fix(workflow_run_trigger): 204 dispatch is success; resolve run via listing (#41)

Gitea's workflow_dispatch endpoint returns 204 No Content with no Location
header. DispatchWorkflow required that header, so every successful dispatch
errored with "missing Location header" and never yielded a run ID — making
the tool unusable for CAD dispatch.

- DispatchWorkflow now returns error-only; 204 = success, no Location needed.
  Body already carried ref+inputs; kept and covered by test.
- Tool snapshots the newest existing workflow_dispatch run before dispatch,
  then polls ListWorkflowRuns after and returns the newest run with ID above
  that baseline (avoids returning a stale prior run). Falls back to an honest
  "dispatched, run not yet registered" result rather than failing.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.aider.conventions.md

@@ -27,6 +27,14 @@ and climate/sustainability tech.
 
 These rules apply to every task across every project, regardless of harness.
 
+0. **Pre-task ritual — before ANY implementation (non-negotiable).** Run this before writing a single line:
+   - **Query the brain** (`brain_query`) for the domain + symptom. If the result changes your approach, surface it before acting. 5 seconds beats 5 hours.
+   - **Load the relevant skill** — see trigger table in *Engineering Skills* below.
+   - **Write the failing test first.** Name the test before the function. If the target is untestable (e.g. `main()` wiring), extract the logic into a testable function first. No implementation without a red test.
+   - **State the observable success criterion** — what specific behavior, output, or passing test proves this is done?
+
+   **TDD is non-negotiable.** "Tests pass" is not proof of correctness — only proof the tests ran. Write tests that would catch the bug before writing code that fixes it.
+
 1. **No assumptions.** Don't hide confusion — surface it. Surface tradeoffs explicitly.
    Think before coding; if the problem is unclear, ask or state assumptions before acting.
 2. **Minimum viable code.** Solve with the smallest change that works. Nothing
@@ -49,6 +57,22 @@ These rules apply to every task across every project, regardless of harness.
    PR flow only when a human reviewer outside the project is required. Document
    the reason in PROJECT.md.
 
+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
+     new/changed ADR, patch for a fix; docs in the same commit). Check the repo's
+     actual last tag — stated versions in docs drift stale.
+   - **Push** main and the tag (CI is the gate).
+   - **Persist generalizable learnings to the brain** (`brain_write`, wing/hall) —
+     the reusable patterns and the footguns that would bite anyone again, never
+     project status. See *Knowledge base — when to write* below.
+   - **File discovered-but-deferred work as tracker issues** on the project's own
+     repo — token-budget gaps, recorded ADR limitations, v2 follow-ups. Don't let
+     "out of scope, recorded" rot in a commit message; make it a ticket with a
+     source pointer.
+   - Surface the brain entries and issue numbers in the closing summary so the
+     trail is auditable.
+
 ## Default stack
 
 | Layer | Default | Fallback | Last resort |
@@ -78,6 +102,26 @@ Exploratory: Rust, Zig — I'll tell you when I want these.
 - **Security**: no secrets in code, govulncheck before adding deps, SOPS for encrypted config
 - **Dependencies**: prefer stdlib. testify, slog, templ, sqlc, google.golang.org/adk (agent projects only) are pre-approved; anything else needs justification in the commit message
 
+## Secret handling (every harness, every command)
+
+Tool output is persisted: terminal → `~/.claude/projects` transcripts →
+claudewatcher → brain/wiki → gitea history. A secret printed once is
+searchable forever, and clearing it means rotating the key. So:
+
+1. **Never print, echo, log, or transform a secret to inspect it.** No
+   `base64`/`xxd`/`cat` of a key, and never pipe a secret through a transform
+   to defeat `op run`'s output masking (it masks raw values; base64 hides them
+   from the mask — that exact trick leaked a key on 2026-06-11).
+2. **Secrets stay in the subprocess.** Reference them only as env vars consumed
+   *inside* `op run --env-file ~/.op-env -- <cmd>`. Never place a literal secret
+   in a command's argv (it lands in the tool call and the transcript).
+3. **Existence check without revealing the value:** `[ -n "$X" ] && echo set` —
+   never `${X:-...}` (returns the value when set) and never echo a substring of it.
+4. **Cross-host secrets:** run the secret-consuming command on the host that has
+   the secret; do not forward a raw key over ssh argv/stdout.
+5. If a secret does leak into output, say so immediately and flag it for rotation —
+   don't bury it.
+
 ## Infrastructure
 
 Three machines on Tailscale:
@@ -157,7 +201,7 @@ 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":"..."}` |
-| **Browser / human inspection** | `https://gitea.d-ma.be/mathias/hyperguild` → `knowledge/` and `wiki/` markdown files |
+| **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`.
 - **Routing**: brain_answer's LLM uses berget.ai as primary, iguana ollama as
@@ -219,15 +263,17 @@ unconditionally on every host, every harness.
 
 ## Engineering Skills
 
-Shared engineering skills are available in `~/dev/.skills/`. Load on demand via the index.
+Shared engineering skills are available in `~/dev/.skills/`. Load at task start — not "on demand" but on schedule, before writing code. See `~/dev/.skills/SKILLS_INDEX.md` for the full list.
 
-See `~/dev/.skills/SKILLS_INDEX.md` for the full list with descriptions and "use when" triggers.
+**Skill trigger table — load before starting, not after getting stuck:**
 
-Key skills:
+| Task type | Load |
-- **TDD**: always write tests first — load `tdd` skill
+|-----------|------|
-- **Code Review**: load `code-review` skill before any review
+| Any feature or bug fix | `tdd` |
-- **SOLID/Clean Code**: load `solid` or `clean-code` skill for design work
+| Refactor or design | `clean-code` or `solid` |
-- **Problem first**: load `problem-analysis` skill before coding non-trivial features
+| Debug | `problem-analysis` |
+| Review code or PRs | `code-review` |
+| Frame a problem before coding | `problem-analysis` |
 
 ---
 

.context/mcp.json

@@ -16,10 +16,7 @@
     },
     "infra": {
       "type": "http",
-      "url": "https://infra-mcp.d-ma.be/mcp",
+      "url": "https://infra-mcp.d-ma.be/mcp"
-      "headers": {
-        "Authorization": "Bearer ${INFRA_MCP_TOKEN}"
-      }
     }
   }
 }

.context/system-prompt.txt

@@ -32,6 +32,14 @@ and climate/sustainability tech.
 
 These rules apply to every task across every project, regardless of harness.
 
