verify-implementation
SKILL.md
Verify Implementation
Run verification commands after implementing changes to ensure correctness.
Verification Sources
Check thoughts/notes/commands.md (created by discover-project-commands skill) for:
- Available test commands
- Linting commands
- Build commands
- Other project-specific verification
IMPORTANT: Always Use Verification Agent
DO NOT run verification commands directly in main context.
Spawn a verification agent to run checks and return summary:
Task(subagent_type="Bash",
prompt="Run verification commands from plan success criteria:
- make test
- make lint
- make build
[list all automated checks from plan]
Return concise summary:
- If all pass: ✅ All verification passed
- If any fail: ❌ Verification failed
- List ONLY the failing checks
- Include first few lines of error
- Omit stack traces and verbose output
Maximum output: 2k tokens")
Why use an agent:
- Raw output can be 10k+ tokens (floods main agent context)
- Agent filters to only essential info (1-2k tokens)
- Keeps main agent focused on implementation
- Part of token management strategy (see
spawn-implementation-agents)
Agent returns:
- ✅ Pass status or ❌ Fail status
- Only failed checks (not successful ones)
- First few lines of errors (not full output)
- Actionable information only
Common Verification Steps
1. Run Tests
Check commands.md, then run appropriate command:
make test- If Make target existsnpm test- If npm script existsgo test ./...- Direct Go commandpytest- Direct Python command
2. Run Linting
Check commands.md, then run:
make lint- If Make target existsnpm run lint- If npm script existsgolangci-lint run- Direct Go commandeslint .- Direct JavaScript command
3. Build Check
Check commands.md, then run:
make build- If Make target existsnpm run build- If npm script existsgo build ./...- Direct Go command
4. Type Checking (if applicable)
npm run typecheck- TypeScriptmypy .- Python- Go builds include type checking
Verification Workflow
- Check for commands.md: Read
thoughts/notes/commands.md - Run automated checks: Use commands from reference doc
- Report results: Show pass/fail for each check
- Manual verification: Prompt for manual testing if needed
Handling Failures
If verification fails:
- Show the error output
- Analyze the error
- Fix the issue
- Re-run verification
- Don't mark task complete until all checks pass
Example
# Read the commands reference
cat thoughts/notes/commands.md
# Run appropriate commands
make test
make lint
make build
# All pass? Mark phase complete in plan
# Any fail? Debug and fix before proceeding
Important
Always spawn agent for verification - never run commands directly in main context.
See spawn-implementation-agents skill for full orchestration pattern.
Weekly Installs
2
Repository
eveld/claudeGitHub Stars
4
First Seen
Feb 28, 2026
Security Audits
Installed on
opencode2
gemini-cli2
codebuddy2
github-copilot2
codex2
kimi-cli2