "could not read the issue's labels — fail-closed; PR left open, not merged"
+	case riskLow:
+		return "" // never asked: only a non-low tier needs explaining
+	case riskNone:
+		return unclassifiedReason
+	default: // the zero value and any tier added later: unclassified is not low
+		return unclassifiedReason
+	}
+}
+
 // shipGate is the decision produced by evaluateShipGate.
 type shipGate struct {
 	merge    bool
@@ -101,29 +168,46 @@ func classifyCI(runs []gitea.WorkflowRun) string {
 }

 // evaluateShipGate is the load-bearing safety of tbd_ship. It returns merge=true
-// only when CI is fully green AND the base branch is not review-protected. Every
-// other state fails closed to PR-only with an explanatory reason (#40).
-func evaluateShipGate(runs []gitea.WorkflowRun, bp *gitea.BranchProtection) shipGate {
+// only when CI is fully green AND the base branch is not review-protected AND the
+// change is classified risk/low. Every other state fails closed to PR-only with an
+// explanatory reason (#40, gitea-mcp#63).
+func evaluateShipGate(runs []gitea.WorkflowRun, bp *gitea.BranchProtection, risk riskTier) shipGate {
 	ci := classifyCI(runs)

 	// Review-protected base can't be auto-merged regardless of CI.
 	if bp != nil && bp.Protected && bp.RequiredApprovals > 0 {
 		return shipGate{false, ci, fmt.Sprintf(
-			"base branch requires %d approving review(s) — auto-merge disabled; PR opened for manual merge", bp.RequiredApprovals)}
+			"base branch requires %d approving review(s) — auto-merge disabled; PR left open for manual merge", bp.RequiredApprovals)}
+	}
+
+	// Only risk/low merges on green. Checked before CI so an unclassified change never reads as "waiting for CI".
+	if risk != riskLow {
+		return shipGate{false, ci, riskBlockReason(risk)}
 	}

 	switch ci {
 	case "success":
 		return shipGate{true, ci, ""}
 	case "pending":
-		return shipGate{false, ci, "CI is still running for the head commit — PR opened; re-invoke once checks complete, or merge manually"}
+		return shipGate{false, ci, "CI is still running for the head commit — PR left open; re-invoke once checks complete, or merge manually"}
 	case "failed":
-		return shipGate{false, ci, "CI failed for the head commit — PR opened, not merged"}
+		return shipGate{false, ci, "CI failed for the head commit — PR left open, not merged"}
 	default: // none
-		return shipGate{false, ci, "no CI runs found for the head commit — fail-closed: a change with no CI gate is never auto-merged to trunk; PR opened for manual merge"}
+		return shipGate{false, ci, "no CI runs found for the head commit — fail-closed: a change with no CI gate is never auto-merged to trunk; PR left open for manual merge"}
 	}
 }

+// evaluateMergeGate is what both merge tools call. It adds one precondition to evaluateShipGate: the PR's head commit
+// must be known. CI is read for that commit and the merge is bound to it (head_commit_id); with an empty sha the runs
+// query would lose its filter (the repo's most recent runs of ANY commit would be classified) and the binding would be
+// dropped, so an unknown head commit never merges.
+func evaluateMergeGate(headSHA string, runs []gitea.WorkflowRun, bp *gitea.BranchProtection, risk riskTier) shipGate {
+	if headSHA == "" {
+		return shipGate{false, "none", "the PR's head commit is unknown — fail-closed (CI cannot be scoped and the merge cannot be bound to it); PR left open, not merged"}
+	}
+	return evaluateShipGate(runs, bp, risk)
+}
+
 var shipSlugRe = regexp.MustCompile(`[^a-z0-9]+`)

 // deriveBranch builds a deterministic short-lived branch name from the change,
@@ -160,6 +244,13 @@ func (t *TBDShip) Call(ctx context.Context, raw json.RawMessage) (json.RawMessag
 	if branch == "" {
 		branch = deriveBranch(args.Message, args.Path, args.Content)
 	}
+	// Both names reach a ref: base is also the protection-probe key, so a
+	// `refs/heads/main` base would read as unprotected (#65).
+	for _, b := range []string{base, branch} {
+		if err := ValidateBranchName(b); err != nil {
+			return nil, err
+		}
+	}
 	deleteBranch := true
 	if args.DeleteBranch != nil {
 		deleteBranch = *args.DeleteBranch
@@ -222,8 +313,9 @@ func (t *TBDShip) Call(ctx context.Context, raw json.RawMessage) (json.RawMessag
 	}

 	// CI gate on the PR head commit.
-	runs := t.pollRuns(ctx, args.Owner, args.Repo, pr.Head.Sha, args.CITimeoutSec)
-	gate := evaluateShipGate(runs, bp)
+	runs := pollRuns(ctx, t.c, args.Owner, args.Repo, pr.Head.Sha, args.CITimeoutSec)
+	risk := riskTierOf(ctx, t.c, args.Owner, args.Repo, args.Issue)
+	gate := evaluateMergeGate(pr.Head.Sha, runs, bp, risk)

 	result := map[string]any{
 		"merged":    false,
@@ -231,6 +323,7 @@ func (t *TBDShip) Call(ctx context.Context, raw json.RawMessage) (json.RawMessag
 		"pr_url":    pr.HTMLURL,
 		"branch":    branch,
 		"ci_status": gate.ciStatus,
+		"risk":      string(risk),
 	}
 	if !gate.merge {
 		result["reason"] = gate.reason
@@ -238,9 +331,9 @@ func (t *TBDShip) Call(ctx context.Context, raw json.RawMessage) (json.RawMessag
 	}

 	// Green + unprotected → squash-merge, then delete the short-lived branch.
-	if err := t.c.MergePullRequest(ctx, args.Owner, args.Repo, pr.Number, gitea.MergePRArgs{Do: "squash"}); err != nil {
+	if err := t.c.MergePullRequest(ctx, args.Owner, args.Repo, pr.Number, gitea.MergePRArgs{Do: "squash", HeadCommitID: pr.Head.Sha}); err != nil {
 		if errors.Is(err, gitea.ErrConflict) {
-			result["reason"] = "base moved and the merge is not clean — PR opened, not merged; resolve the conflict and merge manually"
+			result["reason"] = "base moved and the merge is not clean — PR left open, not merged; resolve the conflict and merge manually"
 			return textOK(result)
 		}
 		return nil, fmt.Errorf("merge PR #%d: %w", pr.Number, err)
@@ -254,10 +347,27 @@ func (t *TBDShip) Call(ctx context.Context, raw json.RawMessage) (json.RawMessag
 	return textOK(result)
 }