+0. **Pre-task ritual — before ANY implementation (non-negotiable).** Run this before writing a single line:
+   - **Query the brain** (`brain_query`) for the domain + symptom. If the result changes your approach, surface it before acting. 5 seconds beats 5 hours.
+   - **Load the relevant skill** — see trigger table in *Engineering Skills* below.
+   - **Write the failing test first.** Name the test before the function. If the target is untestable (e.g. `main()` wiring), extract the logic into a testable function first. No implementation without a red test.
+   - **State the observable success criterion** — what specific behavior, output, or passing test proves this is done?
+
+   **TDD is non-negotiable.** "Tests pass" is not proof of correctness — only proof the tests ran. Write tests that would catch the bug before writing code that fixes it.
+
 1. **No assumptions.** Don't hide confusion — surface it. Surface tradeoffs explicitly.
    Think before coding; if the problem is unclear, ask or state assumptions before acting.
 2. **Minimum viable code.** Solve with the smallest change that works. Nothing
@@ -54,6 +62,22 @@ These rules apply to every task across every project, regardless of harness.
    PR flow only when a human reviewer outside the project is required. Document
    the reason in PROJECT.md.
 
+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
+     new/changed ADR, patch for a fix; docs in the same commit). Check the repo's
+     actual last tag — stated versions in docs drift stale.
+   - **Push** main and the tag (CI is the gate).
+   - **Persist generalizable learnings to the brain** (`brain_write`, wing/hall) —
+     the reusable patterns and the footguns that would bite anyone again, never
+     project status. See *Knowledge base — when to write* below.
+   - **File discovered-but-deferred work as tracker issues** on the project's own
+     repo — token-budget gaps, recorded ADR limitations, v2 follow-ups. Don't let
+     "out of scope, recorded" rot in a commit message; make it a ticket with a
+     source pointer.
+   - Surface the brain entries and issue numbers in the closing summary so the
+     trail is auditable.
+
 ## Default stack
 
 | Layer | Default | Fallback | Last resort |
@@ -83,6 +107,26 @@ Exploratory: Rust, Zig — I'll tell you when I want these.
 - **Security**: no secrets in code, govulncheck before adding deps, SOPS for encrypted config
 - **Dependencies**: prefer stdlib. testify, slog, templ, sqlc, google.golang.org/adk (agent projects only) are pre-approved; anything else needs justification in the commit message
 
+## Secret handling (every harness, every command)
+
+Tool output is persisted: terminal → `~/.claude/projects` transcripts →
+claudewatcher → brain/wiki → gitea history. A secret printed once is
+searchable forever, and clearing it means rotating the key. So:
+
+1. **Never print, echo, log, or transform a secret to inspect it.** No
+   `base64`/`xxd`/`cat` of a key, and never pipe a secret through a transform
+   to defeat `op run`'s output masking (it masks raw values; base64 hides them
+   from the mask — that exact trick leaked a key on 2026-06-11).
+2. **Secrets stay in the subprocess.** Reference them only as env vars consumed
+   *inside* `op run --env-file ~/.op-env -- <cmd>`. Never place a literal secret
+   in a command's argv (it lands in the tool call and the transcript).
+3. **Existence check without revealing the value:** `[ -n "$X" ] && echo set` —
+   never `${X:-...}` (returns the value when set) and never echo a substring of it.
+4. **Cross-host secrets:** run the secret-consuming command on the host that has
+   the secret; do not forward a raw key over ssh argv/stdout.
+5. If a secret does leak into output, say so immediately and flag it for rotation —
+   don't bury it.
+
 ## Infrastructure
 
 Three machines on Tailscale:
@@ -162,7 +206,7 @@ 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":"..."}` |
-| **Browser / human inspection** | `https://gitea.d-ma.be/mathias/hyperguild` → `knowledge/` and `wiki/` markdown files |
+| **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`.
 - **Routing**: brain_answer's LLM uses berget.ai as primary, iguana ollama as
@@ -224,15 +268,17 @@ unconditionally on every host, every harness.
 
 ## Engineering Skills
 
-Shared engineering skills are available in `~/dev/.skills/`. Load on demand via the index.
+Shared engineering skills are available in `~/dev/.skills/`. Load at task start — not "on demand" but on schedule, before writing code. See `~/dev/.skills/SKILLS_INDEX.md` for the full list.
 
-See `~/dev/.skills/SKILLS_INDEX.md` for the full list with descriptions and "use when" triggers.
+**Skill trigger table — load before starting, not after getting stuck:**
 
-Key skills:
+| Task type | Load |
-- **TDD**: always write tests first — load `tdd` skill
+|-----------|------|
-- **Code Review**: load `code-review` skill before any review
+| Any feature or bug fix | `tdd` |
-- **SOLID/Clean Code**: load `solid` or `clean-code` skill for design work
+| Refactor or design | `clean-code` or `solid` |
-- **Problem first**: load `problem-analysis` skill before coding non-trivial features
+| Debug | `problem-analysis` |
+| Review code or PRs | `code-review` |
+| Frame a problem before coding | `problem-analysis` |
 
 ---
 

.cursorrules

@@ -30,6 +30,14 @@ and climate/sustainability tech.
 
 These rules apply to every task across every project, regardless of harness.
 
+0. **Pre-task ritual — before ANY implementation (non-negotiable).** Run this before writing a single line:
+   - **Query the brain** (`brain_query`) for the domain + symptom. If the result changes your approach, surface it before acting. 5 seconds beats 5 hours.
+   - **Load the relevant skill** — see trigger table in *Engineering Skills* below.
+   - **Write the failing test first.** Name the test before the function. If the target is untestable (e.g. `main()` wiring), extract the logic into a testable function first. No implementation without a red test.
+   - **State the observable success criterion** — what specific behavior, output, or passing test proves this is done?
+
+   **TDD is non-negotiable.** "Tests pass" is not proof of correctness — only proof the tests ran. Write tests that would catch the bug before writing code that fixes it.
+
 1. **No assumptions.** Don't hide confusion — surface it. Surface tradeoffs explicitly.
    Think before coding; if the problem is unclear, ask or state assumptions before acting.
 2. **Minimum viable code.** Solve with the smallest change that works. Nothing
@@ -52,6 +60,22 @@ These rules apply to every task across every project, regardless of harness.
    PR flow only when a human reviewer outside the project is required. Document
    the reason in PROJECT.md.
 
