verify-implementation

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 exists
• npm test - If npm script exists
• go test ./... - Direct Go command
• pytest - Direct Python command

2. Run Linting

Check commands.md, then run:

• make lint - If Make target exists
• npm run lint - If npm script exists
• golangci-lint run - Direct Go command
• eslint . - Direct JavaScript command

3. Build Check

Check commands.md, then run:

• make build - If Make target exists
• npm run build - If npm script exists
• go build ./... - Direct Go command