+// riskTierOf classifies the change from the named issue. No issue, or an issue that cannot be read, never yields
+// riskLow. The link between the PR and the issue is the caller's claim: the gate enforces the tier of the issue it
+// is given, it cannot prove that issue is the one the change belongs to (stated limit, gitea-mcp#63).
+func riskTierOf(ctx context.Context, c *gitea.Client, owner, repo string, issue int) riskTier {
+	if issue <= 0 {
+		return riskNone
+	}
+	iss, err := c.GetIssue(ctx, owner, repo, issue)
+	if err != nil || iss == nil {
+		return riskUnreadable
+	}
+	return classifyRisk(iss.Labels)
+}
+
 // pollRuns lists the head commit's workflow runs, polling up to timeoutSec for
 // them to reach a terminal state. timeoutSec == 0 is a single snapshot (no
 // sleep). A listing error yields no runs → the gate fails closed.
-func (t *TBDShip) pollRuns(ctx context.Context, owner, repo, headSHA string, timeoutSec int) []gitea.WorkflowRun {
+func pollRuns(ctx context.Context, c *gitea.Client, owner, repo, headSHA string, timeoutSec int) []gitea.WorkflowRun {
+	if headSHA == "" {
+		return nil // never query without the commit filter: Gitea would return the repo's latest runs of any commit
+	}
 	if timeoutSec < 0 {
 		timeoutSec = 0
 	}
@@ -266,7 +376,7 @@ func (t *TBDShip) pollRuns(ctx context.Context, owner, repo, headSHA string, tim
 	}
 	deadline := time.Now().Add(time.Duration(timeoutSec) * time.Second)
 	for {
-		runs := t.listRuns(ctx, owner, repo, headSHA)
+		runs := listRuns(ctx, c, owner, repo, headSHA)
 		switch classifyCI(runs) {
 		case "success", "failed":
 			return runs // terminal — no point waiting
@@ -297,8 +407,8 @@ func (t *TBDShip) findOpenPR(ctx context.Context, owner, repo, branch string) (*
 	return nil, fmt.Errorf("no open PR found for head %s: %w", branch, gitea.ErrNotFound)
 }

-func (t *TBDShip) listRuns(ctx context.Context, owner, repo, headSHA string) []gitea.WorkflowRun {
-	resp, err := t.c.ListWorkflowRuns(ctx, owner, repo, gitea.ListWorkflowRunsArgs{HeadSHA: headSHA, Limit: 50})
+func listRuns(ctx context.Context, c *gitea.Client, owner, repo, headSHA string) []gitea.WorkflowRun {
+	resp, err := c.ListWorkflowRuns(ctx, owner, repo, gitea.ListWorkflowRunsArgs{HeadSHA: headSHA, Limit: 50})
 	if err != nil || resp == nil {
 		return nil
 	}
@@ -1,6 +1,11 @@
 package tools

 import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
 	"testing"

