一键导入
verify
Verify phase deliverables match acceptance criteria
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Verify phase deliverables match acceptance criteria
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Initialize a Megazord project with configuration and planning structure
Systematic four-phase debugging (reproduce, isolate, root-cause, fix)
Socratic brainstorming to explore approaches before implementation
Execute the current phase plan (subagent or Agent Teams mode)
Show all available Megazord skills with descriptions and usage examples
Complete milestone lifecycle -- audit, archive, deferred items, and next version preparation
| name | verify |
| description | Verify phase deliverables match acceptance criteria |
| disable-model-invocation | false |
Verify that a phase achieved its GOAL by performing goal-backward verification against ROADMAP.md success criteria and PLAN.md must_haves. Spawns a verifier subagent via the Task tool that checks truths, artifacts (3-level), key links, and requirement coverage against the actual codebase. Produces VERIFICATION.md with results.
Note: This skill always works when invoked manually, regardless of the workflow.verifier config setting. The config toggle only controls whether /mz:go automatically suggests verification after execution.
Reference @skills/init/design-system.md for visual output formatting.
Reference @skills/shared/presentation-standards.md for content formatting rules.
Reference @skills/shared/terminology.md for official term definitions.
Reference @skills/verify/verifier.md for verification protocol and spawning patterns.
Output the stage banner:
+===============================================+
| MEGAZORD > VERIFY |
+===============================================+
Read .planning/megazord.config.json. If missing, display error and stop:
## Error
Project not initialized — no megazord.config.json found.
`/mz:init`
If config exists, continue loading:
.planning/STATE.md for current position..planning/ROADMAP.md for phase details and success criteria.Parse the user's message (text after /mz:verify) for arguments:
{phase_num} -- verify a specific phase (optional, defaults to current)--partial -- verify even if some plans are incomplete (skip completion check)--milestone {version} -- run milestone audit mode instead of phase verification (see Step 3b)Determine the plugin path for CLI commands:
plugin_path from the config JSON.plugin_path is not set in config, try ~/.claude/plugins/mz. Check if ~/.claude/plugins/mz/bin/megazord.mjs exists.## Error
Plugin path not configured.
`/mz:settings` to set plugin_path, or `/mz:init` to re-initialize.
Use this resolved path for all node {plugin_path}/bin/megazord.mjs commands below:
node {plugin_path}/bin/megazord.mjs tools plan incomplete --phase-dir {phase_dir}
.planning/phases/05-code-review-and-verification).## Error
Phase {N} not found in roadmap. Valid phases: {list}.
`/mz:status` to check current position.
Display:
> Target
Phase {N}: {Name} — {functional_sentence_from_goal}
The goal is already inline in the functional sentence — no separate "Goal:" line needed.
If --milestone {version} was provided, the skill enters milestone audit mode instead of the regular phase verification flow (Steps 4-7 are skipped).
Display:
> Milestone Audit: {version}
Determine which phases belong to this milestone:
.planning/MILESTONE.md exists with the specified version. If found, read the phases list from it.- [x]) and collect their phase numbers.Run the milestone audit via CLI tool:
node {plugin_path}/bin/megazord.mjs tools milestone audit --phases "{comma-separated phase numbers}"
Parse the JSON audit result.
Display audit results:
If all phases pass:
+===============================================+
| Milestone {version} Audit: PASSED |
+-----------------------------------------------+
| Phases verified: {N}/{N} |
| All verification gates passed. |
+===============================================+
Suggest: "Run /mz:plan to complete the milestone (archive and tag)."
If some phases fail:
+===============================================+
| Milestone {version} Audit: GAPS FOUND |
+-----------------------------------------------+
| Phases verified: {passed}/{total} |
| |
| Failed: |
| - Phase {N}: {Name} — {reason} |
| - Phase {M}: {Name} — No VERIFICATION.md |
+===============================================+
Suggest: "Run /mz:verify {N} for each failed phase."
Write MILESTONE-AUDIT.md in .planning/ with:
passed or gaps_foundDisplay Next Up block:
If passed:
## Next Up
**Milestone {version} audit passed.** Ready to archive.
`/mz:lifecycle`
<sub>`/clear` — start fresh context for the next step</sub>
If gaps found:
## Next Up
**Address gaps before closing milestone.**
`/mz:verify {N}` for Phase {N}: {Name}
<sub>`/clear` — start fresh context for the next step</sub>
Exit (do not proceed to regular phase verification).
If --milestone was NOT provided: Continue to Step 4 (regular phase verification flow).
node {plugin_path}/bin/megazord.mjs tools plan incomplete --phase-dir {phase_dir}
## Error
No plans found for Phase {N}: {Name}.
