verify
Manifest verification runner. Spawns parallel verifiers for Global Invariants and Acceptance Criteria. Called by /do, not directly by users.
安装与使用
请根据来源仓库和你的 AI 工具说明安装此技能。先阅读技能文件和依赖要求，再添加到工作流。
查看来源 ↗使用指南技能说明
/verify - Manifest Verification Runner
Orchestrate verification of all criteria from a Manifest by spawning parallel verifiers. Report results grouped by type.
User request: $ARGUMENTS
Format: <manifest-file-path> <execution-log-path> [--scope=files]
If paths missing: Return error "Usage: /verify "
Principles
| Principle | Rule |
|---|---|
| Orchestrate, don't verify | Spawn agents to verify. You coordinate results, never run checks yourself. |
| ALL criteria, no exceptions | Every INV-G* and AC-. criterion MUST be verified. Skipping any criterion is a critical failure. |
| Maximize parallelism | Launch all verifiers in a SINGLE message with multiple Task tool calls. Never launch one at a time. |
| Globals are critical | Global Invariant failures mean task failure. Highlight prominently. |
| Actionable feedback | Pass through file:line, expected vs actual, fix hints. |
Verification Methods
| Type | What | Handler |
|---|---|---|
bash |
Shell commands (tests, lint, typecheck) | criteria-checker |
codebase |
Code pattern checks | criteria-checker |
subagent |
Specialized reviewer agents | Named agent (e.g., code-bugs-reviewer) |
research |
External info (API docs, dependencies) | criteria-checker |
manual |
Set aside for human verification | /escalate |
Note: criteria-checker handles any automated verification requiring commands, file analysis, reasoning, or web research.
Criterion Types
| Type | Pattern | Failure Impact |
|---|---|---|
| Global Invariant | INV-G{N} | Task fails |
| Acceptance Criteria | AC-{D}.{N} | Deliverable incomplete |
| Process Guidance | PG-{N} | Not verified (guidance only) |
Note: PG-* items guide HOW to work. Followed during /do, not checked by /verify.
Never Do
- Skip criteria (even "obvious" ones)
- Launch verifiers sequentially across multiple messages
- Verify criteria yourself instead of spawning agents
Outcome Handling
| Condition | Action |
|---|---|
| Any Global Invariant failed | Return all failures, globals highlighted |
| Any AC failed | Return failures grouped by deliverable |
| All automated pass, manual exists | Return manual criteria, hint to call /escalate |
| All pass | Call /done |
Output Format
Report verification results grouped by Global Invariants first, then by Deliverable.
On failure - Show for each failed criterion:
- Criterion ID and description
- Verification method
- Failure details: location, expected vs actual, fix hint
On success with manual - List manual criteria with how-to-verify from manifest, suggest /escalate.
On full success - Call /done.
使用示例
查看来源仓库中的使用示例与配置说明。
查看来源 ↗来源历史
- 收录时间
- 2026-01-20
- 索引更新时间
- 2026-02-05
- Branch
- main
收录与同步日期属于本站记录。源码修改历史请查看来源仓库。
GitHub ↗chat Comments (0)
Sign in to join the discussion and leave a comment.
No comments yet. Be the first to share your thoughts!