 	"git.d-ma.be/mathias/gitea-mcp/internal/gitea"
@@ -20,22 +25,31 @@ func TestEvaluateShipGate(t *testing.T) {
 		name      string
 		runs      []gitea.WorkflowRun
 		bp        *gitea.BranchProtection
+		risk      riskTier
 		wantMerge bool
 		wantCI    string
 	}{
-		{"all green → merge", []gitea.WorkflowRun{run("completed", "success")}, open, true, "success"},
-		{"in_progress → no merge", []gitea.WorkflowRun{run("in_progress", "")}, open, false, "pending"},
-		{"queued → no merge", []gitea.WorkflowRun{run("queued", "")}, open, false, "pending"},
-		{"failed → no merge", []gitea.WorkflowRun{run("completed", "failure")}, open, false, "failed"},
-		{"cancelled → no merge (fail-closed)", []gitea.WorkflowRun{run("completed", "cancelled")}, open, false, "failed"},
-		{"no runs → no merge (no CI gate)", nil, open, false, "none"},
-		{"mixed success+queued → no merge", []gitea.WorkflowRun{run("completed", "success"), run("queued", "")}, open, false, "pending"},
-		{"green but protected → no merge", []gitea.WorkflowRun{run("completed", "success")}, protected, false, "success"},
-		{"no runs + protected → no merge", nil, protected, false, "none"},
+		{"all green → merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskLow, true, "success"},
+		{"in_progress → no merge", []gitea.WorkflowRun{run("in_progress", "")}, open, riskLow, false, "pending"},
+		{"queued → no merge", []gitea.WorkflowRun{run("queued", "")}, open, riskLow, false, "pending"},
+		{"failed → no merge", []gitea.WorkflowRun{run("completed", "failure")}, open, riskLow, false, "failed"},
+		{"cancelled → no merge (fail-closed)", []gitea.WorkflowRun{run("completed", "cancelled")}, open, riskLow, false, "failed"},
+		{"no runs → no merge (no CI gate)", nil, open, riskLow, false, "none"},
+		{"mixed success+queued → no merge", []gitea.WorkflowRun{run("completed", "success"), run("queued", "")}, open, riskLow, false, "pending"},
+		{"green but protected → no merge", []gitea.WorkflowRun{run("completed", "success")}, protected, riskLow, false, "success"},
+		{"no runs + protected → no merge", nil, protected, riskLow, false, "none"},
+		// gitea-mcp#63: the tier is an input. Green CI never overrides it; unclassified is not low.
+		{"green + risk/medium → no merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskMedium, false, "success"},
+		{"green + risk/high → no merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskHigh, false, "success"},
+		{"green + no risk label → no merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskNone, false, "success"},
+		{"green + ambiguous labels → no merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskAmbiguous, false, "success"},
+		{"green + unknown value → no merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskUnknown, false, "success"},
+		{"green + unreadable issue → no merge", []gitea.WorkflowRun{run("completed", "success")}, open, riskUnreadable, false, "success"},
+		{"green + the zero tier → no merge (a forgotten field must not mean low)", []gitea.WorkflowRun{run("completed", "success")}, open, riskTier(""), false, "success"},
 	}
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			g := evaluateShipGate(tc.runs, tc.bp)
+			g := evaluateShipGate(tc.runs, tc.bp, tc.risk)
 			if g.merge != tc.wantMerge {
 				t.Errorf("merge = %v, want %v (reason: %q)", g.merge, tc.wantMerge, g.reason)
 			}
@@ -51,3 +65,74 @@ func TestEvaluateShipGate(t *testing.T) {
 		})
 	}
 }
+
+func TestClassifyRisk(t *testing.T) {
+	l := func(names ...string) []gitea.Label {
+		out := make([]gitea.Label, len(names))
+		for i, n := range names {
+			out[i] = gitea.Label{Name: n}
+		}
+		return out
+	}
+	tests := []struct {
+		name   string
+		labels []gitea.Label
+		want   riskTier
+	}{
+		{"low", l("risk/low"), riskLow},
+		{"medium", l("risk/medium"), riskMedium},
+		{"high", l("risk/high"), riskHigh},
+		{"low among others", l("type/fix", "risk/low", "size/s"), riskLow},
+		{"none", l("type/fix"), riskNone},
+		{"no labels", nil, riskNone},
+		{"two different", l("risk/low", "risk/high"), riskAmbiguous},
+		{"two identical", l("risk/low", "risk/low"), riskAmbiguous},
+		{"unknown value", l("risk/extreme"), riskUnknown},
+		{"empty value", l("risk/"), riskUnknown},
+		{"wrong case", l("Risk/Low"), riskNone},
+		{"value case matters: risk/Low is unknown, not low", l("risk/Low"), riskUnknown},
+		{"trailing space in the value is unknown, not low", l("risk/low "), riskUnknown},
+		{"leading space in the value is unknown, not low", l("risk/ low"), riskUnknown},
+		{"prefix only inside a name", l("not-risk/low"), riskNone},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := classifyRisk(tc.labels); got != tc.want {
+				t.Errorf("classifyRisk(%v) = %q, want %q", tc.labels, got, tc.want)
+			}
+		})
+	}
+}
+
+// The merge gate needs a head commit to scope CI to and to bind the merge to: without one nothing merges.
+func TestEvaluateMergeGate_UnknownHeadCommitNeverMerges(t *testing.T) {
+	green := []gitea.WorkflowRun{{Status: "completed", Conclusion: "success"}}
+	g := evaluateMergeGate("", green, &gitea.BranchProtection{}, riskLow)
+	if g.merge {
+		t.Fatalf("merge = true for an empty head sha (reason %q)", g.reason)
+	}
+	if !strings.Contains(g.reason, "head commit") {
+		t.Errorf("reason %q should name the head commit", g.reason)
+	}
+	if ok := evaluateMergeGate("abc", green, &gitea.BranchProtection{}, riskLow); !ok.merge {
+		t.Errorf("a known head sha with green CI and risk/low must still merge, got %q", ok.reason)
+	}
+}
+
+func TestPollRuns_NoShaMakesNoQuery(t *testing.T) {
+	var requests atomic.Int32
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		requests.Add(1)
+		_, _ = w.Write([]byte(`{"total_count":1,"workflow_runs":[{"id":1,"status":"completed","conclusion":"success"}]}`))
+	}))
+	defer srv.Close()
+	c := gitea.NewClient(srv.URL, "tok")
+
+	if runs := pollRuns(context.Background(), c, "o", "r", "", 0); len(runs) != 0 || requests.Load() != 0 {
+		t.Errorf("an empty sha must make no request and return no runs, got %d runs after %d requests", len(runs), requests.Load())
+	}
+	// positive control: with a sha the same server is asked, so the zero above means something
+	if runs := pollRuns(context.Background(), c, "o", "r", "abc", 0); len(runs) != 1 || requests.Load() != 1 {
+		t.Errorf("control: with a sha expected 1 run after 1 request, got %d runs after %d requests", len(runs), requests.Load())
+	}
+}
@@ -3,6 +3,7 @@ package tools_test
 import (
 	"context"
 	"encoding/json"
+	"io"
 	"net/http"
 	"net/http/httptest"
 	"strings"
@@ -26,11 +27,21 @@ func shipFake(t *testing.T, runsJSON string, merged, deleted *atomic.Bool) *http
 // shipFakeOpts adds protection (requiredApprovals>0 → protected) and a merge
 // status code, so tests can drive the review-protected and conflict paths.
 func shipFakeOpts(t *testing.T, runsJSON string, requiredApprovals, mergeStatus int, merged, deleted *atomic.Bool) *httptest.Server {
+	t.Helper()
+	return shipFakeFull(t, runsJSON, requiredApprovals, mergeStatus, `[{"id":1,"name":"risk/low"}]`, http.StatusOK, merged, deleted)
+}
+
+// shipFakeFull also serves the issue whose risk/ label classifies the change (gitea-mcp#63): issueLabels is the
+// labels JSON array, issueStatus the HTTP status of the issue lookup.
+func shipFakeFull(t *testing.T, runsJSON string, requiredApprovals, mergeStatus int, issueLabels string, issueStatus int, merged, deleted *atomic.Bool) *httptest.Server {
 	t.Helper()
 	return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		p := r.URL.Path
 		w.Header().Set("Content-Type", "application/json")
 		switch {
+		case r.Method == http.MethodGet && strings.Contains(p, "/issues/42"):
+			w.WriteHeader(issueStatus)
+			_, _ = w.Write([]byte(`{"number":42,"title":"t","state":"open","labels":` + issueLabels + `}`))
 		case r.Method == http.MethodGet && strings.Contains(p, "/branch_protections/"):
 			if requiredApprovals > 0 {
 				_, _ = w.Write([]byte(`{"required_approvals":` + itoa(requiredApprovals) + `}`))
@@ -49,10 +60,12 @@ func shipFakeOpts(t *testing.T, runsJSON string, requiredApprovals, mergeStatus
 			_, _ = w.Write([]byte(`{"content":{"path":"x","sha":"s"},"commit":{"sha":"c"}}`))
 		case r.Method == http.MethodPost && strings.HasSuffix(p, "/pulls"):
 			w.WriteHeader(http.StatusCreated)
-			_, _ = w.Write([]byte(`{"number":7,"title":"t","html_url":"http://x/pulls/7","state":"open","head":{"ref":"tbd/x","sha":"abc"},"base":{"ref":"main"}}`))
+			_, _ = w.Write([]byte(`{"number":7,"title":"t","html_url":"http://x/pulls/7","state":"open","head":{"ref":"tbd/x","sha":"` + shipPRHeadSha + `"},"base":{"ref":"main"}}`))
 		case r.Method == http.MethodGet && strings.Contains(p, "/actions/runs"):
 			_, _ = w.Write([]byte(`{"total_count":0,"workflow_runs":` + runsJSON + `}`))
 		case r.Method == http.MethodPost && strings.HasSuffix(p, "/pulls/7/merge"):
+			body, _ := io.ReadAll(r.Body)
+			shipLastMergeBody.Store(string(body))
 			merged.Store(true)
 			w.WriteHeader(mergeStatus)
 			_, _ = w.Write([]byte(`{}`))
@@ -83,7 +96,7 @@ func TestTBDShip_GreenCI_Merges(t *testing.T) {
 	srv := shipFake(t, `[{"id":1,"status":"completed","conclusion":"success","head_sha":"abc"}]`, &merged, &deleted)
 	defer srv.Close()

