dev-verify
Verify implemented changes before claiming they are complete, fixed, passing, or ready for delivery. Match checks to requested behavior and risk, inspect current evidence, and report limits. Use for completion checks, not general factual answers or a replacement for code review.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-11
- Stars
- ★ 1.6k
- Forks
- 468
01 前言 Front Matter
- name
- dev-verify
- description
- Verify implemented changes before claiming they are complete, fixed, passing, or ready for delivery. Match checks to requested behavior and risk, inspect current evidence, and report limits. Use for completion checks, not general factual answers or a replacement for code review.
02 资源文件 3
- 参考文档examples.md2.3 KB
- 参考文档references/dev-baseline.md3.5 KB
- 参考文档references/project-verification.md5.3 KB
03 SKILL.md 正文
Dev Verify
Make completion claims no broader than the evidence. This workflow verifies behavior and artifacts; it does not authorize commits, publishing, merge, or cleanup.
Load baseline
Read references/dev-baseline.md before execution. Resolve paths relative to this skill directory.
Choose evidence
Identify what the user expects to be true and the smallest sufficient proof. Use project commands and conventions; inspect their actual scope before relying on their names.
| Claim | Relevant proof |
|---|---|
| Behavior implemented | Tests or direct checks of observable acceptance criteria |
| Bug fixed | Original reproduction or faithful regression passes |
| Regression detects the defect | Observed failure on original behavior, or isolated mutation proof |
| Refactor preserves behavior | Relevant characterization tests pass before and after |
| Build/lint succeeds | Corresponding command completes successfully |
| UI works | Relevant rendered states and interactions, with viewport/device scope stated |
| Skill/document updated | Structure/link checks plus review of changed instructions/content |
Use focused checks for local changes. Broaden for shared contracts, state transitions, persistence, integrations, security, or project-required gates. Do not add tests that merely mirror reversible low-impact edits. A parser or keyword check validates structure, not instruction quality or system behavior.
Run and inspect
- Use fresh evidence from the final relevant file state. Evidence already produced and inspected during the task remains usable if the tested inputs, dependencies, and environment have not materially changed; do not rerun solely because a new message or handoff occurred.
- Record exact commands, exit status, and meaningful output. Inspect failures, skipped tests, and warnings that affect the claim. A command being started is not evidence of completion.
- Keep verification read-only with respect to source and the Git index. Run tools that generate artifacts in an isolated output location where needed. Do not fix code in an independent verification lane unless explicitly reassigned.
- A worker's summary is not enough: inspect its output and relevant artifacts, or independently reproduce the critical check. Scale repeated verification to remaining uncertainty.
- When a check fails, distinguish a regression from a pre-existing/environmental failure using evidence. Continue authorized corrective work in the appropriate lane, then rerun affected checks. Do not quietly waive a required failure.
Unavailable hardware, credentials, production access, or test tooling limits the claim. Complete all feasible checks and report missing proof without presenting a local substitute as equivalent.
Reuse project verification
Prefer existing project scripts, tests, and CI entrypoints. For recurring setup or behavior checks, read references/project-verification.md to reuse or maintain a concise project recipe. A missing recipe does not block verification or authorize creating documentation. Create or repair a recipe or harness only in the main execution lane with existing task authorization; an independent verifier keeps source and index unchanged. A newly recorded recipe is executable only after its stated path has actually run; otherwise label it draft.
SDD alignment and completion
Use the current request and relevant existing SDD artifacts as acceptance criteria. If present, .claude/artifacts/{designs,plans,fixes}/ may contain the source contract; select by actual task/file relevance, not the number of artifacts. Later user corrections override stale artifacts.
For substantial work, map requirements to evidence and list missing coverage. Small changes need only a short verification statement. Report what changed, which checks passed or failed, and material residual limits; do not force a fixed checklist into every response.
Verification does not substitute for independent code review. A READY review is not proof that tests ran, and passing tests do not prove every review concern resolved. Use dev-code-review when available and required, or perform the equivalent independent review.
Multi-Agent Profile
Recommended agent_type: worker
When delegation is permitted, give the verifier the current task contract, change scope, and raw artifacts. The verifier evaluates evidence independently and reports claims checked, commands/results, and limitations without changing source or staging files. Follow docs/multi-agent-policy.md if this repository is available; these rules also apply to standalone installations.
04 同仓库更多 Skills
deja-search
deja-vu memory — search the user's past AI coding sessions with the deja CLI. Holds what they already settled and never wrote into the repo (naming rules, banned or already-tried tools, required checks before merge, preferred values, who owns what). Use before your first edit in a task, before you call a change done or ready to merge, before you recommend a tool, library, design or reviewer, before re-debugging an error that may already be solved, when they say 'didn't we fix this before', 'what did we decide about X' or 'I already have that', and before telling them something on this machine does not exist.
deja-reflect
End-of-session reflection and memory curation with deja-vu. Review what was decided, what commands broke and what repaired them, and record durable project rules or decisions with deja remember so the next session starts with them.
revid
Make, edit, export and post AI videos with Revid (revid.ai) through the Revid MCP server. Use whenever the user wants a short-form or long-form video made (from an idea, a script, a URL, an article, a PDF, a tweet, a product page, their own clips, or a song), wants to find what is going viral on TikTok, YouTube, Instagram or X before making something, wants to edit or re-export an existing Revid project, generate a standalone AI image or clip, make a series episode, or publish and schedule a finished video to connected social accounts. Covers setup (hosted MCP with OAuth, or an API key), the 36 tools, credits, polling and the rules for publishing. Routes to the more specific revid-* skills when one fits.
review-study
Review a QuestionPunk survey before launch for wording, answer choices, branching, and missing settings. Use for check my survey or review this draft before publishing.
manage-study
Find QuestionPunk studies, check status, publish an explicitly requested draft, or retrieve its participant link. Use for find my survey, publish this study, or how many people responded.
create-study
Create a saved QuestionPunk survey draft from a research brief or edit specified questions. Use for requests such as make a customer feedback survey or shorten my onboarding study.