[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75W77MFRAPPZYDA4RJN2Z2W":3,"skill-same:01M75W77MFRAPPZYDA4RJN2Z2W":79},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":29,"updatedAt":30,"install":31,"resources":44},"01M75W77MFRAPPZYDA4RJN2Z2W","godmode",null,"Autonomous product factory. A finite programme delivers a bounded roadmap. A perpetual programme continuously pairs independent models to discover evidence-backed value, create and execute goals, run full regression, repair confirmed fallout, and create the next progression until a hard safety, budget, or deadline boundary fires.","stevengonsalvez\u002Fainb-toolkit","stevengonsalvez","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F9320602?v=4","01M75W59QRYZYANWY3XG8CJB1P","Steven Gonsalvez","01M75W59RD5YW1RT67BDV0EAZF",1,"MIT","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: godmode\ndescription: \"Autonomous product factory. A finite programme delivers a bounded roadmap. A perpetual programme continuously pairs independent models to discover evidence-backed value, create and execute goals, run full regression, repair confirmed fallout, and create the next progression until a hard safety, budget, or deadline boundary fires.\"\n---\n\n# godmode — the autonomous product factory\n\nA runtime orchestrator that OWNS the loop. It composes existing machinery —\n`\u002Floop` (driver), `Workflow` (staged ultracode), beads (state), `\u002Fhere-now`\n(dashboard hosting), `tmux-verify`\u002Fbrowser-harness\u002FPlaywright (verify lanes) —\ninto one self-driving programme. You are the driver each tick; these files are\nyour constitution and playbooks.\n\n## Entry modes\n\n| Invocation | Action |\n|---|---|\n| `\u002Fgodmode \u003Cnorth-star> --mode finite\\|perpetual [--approval none\\|roadmap] [--no-court] [--budget \u003Ctokens>] [--deadline \u003CISO>]` (alias: `\u002Fgodmode init \u003Cnorth-star> ...`) | INIT: generate charter + state + dashboard + beads, then begin finite delivery or perpetual evolution |\n| `\u002Fgodmode run [--take-over]` | Resume\u002Fcontinue the loop from state (any session, incl. post-crash). Claims the driver lease first: refuses while another session's lease is fresh, auto-claims a stale one, `--take-over` forces after an AskUserQuestion confirm. On a fresh machine: `sync.sh discover`, pick the slug, claim, then `sync.sh adopt \u003Cslug>` reconstructs local state |\n| `\u002Fgodmode status` | Standup: read state + beads + dashboard + lease, report, change nothing. NEVER adopts (observer-safe: no scratch state means hooks stay inert). Preflight prints \"status publishing DISABLED: missing \u003CX>\" when here-now\u002Fcredentials are absent |\n| `\u002Fgodmode pause` | Stop re-arming the loop; state stays resumable; release the lease (`lease.sh release`) |\n\n## Provider parity\n\n| Capability | Claude | Codex | Copilot |\n|---|---|---|---|\n| init \u002F run (drive the loop) | native scheduler | native Codex automation required | external scheduler adapter required |\n| status \u002F discover | yes | yes | yes |\n| dashboard publish + sidecar sync + lease guard | all hooks | hooks.json (shared format) | sessionStart pull + staleness nudge |\n\nNever claim a provider is driving when its scheduler or peer-model adapter is\nmissing. Record the capability result in state and defer that lane visibly.\n\n## The three-layer machine\n\n```\nLAYER 1  CHARTER   .agents\u002Fgoals\u002F\u003Cslug>-charter.md   — constitution, re-read EVERY tick\nLAYER 2  DRIVER    \u002Floop dynamic mode                — ScheduleWakeup ~600s, same prompt re-entered\nLAYER 3  WORKFLOWS one ultracode Workflow per stage  — resumable via resumeFromRunId\nSTATE    beads (epic\u002Ffeature nodes) + .agents\u002Fscratch\u002F\u003Cslug>-state.json (working\n         copy) + hook-maintained sidecar mirror {state,charter,lease} on the\n         dedicated ref refs\u002Fgodmode\u002F\u003Cslug> (cross-machine; GODMODE_SYNC=local opts out)\n```\n\n## Pipeline (stage names are the primary vocabulary; W-numbers are aliases)\n\n```\nSIGNAL                   TTL    ONE model gathers CITED external evidence: comparable\n        │                       products, user-reported gaps, constraining standards.\n        │                       Reads the append-only ledger first and returns only what is\n        │                       NEW; parked rows keep their reason and are never re-proposed.\n        │                       \u002Fresearch if available, else WebSearch+WebFetch. Charter-gated;\n        │                       unavailable degrades to repo-only discovery, never a failure.