-	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x"}`)
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42}`)

 	assert.Equal(t, true, res["merged"])
 	assert.Equal(t, "success", res["ci_status"])
@@ -98,7 +111,7 @@ func TestTBDShip_NoCI_FailsClosed(t *testing.T) {
 	srv := shipFake(t, `[]`, &merged, &deleted)
 	defer srv.Close()

-	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x"}`)
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42}`)

 	assert.Equal(t, false, res["merged"])
 	assert.Equal(t, "none", res["ci_status"])
@@ -113,7 +126,7 @@ func TestTBDShip_RedCI_FailsClosed(t *testing.T) {
 	srv := shipFake(t, `[{"id":1,"status":"completed","conclusion":"failure","head_sha":"abc"}]`, &merged, &deleted)
 	defer srv.Close()

-	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x"}`)
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42}`)

 	assert.Equal(t, false, res["merged"])
 	assert.Equal(t, "failed", res["ci_status"])
@@ -128,7 +141,7 @@ func TestTBDShip_ReviewProtected_FailsClosed(t *testing.T) {
 	srv := shipFakeOpts(t, `[{"id":1,"status":"completed","conclusion":"success","head_sha":"abc"}]`, 1, http.StatusOK, &merged, &deleted)
 	defer srv.Close()

-	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x"}`)
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42}`)

 	assert.Equal(t, false, res["merged"])
 	assert.NotEmpty(t, res["reason"])
@@ -142,7 +155,7 @@ func TestTBDShip_MergeConflict_FailsClosed(t *testing.T) {
 	srv := shipFakeOpts(t, `[{"id":1,"status":"completed","conclusion":"success","head_sha":"abc"}]`, 0, http.StatusConflict, &merged, &deleted)
 	defer srv.Close()

-	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x"}`)
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42}`)

 	assert.Equal(t, false, res["merged"])
 	assert.NotEmpty(t, res["reason"])
@@ -160,6 +173,8 @@ func TestTBDShip_Resume_ExistingBranchAndPR(t *testing.T) {
 		p := r.URL.Path
 		w.Header().Set("Content-Type", "application/json")
 		switch {
+		case r.Method == http.MethodGet && strings.Contains(p, "/issues/42"):
+			_, _ = w.Write([]byte(`{"number":42,"title":"t","state":"open","labels":[{"id":1,"name":"risk/low"}]}`))
 		case r.Method == http.MethodGet && strings.Contains(p, "/branch_protections/"):
 			w.WriteHeader(http.StatusNotFound)
 			_, _ = w.Write([]byte(`{"message":"not found"}`))
@@ -196,7 +211,7 @@ func TestTBDShip_Resume_ExistingBranchAndPR(t *testing.T) {
 	}))
 	defer srv.Close()

-	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","branch":"`+branch+`"}`)
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42,"branch":"`+branch+`"}`)

 	assert.Equal(t, true, res["merged"], "resume must merge when CI is green")
 	assert.Equal(t, float64(7), res["pr_number"], "must reuse the existing PR #7")
@@ -204,6 +219,138 @@ func TestTBDShip_Resume_ExistingBranchAndPR(t *testing.T) {
 	assert.True(t, merged.Load())
 }