+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
+     new/changed ADR, patch for a fix; docs in the same commit). Check the repo's
+     actual last tag — stated versions in docs drift stale.
+   - **Push** main and the tag (CI is the gate).
+   - **Persist generalizable learnings to the brain** (`brain_write`, wing/hall) —
+     the reusable patterns and the footguns that would bite anyone again, never
+     project status. See *Knowledge base — when to write* below.
+   - **File discovered-but-deferred work as tracker issues** on the project's own
+     repo — token-budget gaps, recorded ADR limitations, v2 follow-ups. Don't let
+     "out of scope, recorded" rot in a commit message; make it a ticket with a
+     source pointer.
+   - Surface the brain entries and issue numbers in the closing summary so the
+     trail is auditable.
+
 ## Default stack
 
 | Layer | Default | Fallback | Last resort |
@@ -81,6 +105,26 @@ Exploratory: Rust, Zig — I'll tell you when I want these.
 - **Security**: no secrets in code, govulncheck before adding deps, SOPS for encrypted config
 - **Dependencies**: prefer stdlib. testify, slog, templ, sqlc, google.golang.org/adk (agent projects only) are pre-approved; anything else needs justification in the commit message
 
+## Secret handling (every harness, every command)
+
+Tool output is persisted: terminal → `~/.claude/projects` transcripts →
+claudewatcher → brain/wiki → gitea history. A secret printed once is
+searchable forever, and clearing it means rotating the key. So:
+
+1. **Never print, echo, log, or transform a secret to inspect it.** No
+   `base64`/`xxd`/`cat` of a key, and never pipe a secret through a transform
+   to defeat `op run`'s output masking (it masks raw values; base64 hides them
+   from the mask — that exact trick leaked a key on 2026-06-11).
+2. **Secrets stay in the subprocess.** Reference them only as env vars consumed
+   *inside* `op run --env-file ~/.op-env -- <cmd>`. Never place a literal secret
+   in a command's argv (it lands in the tool call and the transcript).
+3. **Existence check without revealing the value:** `[ -n "$X" ] && echo set` —
+   never `${X:-...}` (returns the value when set) and never echo a substring of it.
+4. **Cross-host secrets:** run the secret-consuming command on the host that has
+   the secret; do not forward a raw key over ssh argv/stdout.
+5. If a secret does leak into output, say so immediately and flag it for rotation —
+   don't bury it.
+
 ## Infrastructure
 
 Three machines on Tailscale:
@@ -160,7 +204,7 @@ 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":"..."}` |
-| **Browser / human inspection** | `https://gitea.d-ma.be/mathias/hyperguild` → `knowledge/` and `wiki/` markdown files |
+| **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`.
 - **Routing**: brain_answer's LLM uses berget.ai as primary, iguana ollama as
@@ -222,15 +266,17 @@ unconditionally on every host, every harness.
 
 ## Engineering Skills
 
-Shared engineering skills are available in `~/dev/.skills/`. Load on demand via the index.
+Shared engineering skills are available in `~/dev/.skills/`. Load at task start — not "on demand" but on schedule, before writing code. See `~/dev/.skills/SKILLS_INDEX.md` for the full list.
 
-See `~/dev/.skills/SKILLS_INDEX.md` for the full list with descriptions and "use when" triggers.
+**Skill trigger table — load before starting, not after getting stuck:**
 
-Key skills:
+| Task type | Load |
-- **TDD**: always write tests first — load `tdd` skill
+|-----------|------|
-- **Code Review**: load `code-review` skill before any review
+| Any feature or bug fix | `tdd` |
-- **SOLID/Clean Code**: load `solid` or `clean-code` skill for design work
+| Refactor or design | `clean-code` or `solid` |
-- **Problem first**: load `problem-analysis` skill before coding non-trivial features
+| Debug | `problem-analysis` |
+| Review code or PRs | `code-review` |
+| Frame a problem before coding | `problem-analysis` |
 
 ---
 

AGENTS.md

@@ -27,6 +27,14 @@ and climate/sustainability tech.
 
 These rules apply to every task across every project, regardless of harness.
 
+0. **Pre-task ritual — before ANY implementation (non-negotiable).** Run this before writing a single line:
+   - **Query the brain** (`brain_query`) for the domain + symptom. If the result changes your approach, surface it before acting. 5 seconds beats 5 hours.
+   - **Load the relevant skill** — see trigger table in *Engineering Skills* below.
+   - **Write the failing test first.** Name the test before the function. If the target is untestable (e.g. `main()` wiring), extract the logic into a testable function first. No implementation without a red test.
+   - **State the observable success criterion** — what specific behavior, output, or passing test proves this is done?
+
+   **TDD is non-negotiable.** "Tests pass" is not proof of correctness — only proof the tests ran. Write tests that would catch the bug before writing code that fixes it.
+
 1. **No assumptions.** Don't hide confusion — surface it. Surface tradeoffs explicitly.
    Think before coding; if the problem is unclear, ask or state assumptions before acting.
 2. **Minimum viable code.** Solve with the smallest change that works. Nothing
@@ -49,6 +57,22 @@ These rules apply to every task across every project, regardless of harness.
    PR flow only when a human reviewer outside the project is required. Document
    the reason in PROJECT.md.
 
+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
+     new/changed ADR, patch for a fix; docs in the same commit). Check the repo's
+     actual last tag — stated versions in docs drift stale.
+   - **Push** main and the tag (CI is the gate).
+   - **Persist generalizable learnings to the brain** (`brain_write`, wing/hall) —
+     the reusable patterns and the footguns that would bite anyone again, never
+     project status. See *Knowledge base — when to write* below.
+   - **File discovered-but-deferred work as tracker issues** on the project's own
+     repo — token-budget gaps, recorded ADR limitations, v2 follow-ups. Don't let
+     "out of scope, recorded" rot in a commit message; make it a ticket with a
+     source pointer.
+   - Surface the brain entries and issue numbers in the closing summary so the
+     trail is auditable.
+
 ## Default stack
 
 | Layer | Default | Fallback | Last resort |
@@ -78,6 +102,26 @@ Exploratory: Rust, Zig — I'll tell you when I want these.
 - **Security**: no secrets in code, govulncheck before adding deps, SOPS for encrypted config
 - **Dependencies**: prefer stdlib. testify, slog, templ, sqlc, google.golang.org/adk (agent projects only) are pre-approved; anything else needs justification in the commit message
 
+## Secret handling (every harness, every command)
+
+Tool output is persisted: terminal → `~/.claude/projects` transcripts →
+claudewatcher → brain/wiki → gitea history. A secret printed once is
+searchable forever, and clearing it means rotating the key. So:
+
+1. **Never print, echo, log, or transform a secret to inspect it.** No
+   `base64`/`xxd`/`cat` of a key, and never pipe a secret through a transform
+   to defeat `op run`'s output masking (it masks raw values; base64 hides them
+   from the mask — that exact trick leaked a key on 2026-06-11).
+2. **Secrets stay in the subprocess.** Reference them only as env vars consumed
+   *inside* `op run --env-file ~/.op-env -- <cmd>`. Never place a literal secret
+   in a command's argv (it lands in the tool call and the transcript).
+3. **Existence check without revealing the value:** `[ -n "$X" ] && echo set` —
+   never `${X:-...}` (returns the value when set) and never echo a substring of it.
+4. **Cross-host secrets:** run the secret-consuming command on the host that has
+   the secret; do not forward a raw key over ssh argv/stdout.
+5. If a secret does leak into output, say so immediately and flag it for rotation —
+   don't bury it.
+
 ## Infrastructure
 
 Three machines on Tailscale:
@@ -157,7 +201,7 @@ 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":"..."}` |
-| **Browser / human inspection** | `https://gitea.d-ma.be/mathias/hyperguild` → `knowledge/` and `wiki/` markdown files |
+| **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`.
 - **Routing**: brain_answer's LLM uses berget.ai as primary, iguana ollama as
@@ -219,15 +263,17 @@ unconditionally on every host, every harness.
 