\n        ▼\nDISCOVER                 once   BRAIN brainstorms the nirvana landscape from the\n        │                       north-star + a repo scan + the SIGNAL rows + any existing\n        │                       backlog → writes the feature REGISTRY\n        │                       (.agents\u002Fplans\u002F\u003Cslug>-registry.md), provenance per row\n        │                       (repo | signal | user). User lists appended, not replaced.\n        ▼\nFEASIBILITY COURT (W0)   once   every REGISTRY idea → verdict: feasible | downgrade | park+blocker\n        │                       skipped only by explicit --no-court (registry taken verbatim)\n        ▼\nROADMAP (W1)             once   cluster survivors → ordered epics + parallel-pair candidates,\n        │                       seed beads (one per epic + per feature)\n        ▼\n[OPTIONAL ROADMAP GATE]         only when approval_policy is roadmap\n        ▼\nper epic, serial on stacked branches:\n   PLAN (W2)      planner → adversarial review → revise → VERIFY THE REVISE EDITED THE\n        │         PLAN FILES (mtime\u002Fcontent) — see references\u002Flessons.md\n        ▼\n   EXECUTE (W3)   build (model pair per workstream) → pair review → adversarial epic\n        │         review (+fix) → VERIFY (surface lane, references\u002Fverify-lanes.md)\n        ▼         → bounded fix loop (≤2) → build gate (unpiped!)\n   SHIP           stacked PR (labelled for review) → close epic+feature beads with\n        │         evidence notes → dashboard update\n        ▼\nPERPETUAL         after every ship: full regression + refresh SIGNAL if stale +\n                  next discovery\u002Fplan. Backlog-dry refreshes SIGNAL first and\n                  becomes adaptive evidence research, never DONE\nTERMINATION       finite backlog-dry | --budget exhausted | --deadline reached |\n                  security, production-safety, authority, or lease loss\n```\n\nPerpetual progression: after each epic ships, start cumulative regression in\nan isolated lane. At the same time, the completeness critic re-enters Discover\nand creates evidence-backed candidates for the next generation, refreshing the\nexternal Signal first when it is stale, so a dry backlog is replenished from\nnew outside evidence rather than from re-reading the same repo. The refresh\nappends to the signal ledger and dedupes against every row ever recorded,\nparked ones included, so the programme does not pay to rediscover a dead end it\nalready rejected. A confirmed\ndefect preempts the single mutation lane. Discovery and planning continue in\nparallel. An empty Court result enters adaptive-backoff research, never DONE.\n\n## Creative quorum\n\nEvery creative decision has two independent model perspectives: Discover,\nFeasibility Court, Roadmap, completeness critic, incident root-cause analysis,\nand strategic reprioritisation. Signal is not one of them: gathering cited\nexternal evidence is retrieval, so it runs on one model and is judged by its\ncitations, not by quorum. Resolve the route before launching work with:\n\n```bash\n${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Fprogramme_policy.py creative-route \u003Cavailability.json>\n```\n\n- Claude primary pairs a Claude creative model with `codex:codex-rescue`.\n- Codex primary pairs a Codex model with Claude Fable.\n- Copilot primary uses two distinct Copilot models, then enriches with Claude\n  or Codex when available.\n- Fallback uses any two distinct available models. One model means creative\n  work is deferred, while regression and grounded work continue.\n- Persist both proposals, model identifiers, evidence, disagreement, and the\n  synthesis. A third model synthesises when available; otherwise use the\n  evidence rubric and preserve dissent.\n\nParallel epics: allowed ONLY if blessed AND file-sets provably disjoint AND\nworktrees available; otherwise serialise (single worktree = single checkout).\n\n## INIT procedure\n\n1. Parse the north-star + flags. Derive `\u003Cslug>`.\n2. Generate the charter from `references\u002Fcharter-template.md` — fill outcome,\n   model policy (resolve fable toggle), verify doctrine, termination, stop\n   rules, dashboard slug. Write `.agents\u002Fgoals\u002F\u003Cslug>-charter.md`.\n3. Write `.agents\u002Fscratch\u002F\u003Cslug>-state.json` (schema:\n   `references\u002Fstate-and-beads.md`) with `driver_session_id: null` and a\n   `phase_since` timestamp. You cannot know your own session_id: the\n   PostToolUse hook fills `driver_session_id` from the hook event on this very\n   write (first-writer-wins), which is why the write must precede the claim.