원클릭으로
vibe
Controlled small-change lane for low-risk fixes, docs, tests, and narrowly scoped edits without running the full SDLC pipeline.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Controlled small-change lane for low-risk fixes, docs, tests, and narrowly scoped edits without running the full SDLC pipeline.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Autonomous build loop with Karpathy ratcheting, GAN evaluator, and session chaining. Iterates story groups until all features pass or stopping criteria met.
Change the behavior of existing code — story-driven by default, or --issue N for a GitHub bug fix. Test-first, full verification, code review.
Code generation quality principles — TDD, typing, error handling, logging, API integration, LLM integration.
Brownfield change route — take an existing-code feature request from intent to a reviewed PR, scaling from a single /change to an epic via /spec→/design→/auto. Linear-tracked, backed by a committed DeepWiki.
[Internal pipeline stage — run by /auto self-heal and /implement validation when lint/type error volume is high; invoke directly only as a power user.] Turn compiler/linter output into a sharded work queue; fix per package with optional adversarial review — no full monorepo suite mid-shard.
[Internal pipeline stage — run by /auto (follow with /gate); invoke directly only as a power user.] Generate production code and tests for a story group using agent teams for parallel execution.
| name | vibe |
| description | Controlled small-change lane for low-risk fixes, docs, tests, and narrowly scoped edits without running the full SDLC pipeline. |
| argument-hint | [brief-change-description] |
| context | fork |
Use /vibe for small, low-risk changes where the full BRD → spec → design → auto pipeline would be disproportionate.
This is not permission to free-code. It is a bounded engineering lane with explicit scope, targeted verification, and reviewer enforcement.
Ultracode tip: Leave ultracode off here (
/effort highor lower). This lane exists to keep small, low-risk changes proportionate — fanning out workflows would defeat its entire purpose.
/goal tip (optional unattended iteration): On Claude Code v2.1.139+ you can let
/goaldrive this single bounded session toward a verifiable condition — e.g./goal the targeted test passes and lint is clean, or stop after N turns. Always include the "or stop after N turns" safety clause, and phrase conditions so each turn must produce fresh evidence (re-run the test, show the exit code) to avoid false-positive completion./goal's evaluator (Haiku) only judges what is in the transcript — it does not run tools or read files — so the proof (test output, exit codes) must be printed in the conversation, not routed through subagents. That makes/goalsuitable for this small lane only. Do not use/goalinside/auto: it is single-session and would conflict with session chaining, the GAN evaluator, and sprint contracts./goaldoes not replace the evaluator/sprint-contract gate.
/vibe "fix typo in empty-state copy"
/vibe "add missing null guard in invoice total"
/vibe "update README install command"
Use controlled vibe coding only when all are true:
Escalate to /change, /refactor, /spec, or /auto when any are true:
upgrading-dependencies (patch bumps may stay in /vibe if the suite proves them; minor/major escalate).| Class | Examples | Required Verification |
|---|---|---|
| CV0 docs/config | docs typo, README command, comments, non-runtime config | git diff --check, relevant parser if any |
| CV1 test/tooling | add/adjust tests, lint config, CI command | targeted test/lint command |
| CV2 small behavior | null guard, validation message, small UI state, single bug | failing test or reproduction first, then targeted test |
CV2 is the highest class allowed in /vibe. Anything larger escalates.
State:
If classification is uncertain, use the clarify gate (.claude/skills/clarify/SKILL.md) — ask at most 3 questions, prefer recording assumptions over interrogating. If still uncertain, escalate.
Learned rules: before writing the micro-contract, read .claude/state/learned-rules.md. If it exists and is non-empty, inject its contents verbatim into your working context — a learned rule can affect scope (Step 1) as well as implementation, so read it before the contract is finalized.
Before editing, write 3-6 bullets:
## Micro-Contract
- Change:
- In scope:
- Out of scope:
- Verification:
- Rollback:
Append it to .claude/state/vibe-log.md. Create the file if missing.
Context-first (Iron Law) — when specs/brownfield/code-graph.json exists and is not a placeholder, run a micro pack before production source reads:
node .claude/scripts/context-pack.js --diff --budget 800 "<micro-contract change sentence>"
Read only read_next ranges (not whole god files). If confidence is low, one narrow search then re-pack or escalate. Prefer existing project patterns over new abstractions.
If specs/brownfield/change-strategy.md exists, read it before editing. If it marks the affected area as high-risk, stop and escalate out of /vibe.
If specs/brownfield/code-graph.json exists and the change edits an existing production symbol, run the coverage preflight (checking-coverage-before-change). An UNCOVERED verdict is a hard block for /vibe: silent breakage hides exactly here — escalate to /change or /refactor, where pinning/sprouting applies.
For CV2 behavior changes:
For CV0/CV1, skip TDD only when no runtime behavior changes.
Rules:
Run the narrowest useful checks:
git diff --checknode .claude/scripts/local-regression-gate.js (gap G16). Even a narrowly-scoped /vibe edit can silently break an earlier feature — this computes the diff's blast radius over code-graph.json and re-runs only the prior story-group(s) it could plausibly affect (plus any project-manifest.json#verification.golden_paths), instead of the whole accumulated e2e/ suite. A blocked verdict is a BLOCK — fix the regression before Step 7. The full, unabridged regression-suite-full check (gap G15) still runs at /gate as the final backstop before merge.If verification fails, fix within the micro-contract. If the fix expands beyond the eligibility rules, stop and escalate.
The existing hooks mark production-code writes for review and the Stop hook requires reviewer agents before the turn ends. Do not bypass this. If hooks are unavailable, manually invoke clean-code and security review for changed production files.
Report:
/change./vibe or /refactor task./change --issue N or /spec.