 ## Engineering Skills
 
-Shared engineering skills are available in `~/dev/.skills/`. Load on demand via the index.
+Shared engineering skills are available in `~/dev/.skills/`. Load at task start — not "on demand" but on schedule, before writing code. See `~/dev/.skills/SKILLS_INDEX.md` for the full list.
 
-See `~/dev/.skills/SKILLS_INDEX.md` for the full list with descriptions and "use when" triggers.
+**Skill trigger table — load before starting, not after getting stuck:**
 
-Key skills:
+| Task type | Load |
-- **TDD**: always write tests first — load `tdd` skill
+|-----------|------|
-- **Code Review**: load `code-review` skill before any review
+| Any feature or bug fix | `tdd` |
-- **SOLID/Clean Code**: load `solid` or `clean-code` skill for design work
+| Refactor or design | `clean-code` or `solid` |
-- **Problem first**: load `problem-analysis` skill before coding non-trivial features
+| Debug | `problem-analysis` |
+| Review code or PRs | `code-review` |
+| Frame a problem before coding | `problem-analysis` |
 
 ---
 

cmd/gitea-mcp/main.go

@@ -7,7 +7,7 @@ import (
 	"os"
 	"strings"
 
-	chassisauth "gitea.d-ma.be/mathias/mcp-chassis/auth"
+	chassisauth "git.d-ma.be/mathias/mcp-chassis/auth"
 
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/allowlist"
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/auth"

go.mod

@@ -3,13 +3,12 @@ module gitea.d-ma.be/mathias/gitea-mcp
 go 1.26.2
 
 require (
+	git.d-ma.be/mathias/mcp-chassis v0.2.0
 	github.com/hashicorp/golang-lru/v2 v2.0.7
-	github.com/lestrrat-go/jwx/v2 v2.1.6
 	github.com/stretchr/testify v1.11.1
 )
 
 require (
-	gitea.d-ma.be/mathias/mcp-chassis v0.1.0 // indirect
 	github.com/davecgh/go-spew v1.1.1 // indirect
 	github.com/decred/dcrd/dcrec/secp256k1/v4 v4.4.0 // indirect
 	github.com/goccy/go-json v0.10.3 // indirect
@@ -17,6 +16,7 @@ require (
 	github.com/lestrrat-go/httpcc v1.0.1 // indirect
 	github.com/lestrrat-go/httprc v1.0.6 // indirect
 	github.com/lestrrat-go/iter v1.0.2 // indirect
+	github.com/lestrrat-go/jwx/v2 v2.1.6 // indirect
 	github.com/lestrrat-go/option v1.0.1 // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
 	github.com/segmentio/asm v1.2.0 // indirect

go.sum

@@ -1,5 +1,5 @@
-gitea.d-ma.be/mathias/mcp-chassis v0.1.0 h1:8RXO34+n7Vu8HnUMagars6fc4oemqRpMu7MVtjaj4qY=
+git.d-ma.be/mathias/mcp-chassis v0.2.0 h1:6fLmb7xqRa2nNVWsHaUbbfbArgDXJw/gDhb09clBIjo=
-gitea.d-ma.be/mathias/mcp-chassis v0.1.0/go.mod h1:ajbLlwr2L7FAN3TBU39KucZkKJM02wTbKbDKDEW2YvE=
+git.d-ma.be/mathias/mcp-chassis v0.2.0/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=

internal/gitea/workflows.go

@@ -6,7 +6,6 @@ import (
 	"fmt"
 	"net/url"
 	"strconv"
-	"strings"
 )
 
 // DispatchWorkflowArgs is the request body for a workflow_dispatch trigger.
@@ -15,42 +14,26 @@ type DispatchWorkflowArgs struct {
 	Inputs map[string]any `json:"inputs,omitempty"`
 }
 
-// WorkflowRunTrigger holds the run ID extracted from the Location header.
+// DispatchWorkflow triggers a workflow_dispatch event. Gitea returns 204 No
-type WorkflowRunTrigger struct {
+// Content with NO Location header, so the response carries no run ID — callers
-	RunID int64
+// resolve the new run separately via ListWorkflowRuns. Returns nil on success.
-}
+func (c *Client) DispatchWorkflow(ctx context.Context, owner, repo, workflow string, args DispatchWorkflowArgs) error {
-
-// DispatchWorkflow triggers a workflow_dispatch event and returns the new run ID.
-func (c *Client) DispatchWorkflow(ctx context.Context, owner, repo, workflow string, args DispatchWorkflowArgs) (*WorkflowRunTrigger, error) {
 	p := fmt.Sprintf("/api/v1/repos/%s/%s/actions/workflows/%s/dispatches", owner, repo, workflow)
 	payload, err := json.Marshal(args)
 	if err != nil {
-		return nil, err
+		return err
 	}
 	resp, err := c.doRaw(ctx, "POST", p, payload)
 	if err != nil {
-		return nil, err
+		return err
 	}
 	if resp.Status != 204 {
 		if mapErr := MapStatus(resp.Status, resp.Body); mapErr != nil {
-			return nil, mapErr
+			return mapErr
 		}
-		return nil, fmt.Errorf("unexpected status %d", resp.Status)
+		return fmt.Errorf("unexpected status %d", resp.Status)
 	}
-	location := resp.Headers.Get("Location")
+	return nil
-	if location == "" {
-		return nil, fmt.Errorf("missing Location header in dispatch response")
-	}
-	// Location is e.g. "/api/v1/repos/o/r/actions/runs/123" — take the last segment.
-	parts := strings.Split(strings.TrimRight(location, "/"), "/")
-	if len(parts) == 0 {
-		return nil, fmt.Errorf("malformed Location: %s", location)
-	}
-	runID, err := strconv.ParseInt(parts[len(parts)-1], 10, 64)
-	if err != nil {
-		return nil, fmt.Errorf("parse run id from %q: %w", location, err)
-	}
-	return &WorkflowRunTrigger{RunID: runID}, nil
 }
 
 // WorkflowRun represents a Gitea Actions run.

