cad-atlas/internal/atlas/atlas.json

{
  "substrate": [
    {"n":"koala","k":"RTX 5070 · k3s control-plane · Gitea · LiteLLM :30401 · llama-swap :31234 · searxng"},
    {"n":"iguana","k":"M2 Ultra · Ollama / mlx"},
    {"n":"flamingo","k":"daily driver · ~/dev"},
    {"n":"piguard","k":"NGINX reverse-proxy · ntfy"}
  ],
  "ns": "Tailscale mesh · ns: ai-stack · supervisor(→brain) · gitea-mcp · infra-mcp · council",
  "stages": [
    {"no":"STAGE 00","short":"Notice","title":"Signals","plain_title":"Notice what's happening","plain":"New ideas and developments worth reacting to are collected — mostly an automated daily/weekly scan of AI news, plus things saved by hand.","path":"→ mathias/signals",
     "trans_label":"Does it matter to us?","trans":"A raw signal only advances if it connects to something we actually care about. Most captured signals stop here; the few that touch the mission get pulled up against a goal.","nodes":[
      {"t":"Applied AI Radar","plain_t":"Automated news scan","plain":"An automated scan reads AI news every day (and deeper every week) and keeps only claims backed by a real paper, benchmark, code, or named lab.","d":"Daily Tier-1 + weekly Tier-2 deep pass. Verified-primary bar (paper/benchmark/code/named-lab).","tags":["cron · daily/weekly","→ signals #1–26+"]},
      {"t":"Manual capture","plain_t":"Saved by hand","plain":"Anything interesting spotted by hand gets saved into the same inbox.","d":"claude.ai strategic drop · brain capture tool.","tags":["ad-hoc"]},
      {"t":"Aspirational surfaces","plain_t":"Not built yet","plain":"Planned but not built yet: sending ideas in by Telegram, voice, or a URL.","pill":"var(--dim)","d":"Telegram / voice / URL → inbox. NOT built.","tags":["gap"]}
    ]},
    {"no":"STAGE 01","short":"Why","cls":"telos","title":"TELOS","plain_title":"Why we're here","plain":"The mission, goals, and problems we're actually trying to solve live here — every piece of work downstream has to trace back to one of these goals.","path":"wiki/telos/",
     "trans_label":"Worth a session?","trans":"A goal or problem on the board becomes the seed for a design session when it's decided worth working on now. The goal is the input the session must trace back to.","nodes":[
      {"t":"Intention substrate","plain_t":"The goal board","plain":"The master list of mission, goals, problems, and current status — the yardstick everything downstream is measured against.","pill":"var(--violet)","d":"Mission · goals · problems · strategies · status. Every downstream item traces to a goal.","tags":["brain_query wing=telos"]}
    ]},
    {"no":"STAGE 02","short":"Think","title":"Strategic session","plain_title":"Think it through","plain":"A human and AI models work out what to do and why, debating hard calls and writing down the decision and what \"done\" will mean.","path":"claude.ai frontier + brain MCP",
     "trans_label":"Decision reached","trans":"It advances only when the thinking converges on a decision and is written down as a concrete, testable specification — not while the answer is still open.","nodes":[
      {"t":"Design · ADRs · specs","plain_t":"Decide the approach","plain":"A human and a top-tier AI model figure out the approach and write down the decision plus what a finished result must prove.","d":"Human + frontier model. ISC acceptance criteria written here.","tags":["Define / converge"]},
      {"t":"🏛️ LLM Council","plain_t":"AI review panel","plain":"For hard calls, several AI models answer independently, anonymously critique each other, and a \"chair\" model synthesises one verdict — reducing any single model's bias.","cls":"council","pill":"var(--violet)","d":"fan-out → anonymous cross-review → chairman synth. glm-4.7-flash · qwen36-35b · gemma4-31b (chair).","tags":["hard strategic Q","chat.d-ma.be"]},
      {"t":"Autoresearch Council","plain_t":"Research review panel","plain":"A parallel version of the same review that vets research findings before they're allowed through.","cls":"council","pill":"var(--violet)","d":"Sibling pipe — ratifies research before the gate.","tags":["proposed: → standalone svc"]}
    ]},
    {"no":"STAGE 03","short":"Write order","title":"Spec → Gitea issue","plain_title":"Write the work order","plain":"The decision is turned into a precise, self-contained work order an AI agent can execute unsupervised — with a pass/fail definition of done, a risk rating, and a tamper-proof seal.","path":"agent-ready contract","generate":"gitea-issues",
     "trans_label":"Sealed & agent-ready","trans":"Advances to the gate only when the spec is a complete contract: a pass/fail test, a risk tier, a regulatory note, no open human dependencies, one embedded Oath, and a valid cryptographic signature. A malformed or unsigned order fails closed and never reaches the gate.","nodes":[