\n   Then claim the driver lease:\n   `bash ${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Flease.sh claim \u003Crepo> \u003Cslug>`.\n4. Dashboard publishing is HOOK-OWNED: every state.json write triggers\n   render + publish + sidecar push (PostToolUse). Your duties: keep\n   state.json truthful, write `current_note`, pick a stable\n   `dashboard_slug`. Publish failure never blocks (pending marker +\n   dashboard banner; hooks retry on the next state write).\n5. Run DISCOVER (template: `references\u002Fstage-workflows.md`) → writes the\n   registry. Then, unless `--no-court`, the Feasibility Court workflow, then\n   Roadmap. With `--no-court`: registry taken verbatim, straight to Roadmap.\n6. If `approval_policy` is `roadmap`, present the roadmap, set\n   `human_gate: \"pending\"`, and wait. Otherwise set `human_gate: \"none\"` and\n   start the loop immediately. Perpetual mode normally uses `none`.\n\n## LOOP protocol (every wake — this is the driver contract)\n\n1. Read the charter + state.json. They outrank your memory of last tick. Run\n   `programme_policy.py validate-state \u003Cstate>` before any side effect. If an\n   optional `human_gate` is pending, do not re-arm.\n2. Check the running workflow (task output file \u002F TaskList). If done: read its\n   result AND its journal if the result looks off; persist artifacts; verify\n   commits landed per the commit policy; advance the state machine (consult\n   `bd ready` for dependency order when picking the next epic); launch the\n   next stage workflow.\n3. Heartbeat + lease: run `${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Flease.sh refresh`\n   EVERY wake, state changed or not (quiet multi-tick workflows must not\n   starve the lease). Exit 5 or a `\u003Cslug>-lease-lost` marker: post a handoff\n   note, downgrade to read-only, do NOT re-arm. The dashboard republishes\n   itself on every state write (hook-owned); a pending-publish marker means\n   degraded publishing, mention it and keep going.\n4. Update beads (recipes in `references\u002Fstate-and-beads.md`), state.json, and\n   the token ledger (accounting rules in the same file — fail CLOSED if spend\n   is unmeasurable under a --budget). state.json writes go through the\n   Write\u002FEdit TOOL only (Non-negotiables). On a phase flip also stamp\n   `phase_since`; after SHIP \u002F HUMAN_GATE \u002F DONE write the phase explainer\n   (\u002Fexplain-to-me) and publish it via\n   `${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Fexplainer-publish.sh` (the Stop gate blocks\n   your stop until its receipt exists).\n5. Ask `programme_policy.py next-action \u003Cstate>` which lane owns the next\n   action. Confirmed defects own mutation, shipped epics queue regression, and\n   perpetual empty backlogs refresh Signal then queue research. Re-arm: ScheduleWakeup ~600s,\n   reason = current phase, prompt = the DRIVER\n   RE-ENTRY PROMPT verbatim (below). Honour STOP RULES first.\n\n### Driver re-entry prompt (the exact string for every ScheduleWakeup)\n\n```\n\u002Fgodmode run — programme charter: .agents\u002Fgoals\u002F\u003Cslug>-charter.md, state:\n.agents\u002Fscratch\u002F\u003Cslug>-state.json. Follow the charter's LOOP PROTOCOL exactly.\n```\n\nNever improvise a different prompt; this string is what makes every future\nwake re-enter this skill and re-read the constitution.\n\n## STOP RULES (halt the loop + page the human)\n\nPage = PushNotification + a red banner note on the dashboard, then stop\nre-arming. Triggers:\n\n- Budget or deadline is exhausted, or budget spend is unmeasurable under a cap.\n- Security, production-safety, lost lease, or missing required authority.\n- A declared production deployment breaches its health policy after rollback.\n\nRepeated workflow or validation failure is not a global stop. Use bounded\nretries, create a quarantined incident, release the mutation lane, and keep\nindependent research and planning moving. Any confirmed defect preempts the\nmutation lane until repaired and fully re-verified.