internal/gitea/workflows_test.go

@@ -14,6 +14,9 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+// Gitea's dispatch endpoint returns 204 No Content with NO Location header.
+// Dispatch must succeed and forward ref+inputs in the body; the run ID is
+// resolved separately by the tool via ListWorkflowRuns.
 func TestDispatchWorkflow(t *testing.T) {
 	var gotBody []byte
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
@@ -22,18 +25,17 @@ func TestDispatchWorkflow(t *testing.T) {
 		var err error
 		gotBody, err = io.ReadAll(r.Body)
 		assert.NoError(t, err)
-		w.Header().Set("Location", "/api/v1/repos/o/r/actions/runs/789")
+		// No Location header — matches real Gitea.
 		w.WriteHeader(http.StatusNoContent)
 	}))
 	defer srv.Close()
 
 	c := gitea.NewClient(srv.URL, "tok")
-	result, err := c.DispatchWorkflow(context.Background(), "o", "r", "ci.yml", gitea.DispatchWorkflowArgs{
+	err := c.DispatchWorkflow(context.Background(), "o", "r", "ci.yml", gitea.DispatchWorkflowArgs{
 		Ref:    "main",
 		Inputs: map[string]any{"env": "prod"},
 	})
 	require.NoError(t, err)
-	assert.Equal(t, int64(789), result.RunID)
 
 	var body map[string]any
 	require.NoError(t, json.Unmarshal(gotBody, &body))
@@ -43,19 +45,6 @@ func TestDispatchWorkflow(t *testing.T) {
 	assert.Equal(t, "prod", inputs["env"])
 }
 
-func TestDispatchWorkflowMissingLocation(t *testing.T) {
-	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		// 204 but no Location header
-		w.WriteHeader(http.StatusNoContent)
-	}))
-	defer srv.Close()
-
-	c := gitea.NewClient(srv.URL, "tok")
-	_, err := c.DispatchWorkflow(context.Background(), "o", "r", "ci.yml", gitea.DispatchWorkflowArgs{Ref: "main"})
-	require.Error(t, err)
-	assert.Contains(t, err.Error(), "Location")
-}
-
 func TestDispatchWorkflowError404(t *testing.T) {
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		w.WriteHeader(http.StatusNotFound)
@@ -63,7 +52,7 @@ func TestDispatchWorkflowError404(t *testing.T) {
 	defer srv.Close()
 
 	c := gitea.NewClient(srv.URL, "tok")
-	_, err := c.DispatchWorkflow(context.Background(), "o", "r", "ci.yml", gitea.DispatchWorkflowArgs{Ref: "main"})
+	err := c.DispatchWorkflow(context.Background(), "o", "r", "ci.yml", gitea.DispatchWorkflowArgs{Ref: "main"})
 	require.Error(t, err)
 	assert.True(t, errors.Is(err, gitea.ErrNotFound))
 }

internal/tools/create_project_from_template.go

@@ -8,6 +8,7 @@ import (
 	"fmt"
 	"regexp"
 	"strings"
+	"time"
 
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/allowlist"
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/gitea"
@@ -47,7 +48,7 @@ func NewCreateProjectFromTemplate(c *gitea.Client, a *allowlist.Allowlist, tmplO
 func (t *CreateProjectFromTemplate) Descriptor() registry.ToolDescriptor {
 	return registry.ToolDescriptor{
 		Name:        "create_project_from_template",
-		Description: "Create a new project repo from a template, substituting placeholders (__PROJECT_NAME__, __MODULE_PATH__) in every file's content AND path (e.g. renaming cmd/__PROJECT_NAME__/) so the result builds. Defaults to the server-configured template; pass template_name to override (e.g. template-go-agent).",
+		Description: "Create a new project repo from a template. Best-effort substitution of placeholders (__PROJECT_NAME__, __MODULE_PATH__) in every file's content AND path (e.g. renaming cmd/__PROJECT_NAME__/): it completes only if the generated branch is promptly writable. If gitea's async generate is slow (infra#179) the repo is still created and partial_failure explains how to finalize locally (`hyperguild new-project`). Check files_substituted and partial_failure. Defaults to the server-configured template; pass template_name to override (e.g. template-go-agent). Pass dispatch_allow=true to also inject a .dispatch-allow file so the project is immediately dispatch-eligible (dispatch#3).",
 		InputSchema: json.RawMessage(`{
 			"type":"object",
 			"properties":{
@@ -55,7 +56,8 @@ func (t *CreateProjectFromTemplate) Descriptor() registry.ToolDescriptor {
 				"name":{"type":"string","pattern":"^[a-z][a-z0-9-]{1,38}[a-z0-9]$"},
 				"description":{"type":"string"},
 				"private":{"type":"boolean"},
-				"template_name":{"type":"string","description":"Template repo name to generate from. Defaults to the server-configured template."}
+				"template_name":{"type":"string","description":"Template repo name to generate from. Defaults to the server-configured template."},
+				"dispatch_allow":{"type":"boolean","description":"When true, inject a .dispatch-allow file so the new project is immediately opt-in for headless dispatch (dispatch#3). Default false."}
 			},
 			"required":["owner","name"]
 		}`),
@@ -63,13 +65,20 @@ func (t *CreateProjectFromTemplate) Descriptor() registry.ToolDescriptor {
 }
 
 type createProjectArgs struct {
-	Owner        string `json:"owner"`
+	Owner         string `json:"owner"`
-	Name         string `json:"name"`
+	Name          string `json:"name"`
-	Description  string `json:"description"`
+	Description   string `json:"description"`
-	Private      bool   `json:"private"`
+	Private       bool   `json:"private"`
-	TemplateName string `json:"template_name"`
+	TemplateName  string `json:"template_name"`
+	DispatchAllow bool   `json:"dispatch_allow"`
 }
 
+// dispatchAllowContent is the body injected when dispatch_allow=true. Mirrors the
+// sandbox convention: presence of the file (not its content) marks the repo
+// dispatch-eligible; the comment exists only to explain that to a human reader.
+const dispatchAllowContent = "# Presence of this file marks this repo as opt-in for headless dispatch.\n" +
+	"# See dispatch#3.\n"
+
 type createProjectResult struct {
 	FullName         string   `json:"full_name"`
 	HTMLURL          string   `json:"html_url"`
@@ -175,6 +184,39 @@ func (t *CreateProjectFromTemplate) Call(ctx context.Context, raw json.RawMessag
 		}
 	}
 