      {"t":"Contract enforced","plain_t":"The work-order rules","plain":"The work order must have a clear pass/fail test, a risk rating, a regulatory-risk note, and no unfinished human dependencies before it counts as agent-ready.","d":"Binary ISC · declared risk tier · reg-risk assessment · no open human deps.","tags":["LOW / MED / HIGH"]},
      {"t":"Admission controller","plain_t":"Tamper-proof seal","plain":"The work order is cryptographically signed when created, so any later tampering is detectable and the eventual change can be checked against it.","d":"Ed25519-sign issue body at creation (#36). Verify sig + PR alignment at infra boundary.","tags":["chain of custody"]},
      {"t":"⚖️ var-go Oath","plain_t":"Definition of done","plain":"A machine-checkable \"definition of done\" is embedded in the work order — exactly one, or the order is rejected — later used to prove the result actually meets the spec.","cls":"oath","pill":"var(--gold)","d":"Acceptance contract embedded in the issue as a var fenced block. Exactly one — zero/multiple fail closed. Prose → typed steps; failures anchored to byte spans.","tags":["swedsl · var-go","defined here → enforced @06"]}
    ]},
    {"no":"STAGE 04","short":"Human go","cls":"gate","title":"Human dispatch gate","plain_title":"Human says go","plain":"A person reviews the work order and its risk and decides whether to release it — the one and only checkpoint where work does not move on its own.","path":"the only checkpoint","generate":"stuck-badges",
     "trans_label":"A human said go","trans":"The hard stop. Nothing crosses automatically — a person must review the plan and risk and explicitly release it, and the repo must be on the allow-list, before any agent starts. This is the single human checkpoint in the whole pipeline.","nodes":[
      {"t":"Human triggers execution","plain_t":"The go button","plain":"A person confirms the plan and its risk level, then releases the work — nothing runs until they do.","cls":"gateway","pill":"var(--amber)","d":"Ratify proposed-plan + risk tier, then dispatch.","gate":true},
      {"t":"Session-Dispatch bridge","plain_t":"Hand-off to agents","plain":"The approval flips a switch that hands the signed work order over to the agents to start execution.","cls":"bridge","pill":"var(--blue)","d":"claude.ai MCP → gitea:workflow_run_trigger → cad-dispatch.yml → agentsquad. The final design→execution bridge.","tags":["workflow_dispatch"]}
    ]},
    {"no":"STAGE 05","short":"Build","cls":"exec","title":"Execute · agentsquad + dispatch","plain_title":"Agents do the work","plain":"AI agents actually build the thing — one writes, a second independent one reviews it to avoid marking its own homework — and every step is logged for the audit trail. Two execution paths run today: agentsquad for orchestrated multi-agent jobs, and dispatch for headless single-issue work.","path":"koala · agentsquad-serve + dispatch CronJob","generate":"stuck-badges",
     "trans_label":"Change proposed","trans":"Advances when the agents finish and open a proposed change (a PR) with its audit log attached. Until there's a concrete change to test, nothing moves.","nodes":[
      {"t":"Task API","plain_t":"Start a job","plain":"A request kicks off a job and hands back an id you can poll for progress.","pill":"var(--coral)","d":"POST /tasks → job id · GET /tasks/{id}. taskqueue + serve (v0.12+).","tags":["single agentsquad.yaml"]},
      {"t":"Executor + reviewer loop","plain_t":"Build + independent review","plain":"One agent does the work; a second, independent agent on a different model reviews it — so nothing marks its own homework.","cls":"win","pill":"var(--coral)","d":"ADK Go + LiteLLM. Frontier models (local qwen spirals). Reviewer on distinct tier — echo-chamber prevention.","risk":true},
      {"t":"dma-cli · routing + scope","plain_t":"Router & guardrails","plain":"A router sends each agent to the right AI backend and enforces what it is and isn't allowed to touch, with a confirmation gate as a guardrail.","cls":"bridge","pill":"var(--blue)","d":"Harness-config arm: routes agents to the right LLM backend. Three-layer scope policy + confirmation gate = CAD guardrail.","tags":["backend routing","scope guardrail"]},
      {"t":"assessor-loop ledger","plain_t":"Audit log","plain":"Every step is recorded in a tamper-evident log so the whole run can be audited afterwards.","d":"Attestation ledger (audit trail) + brain session_log on completion.","tags":["audit package"]},
      {"t":"dispatch · headless issue runs","plain_t":"Picking up work by itself","plain":"A watcher checks the tracker every minute for issues marked ready, and runs one agent per issue on its own — no human starts it. Whether the work is really done is decided by the automated checks, never by the agent saying so.","pill":"var(--coral)","d":"CronJob polls status:dispatch in allow-listed repos (.dispatch-allow marker, fail-closed). Claims, runs, reports; CI-green on the pushed commit is the only done-signal.","tags":["1-min poll","CI is ground truth"]},