+// shipPRHeadSha is the head sha the fake reports for the PR it opens (tests may blank it, then restore it).
+var shipPRHeadSha = "abc"
+
+// shipLastMergeBody is the body of the most recent merge request the fake received.
+var shipLastMergeBody atomic.Value
+
+// An empty head sha used to unscope the CI query and drop the merge binding (review of PR #76). Fail closed.
+func TestTBDShip_AnEmptyHeadShaFailsClosed(t *testing.T) {
+	shipPRHeadSha = ""
+	t.Cleanup(func() { shipPRHeadSha = "abc" })
+	var merged, deleted atomic.Bool
+	srv := shipFake(t, shipGreen, &merged, &deleted)
+	defer srv.Close()
+
+	res := callShip(t, srv.URL, shipArgs)
+
+	assert.Equal(t, false, res["merged"])
+	assert.Contains(t, res["reason"], "head commit")
+	assert.False(t, merged.Load(), "an unknown head commit must never merge")
+}
+
+// TOCTOU: the merge is bound to the head commit whose CI the gate evaluated (gitea-mcp#74).
+func TestTBDShip_TheMergeIsBoundToTheCommitTheGateChecked(t *testing.T) {
+	var merged, deleted atomic.Bool
+	srv := shipFake(t, shipGreen, &merged, &deleted)
+	defer srv.Close()
+
+	res := callShip(t, srv.URL, shipArgs)
+
+	require.Equal(t, true, res["merged"])
+	var payload map[string]any
+	require.NoError(t, json.Unmarshal([]byte(shipLastMergeBody.Load().(string)), &payload))
+	assert.Equal(t, "abc", payload["head_commit_id"], "the head sha the CI check used")
+}
+
+// ---- gitea-mcp#63: the risk tier is an input to the gate. Only exactly one risk/low label merges. ----
+const (
+	shipGreen = `[{"id":1,"status":"completed","conclusion":"success","head_sha":"abc"}]`
+	shipArgs  = `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x","issue":42}`
+)
+
+func TestTBDShip_RiskTier_OnlyLowMerges(t *testing.T) {
+	tests := []struct {
+		name         string
+		labels       string
+		wantMerged   bool
+		wantInReason string
+	}{
+		{"risk/low merges", `[{"id":1,"name":"risk/low"}]`, true, ""},
+		{"risk/low among other labels merges", `[{"id":2,"name":"type/fix"},{"id":1,"name":"risk/low"},{"id":3,"name":"size/s"}]`, true, ""},
+		{"risk/medium does not merge", `[{"id":1,"name":"risk/medium"}]`, false, "risk/medium"},
+		{"risk/high does not merge", `[{"id":1,"name":"risk/high"}]`, false, "risk/high"},
+		{"no risk label is unclassified, not low", `[{"id":2,"name":"type/fix"}]`, false, "no risk/"},
+		{"no labels at all", `[]`, false, "no risk/"},
+		{"two risk labels are ambiguous", `[{"id":1,"name":"risk/low"},{"id":2,"name":"risk/high"}]`, false, "more than one"},
+		{"two identical risk/low labels are still ambiguous", `[{"id":1,"name":"risk/low"},{"id":2,"name":"risk/low"}]`, false, "more than one"},
+		{"an unknown risk value does not merge", `[{"id":1,"name":"risk/lowish"}]`, false, "unknown"},
+		{"an empty risk value does not merge", `[{"id":1,"name":"risk/"}]`, false, "unknown"},
+		{"labels are case-sensitive: RISK/LOW is not risk/low", `[{"id":1,"name":"RISK/LOW"}]`, false, "no risk/"},
+		{"a label that merely contains risk/low does not count", `[{"id":1,"name":"not-risk/low"}]`, false, "no risk/"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			var merged, deleted atomic.Bool
+			srv := shipFakeFull(t, shipGreen, 0, http.StatusOK, tc.labels, http.StatusOK, &merged, &deleted)
+			defer srv.Close()
+
+			res := callShip(t, srv.URL, shipArgs)
+
+			assert.Equal(t, tc.wantMerged, res["merged"])
+			assert.Equal(t, tc.wantMerged, merged.Load(), "the merge endpoint is called iff the tier allows it")
+			assert.Equal(t, "success", res["ci_status"], "CI was green: the tier alone decided")
+			assert.Equal(t, float64(7), res["pr_number"], "the PR is opened either way")
+			if !tc.wantMerged {
+				assert.Contains(t, res["reason"], tc.wantInReason)
+			}
+		})
+	}
+}
+
+// No issue named → nothing classifies the change → no merge, even on green CI.
+func TestTBDShip_NoIssue_Unclassified_NoMerge(t *testing.T) {
+	var merged, deleted atomic.Bool
+	srv := shipFake(t, shipGreen, &merged, &deleted)
+	defer srv.Close()
+
+	res := callShip(t, srv.URL, `{"owner":"mathias","repo":"myrepo","path":"docs/x.md","content":"hi","message":"add x"}`)
+
+	assert.Equal(t, false, res["merged"])
+	assert.Contains(t, res["reason"], "issue")
+	assert.False(t, merged.Load(), "an unclassified change must not merge")
+	assert.Equal(t, float64(7), res["pr_number"])
+}
+
+// The issue lookup failing is not a reason to merge: fail closed, PR left open.
+func TestTBDShip_IssueLookupFails_FailsClosed(t *testing.T) {
+	for _, status := range []int{http.StatusInternalServerError, http.StatusNotFound, http.StatusForbidden} {
+		var merged, deleted atomic.Bool
+		srv := shipFakeFull(t, shipGreen, 0, http.StatusOK, `[{"id":1,"name":"risk/low"}]`, status, &merged, &deleted)
+
+		res := callShip(t, srv.URL, shipArgs)
+		srv.Close()
+
+		assert.Equal(t, false, res["merged"], "status %d", status)
+		assert.NotEmpty(t, res["reason"], "status %d", status)
+		assert.False(t, merged.Load(), "status %d: an unreadable issue must not merge", status)
+	}
+}
+
+// risk/low is necessary, not sufficient: red CI still blocks it, and a review-protected base still blocks it.
+func TestTBDShip_RiskLow_DoesNotOverrideTheOtherGates(t *testing.T) {
+	var merged, deleted atomic.Bool
+	red := shipFakeFull(t, `[{"id":1,"status":"completed","conclusion":"failure","head_sha":"abc"}]`, 0, http.StatusOK, `[{"id":1,"name":"risk/low"}]`, http.StatusOK, &merged, &deleted)
+	defer red.Close()
+	assert.Equal(t, false, callShip(t, red.URL, shipArgs)["merged"])
+	assert.False(t, merged.Load())
+
+	prot := shipFakeFull(t, shipGreen, 1, http.StatusOK, `[{"id":1,"name":"risk/low"}]`, http.StatusOK, &merged, &deleted)
+	defer prot.Close()
+	assert.Equal(t, false, callShip(t, prot.URL, shipArgs)["merged"])
+	assert.False(t, merged.Load())
+}
+
+// The decision is visible to the caller.
+func TestTBDShip_ReportsTheRiskTier(t *testing.T) {
+	var merged, deleted atomic.Bool
+	srv := shipFakeFull(t, shipGreen, 0, http.StatusOK, `[{"id":1,"name":"risk/medium"}]`, http.StatusOK, &merged, &deleted)
+	defer srv.Close()
+
+	assert.Equal(t, "medium", callShip(t, srv.URL, shipArgs)["risk"])
+}
+
 func TestTBDShip_AllowlistRejects(t *testing.T) {
 	tool := tools.NewTBDShip(gitea.NewClient("http://unused", ""), allowlist.New([]string{"mathias"}))
 	_, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"evil","repo":"r","path":"p","content":"c","message":"m"}`))
@@ -1,10 +1,14 @@
 package tools

 import (
+	"bytes"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"reflect"
+	"regexp"
 	"strings"
+	"unicode"

