Menü
Run a long-form goal with judge-verified completion.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Creates or updates repository specs, acceptance criteria, durable decisions, and documentation for new work. Use when adding a feature or defining an ambiguous slice, when making architectural decisions or changing contracts (game state, leaderboard, audio, settings), or when recording durable context (decisions, glossary, conventions) for future engineers and agents.
Audits the MEMORYBLOX codebase for dead TypeScript code, orphan files, and unused symbols, triages findings into live dead code or false positives, and optionally removes provably dead code with focused validation. Use when the user asks for a dead-code scan, unused-code audit, orphan-file scan, unused-symbol triage, or cleanup from analyzer findings.
Remove AI-generated code slop from the current branch diff. Use when the user asks to deslop, clean up AI slop, remove unnecessary comments, strip defensive cruft, or flatten needless nesting.
Runs a disciplined diagnosis loop for hard bugs and performance regressions. Reproduce, minimise, hypothesise, instrument, fix, and regression-test. Use when the user says "diagnose this", the bug is flaky or not yet reproducible, root cause is unknown after triage, or a performance regression needs measurement before fixing.
Extremely strict maintainability review hunting code-judo restructurings — abstraction quality, the 1k-line rule, and spaghetti growth. Use when asked to review code, do a code review, audit changes, check for issues, or review uncommitted/staged changes. Covers diff-first review of pending changes and broader codebase audit.
Transfer session state to a fresh agent as a minimal, lossless state-transfer document.
| name | goal |
| description | Run a long-form goal with judge-verified completion. |
| disable-model-invocation | true |
Declare a durable objective with a verifiable end state. The agent plans →
acts → observes → iterates across many turns until every end-state criterion
is confirmed — or you /goal pause or /goal clear.
.github/copilot-instructions.md — repo conventions, quality gates, anti-patterns.docs/ — domain-specific docs relevant to the goal.A goal has three required parts and one built-in safeguard:
| Part | What it is | Good | Bad |
|---|---|---|---|
| Goal | A clear, actionable objective in the imperative. | «Migrate all v1 API calls to v2.» | «Make the API better.» |
| End state | The checklist. Observable, binary, machine-checkable. | «rg "/api/v1" src/ returns 0 matches; npm test exits 0.» | «It should work.» |
| Constraints | Boundaries: scope, style, safety, non-goals. | «Only edit src/ and tests/. Conventional Commits.» | (none — agent will drift) |
The end state is the judge. The agent must verify each criterion independently before declaring done. This stops both premature completion («looks good to me!») and endless spinning («just one more refactor…»).
Type /goal followed by your objective. The agent will ask for the three
parts if any are missing, then run the loop until the end state is confirmed.
Minimal form:
/goal Migrate all deprecated v1 API calls to v2.
End state: rg "/api/v1" src/ tests/ returns 0 matches, npm test exits 0.
Constraints: only src/ and tests/, Conventional Commits.
Full form: use the layered template for complex, multi-hour goals.
| Command | What it does |
|---|---|
/goal status | Show progress, last verification results, remaining criteria. |
/goal pause | Freeze the loop. State is preserved in session memory. |
/goal resume | Continue from the last checkpoint. |
/goal clear | Stop and reset. Use when you need to refine the goal itself. |
After receiving a goal, the agent runs this cycle until the end state is met or a stop condition fires:
The agent never declares victory on its own say-so. Every «done» must be backed by verification output.
This skill inherits the repo's quality gates. When the end state includes test or lint criteria, use the canonical commands:
npm run quality:sanity # markdownlint + eslint + tsc + vitest
npm run quality:full # above + Playwright E2E
npm run test:coverage # coverage report (≥90% threshold per file)
Branch-per-goal convention:
git switch -c goal/<short-name>
Commit before launching so git reset --hard is always a clean escape.
The loop stops when:
/goal pause or /goal clear.BLOCKERS.md with what's left and stops — no guessing.The goal is done when every end-state criterion has been independently verified and the verification output is presented to the user. No criterion is «close enough.» Partial passes are failures.