verify
Verification and auditing. Validate acceptance scenarios, run multi-perspective audits, check type-check/build evidence, and gate phase/sprint completion.
- License 许可
- MIT
- 最近更新
- 2026-11-10
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- verify
- description
- Verification and auditing. Validate acceptance scenarios, run multi-perspective audits, check type-check/build evidence, and gate phase/sprint completion.
02 资源文件 7
- 参考文档audit.md2.5 KB
- 参考文档complete.md3.8 KB
- 参考文档feature.md1.4 KB
- 参考文档file.md1.4 KB
- 参考文档phase.md1.3 KB
- 参考文档recent.md1.3 KB
- 参考文档sprint.md1.9 KB
03 SKILL.md 正文
Verification
Prove work does what the SPEC said, then gate the phase/sprint. /verify:complete is the exit gate /work:done calls — it PRODUCES the evidence the stop hooks look for (checks/03-scenario-evidence, 04-tdd-cycle, 05-progress-evidence) when you signal [PHASE_COMPLETE]. The other commands are lighter, on-demand audits over a chosen surface.
Audit validates code looks correct. Scenario evidence validates code works. A phase needs both. Walk the SPEC line-by-line — never aggregate a "looks done" verdict.
Commands
| Command | Purpose | Workflow |
|---|---|---|
/verify:complete |
Full phase validation — the exit gate | complete.md |
/verify:phase |
Light audit of the current phase | phase.md |
/verify:sprint |
Audit an entire sprint before merge | sprint.md |
/verify:file |
Audit specific file(s) + their callers | file.md |
/verify:feature |
Audit a named feature across the codebase | feature.md |
/verify:audit |
Deploy a multi-perspective audit (the engine) | audit.md |
/verify:recent |
Audit the last few commits | recent.md |
Shared foundations
State chain (read in order — see CLAUDE.md § Session State Management): docs/CURRENT_WORK.md → context/WORKING.md → phase SPEC.md / SCENARIOS.md / PROGRESS.md.
Configured commands: reference the project's type-check and build commands abstractly — read type_check_command / build_command from .copier-answers.yml, or CLAUDE.md § Essential Commands. Never hardcode a stack's tooling.
Multi-perspective audit: every audit command deploys parallel Explore agents (see the patterns skill § Multi-Explorer Pattern) — one per perspective that applies to the changed surface, each returning a bounded (<2000 char) summary. Explorers cannot load skills; after they report, YOU invoke the project's matching best-practices skill(s) (.claude/skills/domain/) and cross-reference their findings. The perspective menu and report template live in audit.md.
Coherence rule: verify code state first (git log, file reads), then update docs/issues — never record from memory alone.
Friction: when verification exposes a bad skill/hook/workflow step, emit the matching [FRICTION:*] signal (see patterns skill § Friction Signals); the stop hook logs it.
/verify:complete
Exit gate for phase completion. Confirm every acceptance scenario has a recorded passing run, walk the SPEC line-by-line into a met/partial/not-met verdict table, run a multi-perspective audit, cross-reference best-practices skills, pass the configured type-check + build, update the state chain, and align docs ↔ issues. Any partial/not-met row blocks. Only on a clean pass do you check off PROGRESS.md and signal [PHASE_COMPLETE] — the moment the stop-hook evidence checks fire. Full workflow: complete.md.
/verify:phase
Lighter than complete — a targeted audit of the current phase's changed surface for pattern alignment and coherence, ending in a WORKING.md phase-transition marker. Use before moving on; use /verify:complete for official completion. Full workflow: phase.md.
/verify:sprint
Full validation before marking a sprint complete and merging. Aggregate scenario coverage across the sprint's phases, run a cross-phase audit, confirm each phase passed its exit gate, and block the merge on any unmet criteria. Full workflow: sprint.md.
/verify:file
Targeted audit of specific file(s). Read the target and its immediate dependencies, find its callers, deploy explorers scoped to the file's role, cross-reference the matching best-practices skill, and assess caller impact. Full workflow: file.md.
/verify:feature
Audit a named feature across its full implementation. Discover every participating file, deploy explorers scoped to them, check completeness across every layer the feature touches, and report gaps. Full workflow: feature.md.
/verify:audit
The multi-perspective audit engine the other commands call. Scope the surface, deploy the subset of perspectives that applies, consolidate bounded explorer findings, and cross-reference best-practices skills. Full workflow: audit.md.
/verify:recent
Quick audit of the last few commits before a PR or handoff. Read the recent diffs, categorize the changed files, deploy the applicable perspectives, and report CLEAN / ISSUES. Full workflow: recent.md.
04 同仓库更多 Skills
ponytail
Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports intensity levels: lite, full (default), ultra. Use on ANY coding task: writing, adding, refactoring, fixing, reviewing, or designing code, and choosing libraries or dependencies. Also use whenever the user says "ponytail", "be lazy", "lazy mode", "simplest solution", "minimal solution", "yagni", "do less", or "shortest path", or complains about over-engineering, bloat, boilerplate, or unnecessary dependencies. Do NOT use for non-coding requests (general knowledge, prose, translation, summaries, recipes).
workflow/skill
Patterns for adding or editing a skill — frontmatter contract, router + sub-file split, line budgets.
workflow
Meta-skill for editing the framework itself — skills, hooks, commands, rules, CLAUDE.md. Use before changing anything under .claude/ or the routing docs, so the change keeps the system's shape (lazy-load, forge markers, hook auto-discovery, harness).
work
Work lifecycle orchestration: start, continue, pause, handoff, done, plus epic/sprint planning. Context gathering + goal-focused state-chain transitions.
team-lead
Coordinate an Agent Team for parallelizable work — implementation, audit, research, or testing. Teaches the mechanical constraints (WAIT pattern, file ownership, model tiers) and lets you design the team. For orchestration_tier=cluster / multi-agent runs.
scope
Plan and design work before building. Invoke when starting an epic, scoping a sprint, or planning any Standard+ change: orient on what exists, equip the planning procedure + domain skills, then produce the SPEC/SCENARIOS/PROGRESS structure and converge it through an audit loop. Feeds /work:epic-plan and the ATDD entry gate.