+	// Opt the new project into headless dispatch if asked: presence of a
+	// .dispatch-allow file on the default branch marks it dispatch-eligible
+	// (dispatch#3). Ride the same upsertRetry path as substitution so it inherits
+	// the infra#179 branch-readiness / partial-failure handling below. Skip if the
+	// loop already stalled — a failed injection then degrades identically.
+	if args.DispatchAllow && result.PartialFailure == "" {
+		const dispatchAllowPath = ".dispatch-allow"
+		if err := t.upsertRetry(ctx, args.Owner, args.Name, dispatchAllowPath, gitea.UpsertFileArgs{
+			Branch:  branch,
+			Content: base64.StdEncoding.EncodeToString([]byte(dispatchAllowContent)),
+			Message: "dispatch: mark project dispatch-eligible (dispatch#3)",
+		}); err != nil {
+			result.PartialFailure = fmt.Sprintf("write %s: %v", dispatchAllowPath, err)
+		} else {
+			result.FilesSubstituted = append(result.FilesSubstituted, dispatchAllowPath)
+		}
+	}
+
+	// If substitution stalled because the generated branch wasn't writable in time,
+	// the repo IS created — say so clearly and point to the local finalize step,
+	// rather than leaking the raw "branch does not exist" (infra#179: gitea's
+	// template-generate is slow-async on this instance, so tool-side substitution
+	// is best-effort).
+	if strings.Contains(result.PartialFailure, "branch does not exist") ||
+		strings.Contains(result.PartialFailure, "not found") {
+		result.PartialFailure = fmt.Sprintf(
+			"repo created, but its branch (%s) was not writable within %ds — gitea's "+
+				"template-generate is slow-async on this instance (infra#179), so substitution "+
+				"is incomplete (%d file(s) done). Finalize locally with `hyperguild new-project` "+
+				"(clone + substitute, no API race). Underlying: %s",
+			branch, substitutionBudget, len(result.FilesSubstituted), result.PartialFailure)
+	}
+
 	// Fail loud: a scaffold that still holds placeholders does not build. Nothing
 	// substituted (with no explicit failure) means the walk found no placeholders —
 	// suspicious for a real template. Surface it instead of returning silent success.
@@ -185,6 +227,35 @@ func (t *CreateProjectFromTemplate) Call(ctx context.Context, raw json.RawMessag
 	return textOK(result)
 }
 
+// substitutionBudget bounds how long we retry the first write while the freshly
+// generated branch becomes writable. gitea's /generate returns (and serves reads)
+// before the branch ref is committed, so writes 404 "branch does not exist" for a
+// window. We keep the budget SHORT so the MCP call stays responsive: a healthy
+// gitea commits in ~1s and this catches it; a slow one (infra#179, observed >40s)
+// fails fast and we defer substitution with clear guidance rather than hang.
+const substitutionBudget = 5
+
+// upsertRetry retries UpsertFile on the transient post-generate "branch does not
+// exist" not-found, up to substitutionBudget. The write itself is the readiness
+// probe — BranchExists reports the branch present before writes succeed.
+func (t *CreateProjectFromTemplate) upsertRetry(ctx context.Context, owner, name, path string, args gitea.UpsertFileArgs) error {
+	var err error
+	for i := 0; i < substitutionBudget; i++ {
+		if _, err = t.c.UpsertFile(ctx, owner, name, path, args); err == nil {
+			return nil
+		}
+		if !errors.Is(err, gitea.ErrNotFound) {
+			return err
+		}
+		select {
+		case <-ctx.Done():
+			return err
+		case <-time.After(time.Second):
+		}
+	}
+	return err
+}
+
 // substituteEntry substitutes placeholders in one blob. If the path carries a
 // placeholder it renames the file (write new + delete old); otherwise it rewrites
 // content in place when changed. Returns a human-readable description of what was
@@ -212,7 +283,7 @@ func (t *CreateProjectFromTemplate) substituteEntry(ctx context.Context, owner,
 	enc := base64.StdEncoding.EncodeToString([]byte(newContent))
 
 	if renamed {
-		if _, err := t.c.UpsertFile(ctx, owner, name, newPath, gitea.UpsertFileArgs{
+		if err := t.upsertRetry(ctx, owner, name, newPath, gitea.UpsertFileArgs{
 			Branch:  branch,
 			Content: enc,
 			Message: fmt.Sprintf("template: substitute + rename %s -> %s", path, newPath),
@@ -229,7 +300,7 @@ func (t *CreateProjectFromTemplate) substituteEntry(ctx context.Context, owner,
 		return path + " -> " + newPath, ""
 	}
 
-	if _, err := t.c.UpsertFile(ctx, owner, name, path, gitea.UpsertFileArgs{
+	if err := t.upsertRetry(ctx, owner, name, path, gitea.UpsertFileArgs{
 		Branch:  branch,
 		Content: enc,
 		Message: "template: substitute placeholders",

internal/tools/create_project_from_template_test.go

@@ -61,13 +61,13 @@ func (f *fakeTemplateServer) handler(t *testing.T, tmpl, dest string) http.Handl
 				return
 			}
 			f.repoGetsPost++
-			_, _ = w.Write([]byte(fmt.Sprintf(`{"name":%q,"full_name":"mathias/%s","default_branch":"main","clone_url":"c","html_url":"h","template":false}`, dest, dest)))
+			_, _ = fmt.Fprintf(w, `{"name":%q,"full_name":"mathias/%s","default_branch":"main","clone_url":"c","html_url":"h","template":false}`, dest, dest)
 
 		case r.Method == http.MethodPost && p == "/api/v1/repos/mathias/"+tmpl+"/generate":
 			f.generated = true
 			w.WriteHeader(http.StatusCreated)
-			_, _ = w.Write([]byte(fmt.Sprintf(`{"name":%q,"full_name":"mathias/%s","default_branch":%q,"clone_url":"http://gitea.example.com/mathias/%s.git","html_url":"http://gitea.example.com/mathias/%s","template":false}`,
+			_, _ = fmt.Fprintf(w, `{"name":%q,"full_name":"mathias/%s","default_branch":%q,"clone_url":"http://gitea.example.com/mathias/%s.git","html_url":"http://gitea.example.com/mathias/%s","template":false}`,
-				dest, dest, f.genBranch, dest, dest)))
+				dest, dest, f.genBranch, dest, dest)
 
 		case r.Method == http.MethodGet && strings.HasPrefix(p, "/api/v1/repos/mathias/"+dest+"/git/trees/"):
 			var entries []string
@@ -76,7 +76,7 @@ func (f *fakeTemplateServer) handler(t *testing.T, tmpl, dest string) http.Handl
 			}
 			// include a tree (directory) entry to exercise the blob filter
 			entries = append(entries, `{"path":"cmd","type":"tree","sha":"treesha"}`)
-			_, _ = w.Write([]byte(fmt.Sprintf(`{"sha":"root","tree":[%s],"truncated":false}`, strings.Join(entries, ","))))
+			_, _ = fmt.Fprintf(w, `{"sha":"root","tree":[%s],"truncated":false}`, strings.Join(entries, ","))
 
 		case r.Method == http.MethodGet && strings.HasPrefix(p, "/api/v1/repos/mathias/"+dest+"/contents/"):
 			path := strings.TrimPrefix(p, "/api/v1/repos/mathias/"+dest+"/contents/")
@@ -86,8 +86,8 @@ func (f *fakeTemplateServer) handler(t *testing.T, tmpl, dest string) http.Handl
 				_, _ = w.Write([]byte(`{"message":"not found"}`))
 				return
 			}
-			_, _ = w.Write([]byte(fmt.Sprintf(`{"path":%q,"sha":"sha-%s","size":1,"content":%q,"encoding":"base64"}`,
+			_, _ = fmt.Fprintf(w, `{"path":%q,"sha":"sha-%s","size":1,"content":%q,"encoding":"base64"}`,
-				path, strings.ReplaceAll(path, "/", "-"), encb64(body))))
+				path, strings.ReplaceAll(path, "/", "-"), encb64(body))
 