 	"git.d-ma.be/mathias/gitea-mcp/internal/gitea"
 	"git.d-ma.be/mathias/gitea-mcp/internal/registry"
@@ -22,12 +26,154 @@ func parseArgs(raw json.RawMessage, dst any) error {
 		if err := json.Unmarshal([]byte("{}"), dst); err != nil {
 			return err
 		}
-	} else if err := json.Unmarshal(normalizeAliases(raw), dst); err != nil {
-		return err
+	} else {
+		// Reject ambiguity BEFORE alias normalisation: re-marshalling through a
+		// map would otherwise silently collapse exact duplicates (#65).
+		if err := rejectAmbiguousKeys(raw, dst); err != nil {
+			return err
+		}
+		if err := json.Unmarshal(normalizeAliases(raw), dst); err != nil {
+			return err
+		}
 	}
 	return validateRequiredIdentifiers(dst)
 }

+// rejectAmbiguousKeys refuses input whose meaning depends on encoding/json's
+// case-insensitive, last-key-wins matching (#65). A client-side check that reads
+// `branch` must see the value the server acts on, so: no object at any depth may
+// carry two keys equal under strings.EqualFold (exact duplicates included), and
+// no top-level key may differ from a field of dst only in case.
+func rejectAmbiguousKeys(raw json.RawMessage, dst any) error {
+	dec := json.NewDecoder(bytes.NewReader(raw))
+	// UseNumber: without it Token() fails on a number outside float64 range,
+	// which json.Unmarshal ignores in an unknown key — a decoy that would end
+	// the scan before the duplicate (PR #66 review).
+	dec.UseNumber()
+	if err := scanJSONValue(dec); err != nil {
+		if errors.Is(err, gitea.ErrValidation) {
+			return err
+		}
+		// Fail closed: input the scanner cannot read is input it cannot vouch for.
+		return fmt.Errorf("arguments are not valid JSON: %v: %w", err, gitea.ErrValidation)
+	}
+	var top map[string]json.RawMessage
+	if err := json.Unmarshal(raw, &top); err != nil {
+		return nil // not an object; nothing to compare against the schema
+	}
+	fields := jsonFieldNames(dst)
+	for k := range top {
+		for _, f := range fields {
+			if k != f && strings.EqualFold(k, f) {
+				return fmt.Errorf("argument %q does not match field %q exactly (keys are case-sensitive): %w", k, f, gitea.ErrValidation)
+			}
+		}
+	}
+	return nil
+}
+
+// scanJSONValue consumes one JSON value from dec, failing with ErrValidation if
+// any object within it has two keys that are equal under strings.EqualFold.
+func scanJSONValue(dec *json.Decoder) error {
+	tok, err := dec.Token()
+	if err != nil {
+		return err
+	}
+	d, ok := tok.(json.Delim)
+	if !ok {
+		return nil
+	}
+	switch d {
+	case '{':
+		seen := map[string]string{}
+		for dec.More() {
+			kt, err := dec.Token()
+			if err != nil {
+				return err
+			}
+			k, _ := kt.(string)
+			fk := foldKey(k)
+			if prev, dup := seen[fk]; dup {
+				return fmt.Errorf("duplicate argument keys %q and %q (keys must be unique ignoring case): %w", prev, k, gitea.ErrValidation)
+			}
+			seen[fk] = k
+			if err := scanJSONValue(dec); err != nil {
+				return err
+			}
+		}
+	case '[':
+		for dec.More() {
+			if err := scanJSONValue(dec); err != nil {
+				return err
+			}
+		}
+	}
+	_, err = dec.Token() // closing delimiter
+	return err
+}
+
+// foldKey maps every rune to the smallest member of its unicode.SimpleFold
+// orbit, so foldKey(a) == foldKey(b) exactly when strings.EqualFold(a, b) —
+// the folding encoding/json uses (long s == s, Kelvin sign == k). A map keyed
+// on it keeps the duplicate scan linear (PR #66 review: a pairwise EqualFold
+// loop cost 7s of CPU on ~90k keys).
+func foldKey(s string) string {
+	var b strings.Builder
+	for _, r := range s {
+		min := r
+		for f := unicode.SimpleFold(r); f != r; f = unicode.SimpleFold(f) {
+			if f < min {
+				min = f
+			}
+		}
+		b.WriteRune(min)
+	}
+	return b.String()
+}
+
+// jsonFieldNames returns the json tag names of dst's top-level struct fields.
+func jsonFieldNames(dst any) []string {
+	v := reflect.ValueOf(dst)
+	if v.Kind() != reflect.Pointer || v.IsNil() || v.Elem().Kind() != reflect.Struct {
+		return nil
+	}
+	t := v.Elem().Type()
+	names := make([]string, 0, t.NumField())
+	for i := 0; i < t.NumField(); i++ {
+		name, _, _ := strings.Cut(t.Field(i).Tag.Get("json"), ",")
+		if name == "" || name == "-" {
+			name = t.Field(i).Name
+		}
+		names = append(names, name)
+	}
+	return names
+}
+
+// branchNameRe is the shape the skills mcp-destructive-guard hook enforces.
+var branchNameRe = regexp.MustCompile(`^[A-Za-z0-9][A-Za-z0-9._/-]*$`)
+
+// ValidateBranchName rejects branch names that Gitea would turn into a ref
+// other than refs/heads/<name> as the caller meant it. `refs/heads/main` in
+// particular creates refs/heads/refs/heads/main, after which every API write to
+// main fails with "matches more than one" until the ref is deleted by hand
+// (#65). Beyond the shape regexp it applies git's own check-ref-format rules
+// that the regexp admits.
+func ValidateBranchName(b string) error {
+	bad := !branchNameRe.MatchString(b) ||
+		strings.HasPrefix(b, "refs/") ||
+		b == "HEAD" ||
+		strings.Contains(b, "..") ||
+		strings.Contains(b, "//") ||
+		strings.Contains(b, "/.") ||
+		strings.HasSuffix(b, "/") ||
+		strings.HasSuffix(b, ".") ||
+		strings.HasSuffix(b, ".lock")
+	if bad {
+		return fmt.Errorf("invalid branch name %q (want a short name like feat/x, no refs/ prefix): %w", b, gitea.ErrValidation)
+	}
+	return nil
+}
+
 // validateRequiredIdentifiers rejects an empty repo identifier at the tool layer
 // before it reaches the client, where it would build a trailing-empty path
 // segment (`/api/v1/repos/{owner}/`) and leak gitea's bare 404 (#37). `repo` and
@@ -58,7 +58,7 @@ if [ "$FORCE" = false ]; then
    while IFS= read -r _match; do
      _lineno="${_match%%:*}"
      _content="${_match#*:}"