      {"t":"Two independent executors","plain_t":"Two different AI providers","plain":"The same job can be run by either of two different AI providers, chosen per issue. When one is unavailable — an empty account stopped everything for six weeks — the other still works.","cls":"win","pill":"var(--blue)","d":"executor:<name> label routes to claude (Anthropic SDK runner) or kimi (berget/Kimi-K2.6 via LiteLLM, EU-metered, Go-native loop). Provider independence after a 6-week single-provider outage.","tags":["executor: label","EU option"],"risk":true},
      {"t":"Fail-closed tool policy","plain_t":"What an agent may touch","plain":"Each agent may only use an approved short list of tools and commands; anything else is refused outright rather than asked about, since no human is watching. Its file access cannot leave its own copy of the repo, and each run has a hard limit on both steps and money spent.","cls":"gateway","pill":"var(--amber)","d":"One executor.Policy for both executors: tool + bash-prefix allowlist, closed env-prefix list, exact MCP tool names. Writes confined to the checkout (symlink-resolved, O_NOFOLLOW). Per-run turn and USD ceilings; spend ledgered against a rolling cap.","tags":["deny by default","per-run ceilings"]},
      {"t":"Read-only PR review","plain_t":"A second agent reads the change","plain":"A pull request can be handed to an agent that may only read and comment, never edit. Its review is thrown away unless it points at a file the change actually touched — which catches a reviewer that skimmed the title instead of the code. New commits re-open the review by themselves.","pill":"var(--coral)","d":"status:review → read-only tools (no write, no shell). Published only if grounded: must cite a file in the PR's changed-files list. Handling anchored to the head sha, so a push re-opens it; attempts capped per revision.","tags":["grounded or dropped","re-opens on push"]}
    ]},
    {"no":"STAGE 06","short":"Check","title":"PR → CI","plain_title":"Automatic quality checks","plain":"The proposed change is run through automated tests and safety checks — including a check that it actually satisfies the work order's definition of done — and only a clean pass lets it continue.","path":"Gitea Actions · cd.yml (live)","generate":"ci-jobs",
     "trans_label":"All checks green","trans":"Advances only if every automated check passes — tests, linters, security scan, and the Oath check proving it meets the original work order. Any red gate stops it here; a passing reviewer is not enough to override a failed Oath.","nodes":[
      {"t":"Automated checks","plain_t":"Quality checks","plain":"Tests, linters, a security scan, plus a check that the change actually meets the work order — all must pass to continue.","d":"go test · vet · lint · govulncheck + var-go/oath gate.","tags":["green = proceed"]}
    ]},
    {"no":"STAGE 07","short":"Ship","cls":"cd","title":"CD → pod","plain_title":"Ship it","plain":"Once everything is green, the change is deployed automatically to the live server — with the rule that merging code alone doesn't ship it; the release has to be pointed at the new version.","path":"Flux GitOps → k3s","generate":"deploy-state",
     "trans_label":"It's live","trans":"Once the new version is actually running on the server, the deployed outcome becomes the input to scoring. Advancing means shipped and observable, not just merged.","nodes":[
      {"t":"Deploy on green","plain_t":"Auto-deploy when green","plain":"When all checks pass, the release system rolls the new version onto the live server automatically — but only once the release is pointed at that version (merging code alone doesn't ship it).","pill":"var(--green)","d":"Flux reconciles image → k3s pod on koala. Push ≠ deploy: bump tag in mathias/infra.","tags":["ntfy on deploy"]}
    ]},
    {"no":"STAGE 08","short":"Learn","cls":"telos","title":"Loop back","plain_title":"Did it work?","plain":"The result is scored against the goal that started it and fed back into the mission board, so the next round of planning learns from what shipped.","path":"→ TELOS (feedback bus)","generate":"stuck-badges",
     "trans_label":"What did we learn?","trans":"The scored outcome flows back into the mission board so goals, problems, and priorities update — the loop that makes the pipeline a cycle rather than a line. Partly manual today; an explicit improvement target.","nodes":[
      {"t":"Close the loop","plain_t":"Score & feed back","plain":"The outcome is scored against the goal that started it and written back to the mission board, so future planning learns from what actually shipped.","pill":"var(--violet)","d":"session_log + attestation → brain. Score deploy outcome vs originating goal. (arc partly manual — improvement target.)","tags":["continuous"]}
    ]}
  ]
}