\n\n## Model policy (defaults; charter may override any line)\n\n| Role | Default | Fallback |\n|---|---|---|\n| BRAIN | host primary plus mandatory creative peer | availability route from Creative quorum |\n| BUILD | opus + codex pair (`codex:codex-rescue` agentType), disagreements surfaced not silently resolved | charter override |\n| TEST \u002F VALIDATE | sonnet | charter override |\n| SCAFFOLD \u002F mechanical | sonnet | charter override |\n\n## VERIFY (≠ validate — the point of this skill)\n\nAuto-detect the epic's surface from touched files + project manifest, then run\nthe mandated lane from `references\u002Fverify-lanes.md`:\nWeb UI → real-browser drive · TUI → tmux+VHS frame-truth · API → real requests\n+ side-effect asserts · library → unit+property(+mutation).\nCross-cutting: mock ONLY the human at the input boundary; READ the artefact\n(never blank-check); evidence uploaded to here.now and linked from the\ndashboard Evidence tab AND the PR body. In perpetual mode, every shipped epic\nruns the entire cumulative regression spine before auto-merge. Regression uses\nan isolated worktree. A confirmed failure pauses unrelated mutation, creates a\nrepair incident, then resumes paused work from a rebased baseline.\n\n## References (load when you reach that step)\n\n- `references\u002Fcharter-template.md` — the constitution scaffold (INIT step 2)\n- `references\u002Fstage-workflows.md` — Court\u002FRoadmap\u002FPlan\u002FExecute Workflow script templates (each stage launch)\n- `references\u002Fverify-lanes.md` — surface detection + the four proof playbooks (Execute's VERIFY step)\n- `references\u002Fstate-and-beads.md` — state.json schema, beads JSONL plumbing, resume procedure\n- `references\u002Flessons.md` — hard-won operational catches; consult BEFORE authoring any stage workflow and whenever a stage result looks wrong\n\n## Scripts\n\n- `scripts\u002Fbeads_remote.sh` (skill-local) — close\u002Fannotate beads on origin\u002Fmain\n  via git plumbing without touching any checkout (safe under concurrent worktrees)\n- Plugin scripts at `${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002F`:\n  `render_dashboard.py` (deterministic dashboard) · `on-state-write.sh`\n  (PostToolUse pipeline) · `explainer-publish.sh` (receipt writer, the ONLY\n  sanctioned explainer publish path) · `explainer-gate.sh` (Stop gate) ·\n  `sync.sh` (pull|push|adopt|discover) · `lease.sh` (claim|refresh|check|release)\n  · `sidecar_remote.sh` (refs\u002Fgodmode\u002F\u003Cslug> plumbing) · `staleness-note.sh`\n  (Copilot nudge)\n\n## Non-negotiables\n\n- Commit policy: atomic single-concern commits, conventional, named paths only\n  (never `git add -A`), no AI attribution, never commit charters\u002Fstate\u002F\n  dashboards\u002Fscratch\u002Fenv files, EXCEPT the sidecar mirror on\n  `refs\u002Fgodmode\u002F\u003Cslug>`, which hooks maintain via git plumbing. Sign only when\n  the gpg cache is warm; never spawn GUI pinentry headless.\n- state.json is written ONLY via the Write\u002FEdit tool, never Bash\n  redirection\u002Fjq\u002Fmv: the status and heartbeat hooks key on the PostToolUse\n  event (belt: the Stop-hook heartbeat entry; a Bash-written state file means\n  a silently stale dashboard and a starving lease).\n- Plugin version policy: ANY change under `plugins\u002Fgodmode\u002F**` bumps at least\n  the patch version, all three provider manifests together.\n- The dashboard is not optional and never goes stale past one tick.\n- Stale-work check: before fixing any backlog item, verify it isn't already\n  fixed in-tree (`git log\u002Fshow` + live probe). Close-with-citation beats re-fix.\n- Build gates run UNPIPED to a log file, backgrounded, gated on `EXIT=0`.\n","# godmode — the autonomous product factory\n\nA runtime orchestrator that OWNS the loop. It composes existing machinery —\n`\u002Floop` (driver), `Workflow` (staged ultracode), beads (state), `\u002Fhere-now`\n(dashboard hosting), `tmux-verify`\u002Fbrowser-harness\u002FPlaywright (verify lanes) —\ninto one self-driving programme. You are the driver each tick; these files are\nyour constitution and playbooks.\n\n## Entry modes\n\n| Invocation | Action |\n|---|---|\n| `\u002Fgodmode \u003Cnorth-star> --mode finite\\|perpetual [--approval none\\|roadmap] [--no-court] [--budget \u003Ctokens>] [--deadline \u003CISO>]` (alias: `\u002Fgodmode init \u003Cnorth-star> ...