-      _token=$(printf '%s' "$_content" | grep -o '{{[^}]*}}' | head -1)
+      _token=$(head -1 < <(grep -o '{{[^}]*}}' <<<"$_content"))
      echo "  $PROJECT_FILE:$_lineno: unfilled placeholder $_token" >&2
    done <<< "$_placeholders"
    echo "" >&2
Author	SHA1	Message	Date
mathias	5c6c4f1e1e	fix(pr_merge): send Gitea's merge_title_field (gitea-mcp#77) (#82 ) CD / Deploy via GitOps (push) Successful in 7s Details CD / Lint / Test / Vet (push) Successful in 12s Details CD / Build & Import (push) Successful in 5s Details Co-authored-by: Mathias <mthbqv@gmail.com>	2026-10-10 08:50:06 +00:00
mathias	e8521c2555	chore(context): restore the root rules in AGENTS.md and system-prompt.txt (#80 ) (#81 ) CD / Lint / Test / Vet (push) Successful in 12s Details CD / Deploy via GitOps (push) Successful in 5s Details CD / Build & Import (push) Successful in 17s Details Co-authored-by: Mathias <mthbqv@gmail.com>	2026-10-09 06:02:43 +00:00
mathiasandClaude Sonnet 5.5	e696db2e68	docs: main is protected, ship by PR (local-dev#42) CD / Lint / Test / Vet (push) Successful in 13s Details CD / Build & Import (push) Successful in 5s Details CD / Deploy via GitOps (push) Successful in 5s Details Gitea will refuse direct pushes to main in this repo. Rule 5's reviewed lane was enforced on the merge tools (tbd_ship / pr_merge, gitea-mcp v0.17.0/v0.18.0) but not on a direct commit; this overrides rule 5's "risk/low straight to main" for this repo and says why. Derived adapter files regenerated where the repo has them. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZkhi49hfeSqBp5aZJNXsM	2026-10-08 08:34:50 +02:00
mathias	d1e501cc64	feat(pr_merge): gate it like tbd_ship — green CI, unprotected base, exactly one risk/low issue (#74 ) (#76 ) CD / Lint / Test / Vet (push) Successful in 15s Details CD / Build & Import (push) Successful in 7s Details CD / Deploy via GitOps (push) Skipped Details Co-authored-by: Mathias <mthbqv@gmail.com>	2026-10-08 06:18:07 +00:00
mathias	d3faffc4c4	feat(tbd_ship): gate auto-merge on the risk tier; unclassified is not low (#63 ) (#73 ) CD / Build & Import (push) Successful in 4s Details CD / Lint / Test / Vet (push) Successful in 11s Details CD / Deploy via GitOps (push) Skipped Details Co-authored-by: Mathias <mthbqv@gmail.com>	2026-10-08 06:02:34 +00:00
mathiasandClaude Opus 5.5	dbf78a3b34	ci: move template expressions out of run: shells (signals#47) CD / Lint / Test / Vet (push) Successful in 12s Details CD / Build & Import (push) Successful in 6s Details CD / Deploy via GitOps (push) Successful in 5s Details "Derive image tags" expanded ${{ github.ref }} into the run: body. A ref name may contain $(...) and the expansion happens before the shell parses the script, so it now arrives via env: REF instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-05 09:33:44 +02:00
mathias	c3f8d6c002	ci: run the check job on pull requests (#68 ) (#69 ) CD / Deploy via GitOps (push) Successful in 5s Details CD / Lint / Test / Vet (push) Successful in 13s Details CD / Build & Import (push) Successful in 6s Details Co-authored-by: Mathias <mthbqv@gmail.com>	2026-10-05 07:25:38 +00:00
mathias	dc127a05b5	Merge pull request 'fix(tools): reject case-variant arg keys and ref-shaped branch names (#65 )' (#66 ) from fix/65-strict-args-branch-names into main CD / Lint / Test / Vet (push) Successful in 14s Details CD / Build & Import (push) Successful in 6s Details CD / Deploy via GitOps (push) Skipped Details	2026-10-05 05:10:51 +00:00
mathiasandClaude Opus 5.5	cd61239451	perf(tools): linear duplicate-key scan keyed on SimpleFold orbit PR #66 review round 2: the pairwise EqualFold loop was quadratic, 7s of CPU for ~90k keys under the 1 MiB body cap. foldKey maps each rune to the smallest member of its unicode.SimpleFold orbit, equal exactly when strings.EqualFold is, so the scan is a map lookup again. The scaling test fails on the previous code (7.1s). Refs #65 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-05 07:05:00 +02:00
mathiasandClaude Opus 5.5	31a3dd3e0e	fix(tools): fail closed when the key scan cannot read the input PR #66 review: json.Decoder.Token() fails on a number outside float64 range (1e999), which json.Unmarshal ignores in an unknown key, so a decoy key ended the scan and the #65 repro wrote to main. The decoder now uses UseNumber, and any scan error is a validation error. The duplicate check folds with strings.EqualFold (as encoding/json does) instead of ToLower, so long s and the Kelvin sign are caught in nested objects too. Refs #65 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-05 07:02:27 +02:00
mathiasandClaude Opus 5.5	ae50501ba3	fix(tools): reject case-variant arg keys and ref-shaped branch names parseArgs now refuses, before any request: two keys equal under strings.EqualFold in any object (exact duplicates included), and a top-level key that differs from a schema field only in case. Without this, encoding/json's case-insensitive last-wins match let {"branch":"feat/x","Branch":"main"} write to main while a client-side hook read feat/x. file_write_branch (branch) and tbd_ship (branch and base) validate the branch name: ^[A-Za-z0-9][A-Za-z0-9._/-]*$ plus no refs/ prefix, HEAD, "..", "//", "/.", trailing "/" or ".", or ".lock". refs/heads/main otherwise created refs/heads/refs/heads/main and wedged API writes to main. tbd_ship's base is the protection-probe key, so a refs/-prefixed base would also have read as unprotected. Closes #65 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-05 06:58:31 +02:00
mathiasandClaude Opus 5.5	9396b4daaa	chore(context): resync derived adapters from canonical AGENT.md CD / Lint / Test / Vet (push) Successful in 28s Details CD / Build & Import (push) Successful in 14s Details CD / Deploy via GitOps (push) Successful in 6s Details Clears ContextAdapterDrift; canonical changed in local-dev 12f694b/218cf0d. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-04 22:49:57 +02:00
