ReddRadar
Agent skill
verify-spec
Verify that "Must Have" requirements from SPEC.md are implemented in the codebase.
Install this agent skill to your Project
npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/data/verify-spec
SKILL.md
Core responsibilities:
- Read requirements from SPEC.md
- Inspect codebase for proof of implementation
- Verify "Must Have" items strictly
- Report status of each requirement (Implemented/Missing/Partial)
Flow: Load Spec → Extract Requirements → Inspect Code → Verify → Report
- Task Name (from arguments or prompt)
./.gtd/<task_name>/SPEC.md— Source of truth
Agents used:
research— To find evidence in the code
1. Load Specification
Get Task Name:
- If provided in
$ARGUMENTS, use it. - If not, check if
d-workor similar has set a context, otherwise ask user: "Which task (spec) do you want to verify?"
Read Spec:
Read ./.gtd/<task_name>/SPEC.md.
If not found, error: "SPEC.md not found for task <task_name>".
2. Spawn Verification Agents
Group related "Must Have" items (e.g., by feature or domain) and spawn a researcher for each group.
Concurrency: As many as needed. Trigger: After grouping requirements.
Fill prompt and spawn:
<objective>
Verify requirement group: {group_name}
</objective>
<requirements>
1. {req_1}
2. {req_2}
</requirements>
<context>
- Task: {task_name}
- Spec File: {spec_path}
</context>
<investigation_checklist>
1. Search codebase for feature implementation (frontend/backend)
2. Trace logic to ensure it meets the requirement
3. Verify edge cases or constraints mentioned
4. Capture file paths and lines as evidence
</investigation_checklist>
<output_format>
Verification Result:
- Status: PASS / FAIL / PARTIAL
- Evidence: (File:Line, Logic description)
- Notes: (Missing edge cases, validation gaps)
</output_format>
Task(
prompt=filled_prompt,
subagent_type="researcher",
description="Verifying requirement: {req_short_desc}"
)
Note: Main agent aggregates results into the report.
3. Report Findings
Create a verification report (you can output this directly to the user or write to a file if lists are long).
Format:
# Verification Report: {task_name}
**Spec:** ./.gtd/{task_name}/SPEC.md
**Status:** {PASS / FAIL}
## Must Have Requirements
| Requirement | Status | Evidence/Notes |
| :---------- | :--------- | :---------------------------------------------- |
| {Req 1} | ✅ PASS | Found in `file.ts:Method`. Handles X correctly. |
| {Req 2} | ❌ FAIL | No code found for feature Y. |
| {Req 3} | ⚠️ PARTIAL | logic exists in `Z.ts` but missing validation. |
## Summary
- **Implemented:** X/Y
- **Missing:** Z/Y
**Recommendation:**
{Proceed to Verification/Fix Missing Items}
<offer_next>
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
GTD ► SPEC VERIFICATION COMPLETE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Task: {task_name}
Status: {PASS/FAIL}
[ ] {Req 1} ...
[ ] {Req 2} ...
─────────────────────────────────────────────────────
</offer_next>
<forced_stop> STOP. The workflow is complete. Do NOT automatically run the next command. Wait for the user. </forced_stop>
Didn't find tool you were looking for?