From autonomy-loop
The operating rhythm for the two-terminal autonomy loop — the gate, the honesty mandate, the 5-lens review, the research lane, baton discipline, and the accumulated "what almost broke + the rule that caught it" learnings. Read at the start of every loop session.
How this skill is triggered — by the user, by Claude, or both
Slash command
/autonomy-loop:autonomy-operateThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Read `autonomy.config.json` for this repo's knobs. Read `STATE.md` first every session.
Read autonomy.config.json for this repo's knobs. Read STATE.md first every session.
LOOP-STATE.md turn: is your role. Never infer the turn from the commit log.gate.test + the frozen invariant intact (gate.frozenInvariant) + gate.build + gate.lint on touched files. If gate.coverage is set, also run node ${CLAUDE_PLUGIN_ROOT}/hooks/coverage-ratchet.mjs after the coverage command: a coverage drop below the .autonomy-coverage.json floor is a RED gate (the third gate, the drift guard). If gate.patchTarget is greater than 0, also run node ${CLAUDE_PLUGIN_ROOT}/hooks/patch-coverage.mjs --threshold=<patchTarget> so the wave's own changed lines are tested (the fourth gate; the reviewer passes --base=<last-reviewed-sha>). All green or REVERT — never commit red. If gate.envFidelity is set, the LOCAL gate must match CI (an env-naked pass is not proof CI is green).honestyRule): no fabricated numbers; every rate carries N + CI or "building — N/30"; a capability with no real data ABSTAINS visibly — never faked, never silently dead.gateList): both terminals STOP and ask the human for risky/irreversible actions. Never self-authorize one — escalate to FOR-REVIEW.md, set turn: human.Two modes — IDEATE (research lane) and BUILD. ultrathink the plan. Smallest high-signal diff; PURE module + RED-before-green test for new logic; zero deletions on shared files; re-read fresh. Pre-empt the reviewer by answering the 5-lens rubric (Correctness · Honesty · Regression/frozen-drift · Security · UX) in your build note, with a "red-team the opposite" paragraph.
ultrathink every review. Spawn 5 critic subagents (math · honesty · regression+frozen-drift · security · UX); unanimous PASS required. Red-team-the-opposite before any PASS. Bite-check the regression test (reintroduce the bug → RED → restore). Frozen-drift/protected-path → Builder→Critic→Judge debate, PARK to FOR-REVIEW. Fix what you safely can; flag the rest with severity.
An empty bug/feature backlog is NOT a reason to set turn: human. When pending-for-builder is empty, the backlog is drained, and everything left is owner-gated, the builder runs the Research & Ideation lane (tasks/RESEARCH-LANE.md), it does not stand down:
tasks/IDEAS.md): problem · sourced evidence · proposed feature · anchor files · risk tier · acceptance · gate plan.FOR-REVIEW.md as an approval-menu entry the human picks from — and KEEP GOING (next non-gated wave, else the next research cycle). Gated items are parked, never stops.Set turn: human ONLY when truly blocked on all fronts at once (every remaining path needs a human answer AND research is genuinely dry) or the human explicitly pauses the loop — that should be rare. The reviewer mirrors this: never write "stand down" into pending-for-builder; point the builder at the next wave, else the research lane.
These generalized out of a real run — keep adding:
gate.envFidelity so it's automatic).grep -a before concluding a symbol is unused.npx claudepluginhub inferencegod/autonomy-loop --plugin autonomy-loopProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.