`) | INIT: generate charter + state + dashboard + beads, then begin finite delivery or perpetual evolution |\n| `\u002Fgodmode run [--take-over]` | Resume\u002Fcontinue the loop from state (any session, incl. post-crash). Claims the driver lease first: refuses while another session's lease is fresh, auto-claims a stale one, `--take-over` forces after an AskUserQuestion confirm. On a fresh machine: `sync.sh discover`, pick the slug, claim, then `sync.sh adopt \u003Cslug>` reconstructs local state |\n| `\u002Fgodmode status` | Standup: read state + beads + dashboard + lease, report, change nothing. NEVER adopts (observer-safe: no scratch state means hooks stay inert). Preflight prints \"status publishing DISABLED: missing \u003CX>\" when here-now\u002Fcredentials are absent |\n| `\u002Fgodmode pause` | Stop re-arming the loop; state stays resumable; release the lease (`lease.sh release`) |\n\n## Provider parity\n\n| Capability | Claude | Codex | Copilot |\n|---|---|---|---|\n| init \u002F run (drive the loop) | native scheduler | native Codex automation required | external scheduler adapter required |\n| status \u002F discover | yes | yes | yes |\n| dashboard publish + sidecar sync + lease guard | all hooks | hooks.json (shared format) | sessionStart pull + staleness nudge |\n\nNever claim a provider is driving when its scheduler or peer-model adapter is\nmissing. Record the capability result in state and defer that lane visibly.\n\n## The three-layer machine\n\n```\nLAYER 1  CHARTER   .agents\u002Fgoals\u002F\u003Cslug>-charter.md   — constitution, re-read EVERY tick\nLAYER 2  DRIVER    \u002Floop dynamic mode                — ScheduleWakeup ~600s, same prompt re-entered\nLAYER 3  WORKFLOWS one ultracode Workflow per stage  — resumable via resumeFromRunId\nSTATE    beads (epic\u002Ffeature nodes) + .agents\u002Fscratch\u002F\u003Cslug>-state.json (working\n         copy) + hook-maintained sidecar mirror {state,charter,lease} on the\n         dedicated ref refs\u002Fgodmode\u002F\u003Cslug> (cross-machine; GODMODE_SYNC=local opts out)\n```\n\n## Pipeline (stage names are the primary vocabulary; W-numbers are aliases)\n\n```\nSIGNAL                   TTL    ONE model gathers CITED external evidence: comparable\n        │                       products, user-reported gaps, constraining standards.\n        │                       Reads the append-only ledger first and returns only what is\n        │                       NEW; parked rows keep their reason and are never re-proposed.\n        │                       \u002Fresearch if available, else WebSearch+WebFetch. Charter-gated;\n        │                       unavailable degrades to repo-only discovery, never a failure.\n        ▼\nDISCOVER                 once   BRAIN brainstorms the nirvana landscape from the\n        │                       north-star + a repo scan + the SIGNAL rows + any existing\n        │                       backlog → writes the feature REGISTRY\n        │                       (.agents\u002Fplans\u002F\u003Cslug>-registry.md), provenance per row\n        │                       (repo | signal | user). User lists appended, not replaced.\n        ▼\nFEASIBILITY COURT (W0)   once   every REGISTRY idea → verdict: feasible | downgrade | park+blocker\n        │                       skipped only by explicit --no-court (registry taken verbatim)\n        ▼\nROADMAP (W1)             once   cluster survivors → ordered epics + parallel-pair candidates,\n        │                       seed beads (one per epic + per feature)\n        ▼\n[OPTIONAL ROADMAP GATE]         only when approval_policy is roadmap\n        ▼\nper epic, serial on stacked branches:\n   PLAN (W2)      planner → adversarial review → revise → VERIFY THE REVISE EDITED THE\n        │         PLAN FILES (mtime\u002Fcontent) — see references\u002Flessons.md\n        ▼\n   EXECUTE (W3)   build (model pair per workstream) → pair review → adversarial epic\n        │         review (+fix) → VERIFY (surface lane, references\u002Fverify-lanes.md)\n        ▼         → bounded fix loop (≤2) → build gate (unpiped!)\n   SHIP           stacked PR (labelled for review) → close epic+feature beads with\n        │         evidence notes → dashboard update\n        ▼\nPERPETUAL         after every ship: full regression + refresh SIGNAL if stale +\n                  next discovery\u002Fplan. Backlog-dry refreshes SIGNAL first and\n                  becomes adaptive evidence research, never DONE\nTERMINATION       finite backlog-dry | --budget exhausted | --deadline reached |\n                  security, production-safety, authority, or lease loss\n```\n\nPerpetual progression: after each epic ships, start cumulative regression in\nan isolated lane. At the same time, the completeness critic re-enters Discover\nand creates evidence-backed candidates for the next generation, refreshing the\nexternal Signal first when it is stale, so a dry backlog is replenished from\nnew outside evidence rather than from re-reading the same repo. The refresh\nappends to the signal ledger and dedupes against every row ever recorded,\nparked ones included, so the programme does not pay to rediscover a dead end it\nalready rejected. A confirmed\ndefect preempts the single mutation lane. Discovery and planning continue in\nparallel. An empty Court result enters adaptive-backoff research, never DONE.