 		// POST = create (new/renamed file, no sha), PUT = update (existing, with sha).
 		case (r.Method == http.MethodPost || r.Method == http.MethodPut) && strings.HasPrefix(p, "/api/v1/repos/mathias/"+dest+"/contents/"):
@@ -226,6 +226,38 @@ func TestCreateProject_WriteFailure_PartialFailure(t *testing.T) {
 	assert.Contains(t, out.PartialFailure, "go.mod")
 }
 
+// dispatch_allow injects a .dispatch-allow file (dispatch#3) only when true.
+func TestCreateProject_DispatchAllow(t *testing.T) {
+	tests := []struct {
+		name     string
+		argsJSON string
+		wantFile bool
+	}{
+		{"true injects .dispatch-allow", `{"owner":"mathias","name":"new-svc","dispatch_allow":true}`, true},
+		{"false does not inject", `{"owner":"mathias","name":"new-svc","dispatch_allow":false}`, false},
+		{"omitted does not inject", `{"owner":"mathias","name":"new-svc"}`, false},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			files := map[string]string{"go.mod": "module __MODULE_PATH__\n"}
+			f := newFakeTemplateServer(files, "main")
+			srv := httptest.NewServer(f.handler(t, "template-go-agent", "new-svc"))
+			defer srv.Close()
+
+			out := callTool(t, srv.URL, "template-go-agent", tc.argsJSON)
+			require.Empty(t, out.PartialFailure)
+			if tc.wantFile {
+				assert.Contains(t, out.FilesSubstituted, ".dispatch-allow")
+				assert.Contains(t, f.puts, ".dispatch-allow")
+				assert.Contains(t, f.putBodies[".dispatch-allow"], "dispatch#3")
+			} else {
+				assert.NotContains(t, out.FilesSubstituted, ".dispatch-allow")
+				assert.NotContains(t, f.puts, ".dispatch-allow")
+			}
+		})
+	}
+}
+
 // ── guardrails unchanged by the rewrite ──────────────────────────────────────
 
 func TestCreateProject_NameRegexFailure(t *testing.T) {

internal/tools/workflow_run_trigger.go

@@ -4,12 +4,22 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"time"
 
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/allowlist"
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/gitea"
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/registry"
 )
 
+// dispatchResolveBudget bounds how many times we poll for the dispatched run to
+// appear, and dispatchResolvePoll is the gap between polls. Gitea returns 204
+// with no run ID, so we list workflow_dispatch runs and take the newest above
+// the pre-dispatch baseline; it may register a beat after the 204.
+const (
+	dispatchResolveBudget = 5
+	dispatchResolvePoll   = time.Second
+)
+
 // WorkflowRunTrigger triggers a Gitea Actions workflow_dispatch run.
 type WorkflowRunTrigger struct {
 	c       *gitea.Client
@@ -68,17 +78,82 @@ func (t *WorkflowRunTrigger) Call(ctx context.Context, raw json.RawMessage) (jso
 		}
 	}
 
-	result, err := t.c.DispatchWorkflow(ctx, args.Owner, args.Name, args.Workflow, gitea.DispatchWorkflowArgs{
+	// Snapshot the newest existing workflow_dispatch run BEFORE dispatching, so we
+	// can tell our fresh run apart from a prior one (Gitea's 204 carries no run ID).
+	baseline := t.newestDispatchRunID(ctx, args.Owner, args.Name, args.Workflow, ref)
+
+	if err := t.c.DispatchWorkflow(ctx, args.Owner, args.Name, args.Workflow, gitea.DispatchWorkflowArgs{
 		Ref:    ref,
 		Inputs: args.Inputs,
-	})
+	}); err != nil {
-	if err != nil {
 		return nil, err
 	}
 
-	htmlURL := fmt.Sprintf("%s/%s/%s/actions/runs/%d", t.baseURL, args.Owner, args.Name, result.RunID)
+	// Resolve the new run by listing workflow_dispatch runs and taking the newest
+	// one whose ID exceeds the baseline. Poll briefly: the run can register a beat
+	// after the 204.
+	var run *gitea.WorkflowRun
+	for i := 0; i < dispatchResolveBudget; i++ {
+		if i > 0 {
+			select {
+			case <-ctx.Done():
+				return nil, ctx.Err()
+			case <-time.After(dispatchResolvePoll):
+			}
+		}
+		if r := t.newestDispatchRun(ctx, args.Owner, args.Name, args.Workflow, ref); r != nil && r.ID > baseline {
+			run = r
+			break
+		}
+	}
+
+	// Dispatch succeeded (204). If the run has not surfaced yet, say so honestly
+	// rather than failing — the workflow IS firing; the caller can list runs.
+	if run == nil {
+		return textOK(map[string]any{
+			"dispatched": true,
+			"note": "dispatch accepted but the run did not register within the resolve window; " +
+				"list recent workflow_dispatch runs to find it",
+		})
+	}
+
+	htmlURL := run.HTMLURL
+	if htmlURL == "" {
+		htmlURL = fmt.Sprintf("%s/%s/%s/actions/runs/%d", t.baseURL, args.Owner, args.Name, run.ID)
+	}
 	return textOK(map[string]any{
-		"run_id":   result.RunID,
+		"dispatched": true,
-		"html_url": htmlURL,
+		"run_id":     run.ID,
+		"html_url":   htmlURL,
 	})
 }
+
+// newestDispatchRun returns the most recent workflow_dispatch run for the given
+// workflow and ref, or nil if none / on listing error (best-effort resolution).
+func (t *WorkflowRunTrigger) newestDispatchRun(ctx context.Context, owner, name, workflow, ref string) *gitea.WorkflowRun {
+	resp, err := t.c.ListWorkflowRuns(ctx, owner, name, gitea.ListWorkflowRunsArgs{
+		Event:    "workflow_dispatch",
+		Workflow: workflow,
+		Branch:   ref,
+		Limit:    20,
+	})
+	if err != nil || resp == nil {
+		return nil
+	}
+	var newest *gitea.WorkflowRun
+	for i := range resp.WorkflowRuns {
+		r := &resp.WorkflowRuns[i]
+		if newest == nil || r.ID > newest.ID {
+			newest = r
+		}
+	}
+	return newest
+}
+
+// newestDispatchRunID is newestDispatchRun's ID, or 0 if none.
+func (t *WorkflowRunTrigger) newestDispatchRunID(ctx context.Context, owner, name, workflow, ref string) int64 {
+	if r := t.newestDispatchRun(ctx, owner, name, workflow, ref); r != nil {
+		return r.ID
+	}
+	return 0
+}

internal/tools/workflow_run_trigger_test.go

@@ -3,8 +3,10 @@ package tools_test
 import (
 	"context"
 	"encoding/json"
+	"io"
 	"net/http"
 	"net/http/httptest"
+	"strings"
 	"testing"
 