`/mz:plan {N}`
--partial was NOT provided:
Display warning listing incomplete plans. Ask user whether to proceed with partial verification or wait:
> Warning: Incomplete Plans
The following plans have not been executed:
- {plan_file_1}
- {plan_file_2}
Verification may report false gaps for artifacts in unexecuted plans.
Proceed with partial verification? (yes/no)
Or run with --partial to skip this check.
--partial was provided: proceed with a note:
> Note: Partial verification (--partial). Skipping completion check.
{plugin_path}/agents/mz-verifier.md content using the Read tool.<agent_role>
{content of agents/mz-verifier.md}
</agent_role>
<phase_goal>
{Phase goal from ROADMAP.md}
</phase_goal>
<success_criteria>
{Success criteria list from ROADMAP.md, numbered}
</success_criteria>
<plans>
{For each PLAN.md: filename, must_haves section from frontmatter, requirements field}
</plans>
<summaries>
{For each SUMMARY.md: filename, key accomplishments section, key-files, key-decisions}
</summaries>
<requirements>
{Requirement IDs mapped to this phase with their descriptions from REQUIREMENTS.md}
</requirements>
<verification_rules>
- Phase: {phase_number}
- Phase directory: {phase_dir}
- Report path: {phase_dir}/{padded}-VERIFICATION.md
- Mode: hybrid (automated first, user confirmation for ambiguous)
- Criteria sources: ROADMAP.md success criteria + PLAN.md must_haves
</verification_rules>
Spawn via Task tool:
subagent_type: "general-purpose"description: "Verify Phase {phase_number}: {phase name}"Wait for completion.
Parse the structured result (look for ## VERIFICATION COMPLETE).
Note: All file contents are read BEFORE spawning the Task subagent and embedded as inline text. @file references do NOT work across Task boundaries. This is the same pattern used by /mz:go for executor spawning (see skills/go/executor.md).
Parse the verification status from the structured result:
All truths verified, all artifacts exist and are wired, all requirements covered.
Display the strong phase completion banner first:
═══════════════════════════════════════════════════
### Phase {N}: {Name} — Complete
{Brief recap: what this phase delivered, 1-2 sentences}
═══════════════════════════════════════════════════
Then display verification stats:
+===============================================+
| Phase {N}: {Name} — Verification: PASSED |
+===============================================+
| Truths: {N}/{N} passed |
| Artifacts: {N}/{N} verified |
| Key Links: {N}/{N} wired |
| Requirements: {N}/{N} covered |
+===============================================+
Suggest advancing to next phase.
Some truths failed, artifacts missing, or requirements uncovered.
Display gaps clearly with details from VERIFICATION.md:
+===============================================+
| Phase {N}: {Name} — Verification: GAPS FOUND|
+===============================================+
| Truths: {passed}/{total} ({failed} failed) |
| Artifacts: {passed}/{total} |
| Key Links: {wired}/{total} |
| Requirements: {covered}/{total} |
+===============================================+
Then display each gap:
> Gaps
1. [FAILED] {truth/artifact/link description}
Evidence: {what was checked}
Expected: {what should exist}
Suggest running /mz:go to address gaps or re-planning.
Do NOT auto-advance the phase.
Some items need user confirmation.
Display the automated results (passed/failed items) first. Then present each UNCERTAIN item to the user one at a time:
> Human Verification Needed
1. {uncertain truth/artifact description}
Evidence checked: {what was verified}
Question: Does this satisfy the criteria? (yes/no/partial)
Collect user responses. Update the overall status based on responses:
Re-write VERIFICATION.md with user confirmations included.
Display Next Up block:
If passed:
## Next Up
**Plan Phase {N+1}: {NextName}** — {functional_sentence_of_next_phase}
`/mz:plan {N+1}`
<sub>`/clear` — start fresh context for the next step</sub>
If gaps_found:
## Next Up
**Address gaps in Phase {N}: {Name}** — fix issues, then re-verify
`/mz:go`
- Re-verify after fixes: `/mz:verify {N}`
<sub>`/clear` — start fresh context for the next step</sub>
If human_needed (after user confirms all): Update and proceed as passed or gaps_found based on user responses.
| Error | Step | Action |
|---|---|---|
| Config missing | Step 2 | Error box, suggest /mz:init. Stop. |
| ROADMAP.md missing | Step 2 | Error box, suggest /mz:plan. Stop. |
| Phase not found in ROADMAP | Step 3 | Error message with valid phase numbers. Stop. |
| No plans in phase | Step 4 | Error message, suggest /mz:plan. Stop. |
| Verifier subagent fails | Step 5 | Save error, display failure details. Stop. |
{plugin_path}/agents/mz-verifier.md) performs all verification logic. This skill orchestrates: load context, check completion, spawn verifier, handle result.{plugin_path} for CLI commands and agent files is resolved from config.plugin_path, falling back to ~/.claude/plugins/mz.