\n\n## Creative quorum\n\nEvery creative decision has two independent model perspectives: Discover,\nFeasibility Court, Roadmap, completeness critic, incident root-cause analysis,\nand strategic reprioritisation. Signal is not one of them: gathering cited\nexternal evidence is retrieval, so it runs on one model and is judged by its\ncitations, not by quorum. Resolve the route before launching work with:\n\n```bash\n${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Fprogramme_policy.py creative-route \u003Cavailability.json>\n```\n\n- Claude primary pairs a Claude creative model with `codex:codex-rescue`.\n- Codex primary pairs a Codex model with Claude Fable.\n- Copilot primary uses two distinct Copilot models, then enriches with Claude\n  or Codex when available.\n- Fallback uses any two distinct available models. One model means creative\n  work is deferred, while regression and grounded work continue.\n- Persist both proposals, model identifiers, evidence, disagreement, and the\n  synthesis. A third model synthesises when available; otherwise use the\n  evidence rubric and preserve dissent.\n\nParallel epics: allowed ONLY if blessed AND file-sets provably disjoint AND\nworktrees available; otherwise serialise (single worktree = single checkout).\n\n## INIT procedure\n\n1. Parse the north-star + flags. Derive `\u003Cslug>`.\n2. Generate the charter from `references\u002Fcharter-template.md` — fill outcome,\n   model policy (resolve fable toggle), verify doctrine, termination, stop\n   rules, dashboard slug. Write `.agents\u002Fgoals\u002F\u003Cslug>-charter.md`.\n3. Write `.agents\u002Fscratch\u002F\u003Cslug>-state.json` (schema:\n   `references\u002Fstate-and-beads.md`) with `driver_session_id: null` and a\n   `phase_since` timestamp. You cannot know your own session_id: the\n   PostToolUse hook fills `driver_session_id` from the hook event on this very\n   write (first-writer-wins), which is why the write must precede the claim.\n   Then claim the driver lease:\n   `bash ${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Flease.sh claim \u003Crepo> \u003Cslug>`.\n4. Dashboard publishing is HOOK-OWNED: every state.json write triggers\n   render + publish + sidecar push (PostToolUse). Your duties: keep\n   state.json truthful, write `current_note`, pick a stable\n   `dashboard_slug`. Publish failure never blocks (pending marker +\n   dashboard banner; hooks retry on the next state write).\n5. Run DISCOVER (template: `references\u002Fstage-workflows.md`) → writes the\n   registry. Then, unless `--no-court`, the Feasibility Court workflow, then\n   Roadmap. With `--no-court`: registry taken verbatim, straight to Roadmap.\n6. If `approval_policy` is `roadmap`, present the roadmap, set\n   `human_gate: \"pending\"`, and wait. Otherwise set `human_gate: \"none\"` and\n   start the loop immediately. Perpetual mode normally uses `none`.\n\n## LOOP protocol (every wake — this is the driver contract)\n\n1. Read the charter + state.json. They outrank your memory of last tick. Run\n   `programme_policy.py validate-state \u003Cstate>` before any side effect. If an\n   optional `human_gate` is pending, do not re-arm.\n2. Check the running workflow (task output file \u002F TaskList). If done: read its\n   result AND its journal if the result looks off; persist artifacts; verify\n   commits landed per the commit policy; advance the state machine (consult\n   `bd ready` for dependency order when picking the next epic); launch the\n   next stage workflow.\n3. Heartbeat + lease: run `${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Flease.sh refresh`\n   EVERY wake, state changed or not (quiet multi-tick workflows must not\n   starve the lease). Exit 5 or a `\u003Cslug>-lease-lost` marker: post a handoff\n   note, downgrade to read-only, do NOT re-arm. The dashboard republishes\n   itself on every state write (hook-owned); a pending-publish marker means\n   degraded publishing, mention it and keep going.\n4. Update beads (recipes in `references\u002Fstate-and-beads.md`), state.json, and\n   the token ledger (accounting rules in the same file — fail CLOSED if spend\n   is unmeasurable under a --budget). state.json writes go through the\n   Write\u002FEdit TOOL only (Non-negotiables). On a phase flip also stamp\n   `phase_since`; after SHIP \u002F HUMAN_GATE \u002F DONE write the phase explainer\n   (\u002Fexplain-to-me) and publish it via\n   `${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002Fexplainer-publish.sh` (the Stop gate blocks\n   your stop until its receipt exists).