 	"gitea.d-ma.be/mathias/gitea-mcp/internal/allowlist"
@@ -14,10 +16,67 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-func TestWorkflowRunTriggerSuccess(t *testing.T) {
+// runsListJSON is a workflow_runs listing body with a single run of the given id.
-	// Fake server handles both the repo endpoint (default_branch) and the dispatch endpoint.
+func runsListJSON(id int) string {
+	return `{"total_count":1,"workflow_runs":[{"id":` +
+		fmtInt(id) +
+		`,"status":"queued","event":"workflow_dispatch","html_url":"http://gitea.example/mathias/myrepo/actions/runs/` +
+		fmtInt(id) + `"}]}`
+}
+
+func fmtInt(i int) string { b, _ := json.Marshal(i); return string(b) }
+
+// The dispatch endpoint returns 204 with NO Location header (real Gitea). The
+// tool must treat that as success, forward inputs, and resolve the new run by
+// listing workflow_dispatch runs and picking the newest one above the
+// pre-dispatch baseline.
+func TestWorkflowRunTriggerResolvesRunViaListing(t *testing.T) {
+	dispatched := false
+	var gotBody []byte
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch {
+		case r.URL.Path == "/api/v1/repos/mathias/myrepo/actions/workflows/ci.yml/dispatches" && r.Method == http.MethodPost:
+			gotBody, _ = io.ReadAll(r.Body)
+			dispatched = true
+			w.WriteHeader(http.StatusNoContent) // no Location header
+		case strings.HasPrefix(r.URL.Path, "/api/v1/repos/mathias/myrepo/actions/runs"):
+			w.Header().Set("Content-Type", "application/json")
+			if !dispatched {
+				_, _ = w.Write([]byte(`{"total_count":0,"workflow_runs":[]}`)) // baseline: none yet
+				return
+			}
+			_, _ = w.Write([]byte(runsListJSON(100)))
+		default:
+			http.NotFound(w, r)
+		}
+	}))
+	defer srv.Close()
+
+	tool := tools.NewWorkflowRunTrigger(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"mathias"}), srv.URL)
+	out, err := tool.Call(context.Background(), json.RawMessage(
+		`{"owner":"mathias","name":"myrepo","workflow":"ci.yml","ref":"main","inputs":{"issue_number":"36","harness":"agentsquad"}}`))
+	require.NoError(t, err)
+	assert.True(t, dispatched, "expected POST dispatch")
+
+	// inputs forwarded to the dispatch body
+	var body map[string]any
+	require.NoError(t, json.Unmarshal(gotBody, &body))
+	assert.Equal(t, "main", body["ref"])
+	inputs, ok := body["inputs"].(map[string]any)
+	require.True(t, ok, "inputs must be present in dispatch body")
+	assert.Equal(t, "36", inputs["issue_number"])
+	assert.Equal(t, "agentsquad", inputs["harness"])
+
+	// run id resolved via listing (not a Location header)
+	var result map[string]any
+	require.NoError(t, json.Unmarshal(out, &result))
+	assert.Equal(t, float64(100), result["run_id"])
+	assert.Contains(t, result["html_url"], "/actions/runs/100")
+}
+
+func TestWorkflowRunTriggerDefaultBranch(t *testing.T) {
 	repoHit := false
-	dispatchHit := false
+	dispatched := false
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		switch {
 		case r.URL.Path == "/api/v1/repos/mathias/myrepo" && r.Method == http.MethodGet:
@@ -25,9 +84,15 @@ func TestWorkflowRunTriggerSuccess(t *testing.T) {
 			w.Header().Set("Content-Type", "application/json")
 			_, _ = w.Write([]byte(`{"name":"myrepo","full_name":"mathias/myrepo","default_branch":"main"}`))
 		case r.URL.Path == "/api/v1/repos/mathias/myrepo/actions/workflows/ci.yml/dispatches" && r.Method == http.MethodPost:
-			dispatchHit = true
+			dispatched = true
-			w.Header().Set("Location", "/api/v1/repos/mathias/myrepo/actions/runs/42")
 			w.WriteHeader(http.StatusNoContent)
+		case strings.HasPrefix(r.URL.Path, "/api/v1/repos/mathias/myrepo/actions/runs"):
+			w.Header().Set("Content-Type", "application/json")
+			if !dispatched {
+				_, _ = w.Write([]byte(`{"total_count":0,"workflow_runs":[]}`))
+				return
+			}
+			_, _ = w.Write([]byte(runsListJSON(55)))
 		default:
 			http.NotFound(w, r)
 		}
@@ -38,37 +103,10 @@ func TestWorkflowRunTriggerSuccess(t *testing.T) {
 	out, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"mathias","name":"myrepo","workflow":"ci.yml"}`))
 	require.NoError(t, err)
 	assert.True(t, repoHit, "expected GET /repo for default branch")
-	assert.True(t, dispatchHit, "expected POST dispatch")
 
 	var result map[string]any
 	require.NoError(t, json.Unmarshal(out, &result))
-	assert.Equal(t, float64(42), result["run_id"])
+	assert.Equal(t, float64(55), result["run_id"])
-	assert.Contains(t, result["html_url"], "/mathias/myrepo/actions/runs/42")
-}
-
-func TestWorkflowRunTriggerExplicitRef(t *testing.T) {
-	repoHit := false
-	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path == "/api/v1/repos/mathias/myrepo" {
-			repoHit = true
-		}
-		if r.URL.Path == "/api/v1/repos/mathias/myrepo/actions/workflows/ci.yml/dispatches" {
-			w.Header().Set("Location", "/api/v1/repos/mathias/myrepo/actions/runs/99")
-			w.WriteHeader(http.StatusNoContent)
-			return
-		}
-		http.NotFound(w, r)
-	}))
-	defer srv.Close()
-
-	tool := tools.NewWorkflowRunTrigger(gitea.NewClient(srv.URL, "tok"), allowlist.New([]string{"mathias"}), srv.URL)
-	out, err := tool.Call(context.Background(), json.RawMessage(`{"owner":"mathias","name":"myrepo","workflow":"ci.yml","ref":"develop"}`))
-	require.NoError(t, err)
-	assert.False(t, repoHit, "should not call GET /repo when ref is provided")
-
-	var result map[string]any
-	require.NoError(t, json.Unmarshal(out, &result))
-	assert.Equal(t, float64(99), result["run_id"])
 }
 
 func TestWorkflowRunTriggerAllowlistRejects(t *testing.T) {