mathiasandClaude Opus 5.5	0e2bd3d90e	fix(mcp): audit record keeps legacy name/index alias targets CD / Deploy via GitOps (push) Successful in 5s Details CD / Lint / Test / Vet (push) Successful in 15s Details CD / Build & Import (push) Successful in 7s Details A caller using the #38 aliases (name for repo, index for number) left no target in the tools/call audit record. Refs #64 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-01 20:41:51 +02:00
mathiasandClaude Opus 5.5	b149dccdbe	chore(context): re-sync adapters after root brain-scoping correction Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-01 20:41:47 +02:00
mathiasandClaude Opus 5.5	724a997002	feat(mcp): audit-log every tools/call (OWASP MCP08) CD / Deploy via GitOps (push) Successful in 5s Details CD / Lint / Test / Vet (push) Successful in 31s Details CD / Build & Import (push) Successful in 14s Details tools/call dispatched with no record, so repo_delete, pr_merge and file_delete left no trail in gitea-mcp; only auth acceptance was logged. One slog record per call: tool, caller, token fingerprint, target (owner/repo/number/branch/path/tag), outcome, duration. Every other argument is reduced to args_bytes so the audit line never copies file content or issue bodies. Closes #64 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-01 20:34:59 +02:00
mathiasandClaude Opus 5	59dabc9455	chore(context): adopt canonical context-sync.sh and re-sync adapters CD / Lint / Test / Vet (push) Successful in 13s Details CD / Build & Import (push) Successful in 17s Details CD / Deploy via GitOps (push) Successful in 5s Details infra's scripts/context-sync.sh is the estate canonical, and it moved in infra 262369b (a SIGPIPE fix on the placeholder-reporting path: `printf \| grep -o \| head -1` became `head -1 < <(grep -o ...)`). Every other repo still carried the old line, so ContextSyncScriptDrift and ContextAdapterDrift were both firing. Also picks up the canonical AGENT.md change from local-dev a01ee80 — rule 5's GitOps review lane and the new precedence rule — which every repo's adapters need regenerated for regardless of the script. Refs mathias/infra#470 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-30 10:36:32 +02:00
mathiasandClaude Opus 5	6b57a8ba8a	chore(context): regenerate derived adapters from the canonical root context CD / Lint / Test / Vet (push) Successful in 41s Details CD / Build & Import (push) Successful in 22s Details CD / Deploy via GitOps (push) Successful in 10s Details ContextAdapterDrift was firing for 13 repos. Picks up behaviour rule 7 ("Finish what's closable"), which landed in ~/dev/.context/AGENT.md and never reached any repo's derived files. Where mcp.json changes, it also picks up root MCP config the repo was stale on: brain/gitea/infra move from public hostnames to tailnet Serve URLs, and gitea's bearer changes from GITEA_MCP_TOKEN to DMABE_GITEA_API_TOKEN — gitea-mcp authenticates by pass-through and 401s on the former, so a harness syncing from the old file had a dead gitea server and no signal saying so. No literal secrets: mcp.json carries ${VAR} placeholders only. CLAUDE.md deliberately unchanged — context-sync writes it from PROJECT.md alone, since Claude Code walks up to ~/dev/CLAUDE.md for root context. Refs mathias/infra ContextAdapterDrift Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-25 07:34:24 +02:00
mathiasandClaude Opus 5	e1e9a2e30e	feat(auth): every accepted request names the path that granted it CD / Deploy via GitOps (push) Successful in 5s Details CD / Lint / Test / Vet (push) Successful in 16s Details CD / Build & Import (push) Successful in 16s Details Closes #60. Closes #58 (fixed in mcp-chassis v0.7.0, consumed here). Three auth paths reach this server and none of them said so on success. PassthroughMiddleware logged nothing at all, success or failure. While verifying #59 end to end there was no way to confirm from the logs whether calls were using the new PAT pass-through — the caller's own Gitea PAT, the caller's own identity — or falling through to the shared JWT/static path. The only available reasoning was inference from which failures were absent. INFO mcp auth accepted auth_path=passthrough caller=alice token_fp=... INFO mcp auth accepted auth_path=jwt caller=mathiasbq token_fp=... INFO mcp auth accepted auth_path=static token_fp=... The first line matters more than the other two. Pass-through is the path where a request acts as a SPECIFIC Gitea user; an audit trail that cannot say which identity performed a write is not an audit trail. chassisauth.Fingerprint, not a local equivalent. The chassis emits token_fp on its own accepted and rejected lines, and two different fingerprints for the same token cannot be correlated — which is the only reason the field exists. Exported from the chassis in v0.7.1 for exactly this. A token that is not a valid PAT still falls through SILENTLY. The chassis owns the verdict, and logging here as well would double-count every static-token request, since all of them reach this middleware first. Asserted, because "logs nothing on the failure path" is indistinguishable from "was never called". Also lands #58 by consuming mcp-chassis v0.7.0: a JWT presented to a server with no JWT auth configured now reports jwt_auth_not_configured rather than static_token_mismatch. That log line is what sent a debugging session into re-running the OAuth flow while the credentials were valid the whole time. Verified: task check EXIT=0. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-14 20:59:04 +02:00