\n5. Ask `programme_policy.py next-action \u003Cstate>` which lane owns the next\n   action. Confirmed defects own mutation, shipped epics queue regression, and\n   perpetual empty backlogs refresh Signal then queue research. Re-arm: ScheduleWakeup ~600s,\n   reason = current phase, prompt = the DRIVER\n   RE-ENTRY PROMPT verbatim (below). Honour STOP RULES first.\n\n### Driver re-entry prompt (the exact string for every ScheduleWakeup)\n\n```\n\u002Fgodmode run — programme charter: .agents\u002Fgoals\u002F\u003Cslug>-charter.md, state:\n.agents\u002Fscratch\u002F\u003Cslug>-state.json. Follow the charter's LOOP PROTOCOL exactly.\n```\n\nNever improvise a different prompt; this string is what makes every future\nwake re-enter this skill and re-read the constitution.\n\n## STOP RULES (halt the loop + page the human)\n\nPage = PushNotification + a red banner note on the dashboard, then stop\nre-arming. Triggers:\n\n- Budget or deadline is exhausted, or budget spend is unmeasurable under a cap.\n- Security, production-safety, lost lease, or missing required authority.\n- A declared production deployment breaches its health policy after rollback.\n\nRepeated workflow or validation failure is not a global stop. Use bounded\nretries, create a quarantined incident, release the mutation lane, and keep\nindependent research and planning moving. Any confirmed defect preempts the\nmutation lane until repaired and fully re-verified.\n\n## Model policy (defaults; charter may override any line)\n\n| Role | Default | Fallback |\n|---|---|---|\n| BRAIN | host primary plus mandatory creative peer | availability route from Creative quorum |\n| BUILD | opus + codex pair (`codex:codex-rescue` agentType), disagreements surfaced not silently resolved | charter override |\n| TEST \u002F VALIDATE | sonnet | charter override |\n| SCAFFOLD \u002F mechanical | sonnet | charter override |\n\n## VERIFY (≠ validate — the point of this skill)\n\nAuto-detect the epic's surface from touched files + project manifest, then run\nthe mandated lane from `references\u002Fverify-lanes.md`:\nWeb UI → real-browser drive · TUI → tmux+VHS frame-truth · API → real requests\n+ side-effect asserts · library → unit+property(+mutation).\nCross-cutting: mock ONLY the human at the input boundary; READ the artefact\n(never blank-check); evidence uploaded to here.now and linked from the\ndashboard Evidence tab AND the PR body. In perpetual mode, every shipped epic\nruns the entire cumulative regression spine before auto-merge. Regression uses\nan isolated worktree. A confirmed failure pauses unrelated mutation, creates a\nrepair incident, then resumes paused work from a rebased baseline.\n\n## References (load when you reach that step)\n\n- `references\u002Fcharter-template.md` — the constitution scaffold (INIT step 2)\n- `references\u002Fstage-workflows.md` — Court\u002FRoadmap\u002FPlan\u002FExecute Workflow script templates (each stage launch)\n- `references\u002Fverify-lanes.md` — surface detection + the four proof playbooks (Execute's VERIFY step)\n- `references\u002Fstate-and-beads.md` — state.json schema, beads JSONL plumbing, resume procedure\n- `references\u002Flessons.md` — hard-won operational catches; consult BEFORE authoring any stage workflow and whenever a stage result looks wrong\n\n## Scripts\n\n- `scripts\u002Fbeads_remote.sh` (skill-local) — close\u002Fannotate beads on origin\u002Fmain\n  via git plumbing without touching any checkout (safe under concurrent worktrees)\n- Plugin scripts at `${CLAUDE_PLUGIN_ROOT}\u002Fscripts\u002F`:\n  `render_dashboard.py` (deterministic dashboard) · `on-state-write.sh`\n  (PostToolUse pipeline) · `explainer-publish.sh` (receipt writer, the ONLY\n  sanctioned explainer publish path) · `explainer-gate.sh` (Stop gate) ·\n  `sync.sh` (pull|push|adopt|discover) · `lease.sh` (claim|refresh|check|release)\n  · `sidecar_remote.sh` (refs\u002Fgodmode\u002F\u003Cslug> plumbing) · `staleness-note.sh`\n  (Copilot nudge)\n\n## Non-negotiables\n\n- Commit policy: atomic single-concern commits, conventional, named paths only\n  (never `git add -A`), no AI attribution, never commit charters\u002Fstate\u002F\n  dashboards\u002Fscratch\u002Fenv files, EXCEPT the sidecar mirror on\n  `refs\u002Fgodmode\u002F\u003Cslug>`, which hooks maintain via git plumbing. Sign only when\n  the gpg cache is warm; never spawn GUI pinentry headless.