Feature Verification Skill

Trigger

• Keywords: verify, investigate, diagnose, check if working, anomaly, validate
• User provides credentials for data sources

When NOT to Use

• Need to modify data (this skill is read-only only)
• Pure code review (use codex-review)
• Feature development (use feature-dev)

Core Principle

ALL OPERATIONS MUST BE READ-ONLY

Claude independent analysis -> Form conclusion -> Codex third-perspective confirmation -> Integrated report

Workflow

Phase 1: Explore    -> Understand system architecture, data flow, trigger points
Phase 2: Plan       -> Build verification checklist, present to user for confirmation
Phase 3: Execute    -> Execute read-only queries, record expected vs actual
Phase 4: Analyze    -> Claude independently forms diagnostic conclusion
Phase 5: Confirm    -> /codex-brainstorm third-perspective verification
Phase 6: Integrate  -> Synthesize dual perspectives, produce final report

Safety Rules

Rule	Description
READ-ONLY	No write/update/delete operations allowed
PLAN-FIRST	Present verification plan before executing queries
CREDENTIAL-SAFETY	Do not expose full credentials in output
INDEPENDENT-FIRST	Claude forms conclusion first, then asks Codex to confirm

Verification Checklist

Category	Checks
Feature Checks	API endpoints, jobs, error handling
Data Checks	Collections, aggregations, external
Integration Checks	End-to-end flow

Verification

• Report includes Executive Summary + Status
• Each check has expected/actual/status
• Both Claude + Codex perspectives are documented
• Recommendations split into Immediate / Further / Long-term

References

• references/queries.md

  - Query templates + safety rules

• references/output-template.md

  - Report format

Examples

Input: /feature-verify User Auth
       MongoDB: mongodb://...
       {ANALYTICS}: project_id={PROJECT_ID}
Action: Explore -> Plan -> Execute queries -> Analyze -> Codex confirm -> Report

Input: verify user auth is working correctly
Action: Phase 1-6 flow -> Output diagnostic report