\n- state.json is written ONLY via the Write\u002FEdit tool, never Bash\n  redirection\u002Fjq\u002Fmv: the status and heartbeat hooks key on the PostToolUse\n  event (belt: the Stop-hook heartbeat entry; a Bash-written state file means\n  a silently stale dashboard and a starving lease).\n- Plugin version policy: ANY change under `plugins\u002Fgodmode\u002F**` bumps at least\n  the patch version, all three provider manifests together.\n- The dashboard is not optional and never goes stale past one tick.\n- Stale-work check: before fixing any backlog item, verify it isn't already\n  fixed in-tree (`git log\u002Fshow` + live probe). Close-with-citation beats re-fix.\n- Build gates run UNPIPED to a log file, backgrounded, gated on `EXIT=0`.\n",{"name":5,"description":7},"main","9cf32d1e636db09530158af14ba301bfa5d401e9","plugins\u002Fgodmode\u002Fskills\u002Fgodmode","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez","stevengonsalvez.com","2026-11-12 00:50:59",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add stevengonsalvez\u002Fainb-toolkit --skill godmode",{"tool":37,"command":38},"bunx","bunx skills add stevengonsalvez\u002Fainb-toolkit --skill godmode",{"tool":40,"command":41},"pnpm","pnpm dlx skills add stevengonsalvez\u002Fainb-toolkit --skill godmode","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002FSKILL.md",true,[45,51,57,61,65,69,73],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"assets\u002Fprogramme-dashboard.html","ASSET","资产文件",4764,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Fassets\u002Fprogramme-dashboard.html",{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fcharter-template.md","REFERENCE","参考文档",5981,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Freferences\u002Fcharter-template.md",{"path":58,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Flessons.md",5275,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Freferences\u002Flessons.md",{"path":62,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Fstage-workflows.md",12506,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Freferences\u002Fstage-workflows.md",{"path":66,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":67,"sourceHtmlUrl":68},"references\u002Fstate-and-beads.md",8418,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Freferences\u002Fstate-and-beads.md",{"path":70,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":71,"sourceHtmlUrl":72},"references\u002Fverify-lanes.md",4876,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Freferences\u002Fverify-lanes.md",{"path":74,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":77,"sourceHtmlUrl":78},"scripts\u002Fbeads_remote.sh","SCRIPT","脚本",3191,"https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fplugins\u002Fgodmode\u002Fskills\u002Fgodmode\u002Fscripts\u002Fbeads_remote.sh",[80,87,94,100,107,114],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":19},"01M75WK9B8DKVJBJJ4CX2QMJMX","well-formed","Fixture skill that satisfies every mechanical rule, used to prove the conformance check passes a clean tree instead of failing everything.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fclean\u002Fwell-formed\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":19},"01M75WK6ZGDXJJN8879K49QEYG","underscore-tools","Fixture skill that spells the tool allowlist key with an underscore instead of the hyphen the spec requires.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Funderscore-tools\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":19},"01M75WK524MVT2H9E27ZCEA3HR","no-frontmatter","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Fno-frontmatter\u002FSKILL.md",[],[],{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":19},"01M75WJWZBMKR8R60PG3T3MCCC","long-description","Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Flong-description\u002FSKILL.md",[],[],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":19},"01M75WJVRH3J22WKED7E5D01S8","some-other-name","Fixture skill whose frontmatter name does not match its parent directory name.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Fdir-mismatch\u002FSKILL.md",[],[],{"skillNo":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":19},"01M75WJTC7H5FRP6S129VBSMZC","bad_Name","Fixture skill whose name uses an underscore and a capital letter, so it violates the agentskills.io name pattern.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Fbad_Name\u002FSKILL.md",[],[]]