dimetron / code-review-pi

Code Review

Review code for quality, test coverage, and fix problems found.

The canonical output artifact for this skill is ./specs/issues/003-code-review-pi/PROMPT.md.
Only report success after writing the final review summary there.

Steps

1. Identify scope: Run git diff --name-only to find modified files. If there are changes, review only modified packages. If no uncommitted changes, review all packages (./...).

2. Run linters: Execute linters on target packages:

   • go vet ./...
   • golangci-lint run --new-from-rev=HEAD~1 ./... (if available)

3. Run tests and check coverage:

   • go test -race -count=1 -coverprofile=coverage.out ./...
   • go tool cover -func=coverage.out
   • Identify failing tests, functions with 0% coverage, packages below 80%

4. Review code quality: Read target files and check for:

   • Missing tests for new/changed functions
   • Error handling (missing checks, swallowed errors)
   • Resource leaks (unclosed files, connections)
   • Concurrency issues (race conditions, missing mutexes)

5. Fix issues in parallel using subagents: For each package needing coverage fixes, spawn a subagent to work on it concurrently:

   • Use the agent tool to spawn one subagent per package that needs test coverage improvement
   • Each subagent prompt: "Write tests for uncovered functions in package <pkg>. Read the source files, identify untested functions using the coverage data, write table-driven tests covering happy path and error cases. Run go test -race ./<pkg>/... to verify. NEVER delete existing tests."
   • While subagents write tests, fix linter warnings and code quality issues yourself
   • Wait for all subagents to complete, then verify their work compiles

6. Verify gates: All gates must pass before reporting success:

   • GATE 1: go build ./... — zero errors
   • GATE 2: go vet ./... — zero warnings
   • GATE 3: go test -race ./... — all tests pass
   • GATE 4: go test -cover ./... — >= 80% per package
   • If any gate fails, fix and re-run until all pass

7. Report: Write the final summary to ./specs/issues/003-code-review-pi/PROMPT.md and present:

   • Gates: PASS / FAIL for each
   • Test results: pass/fail count
   • Coverage: per-package percentages (before → after)
   • Issues fixed with file:line references

Gates

All gates are mandatory. Do not report success unless every gate passes.

Examples

• /code-review — Full review: lint, test, coverage, parallel fix, verify gates

Guidelines

• If uncommitted changes exist, focus on modified packages
• If no changes, review all packages for full codebase quality check
• NEVER delete existing tests — only add or fix tests
• Use subagents to parallelize test writing across packages (one subagent per package)
• While subagents handle coverage, fix linter and quality issues in parallel
• Tests are mandatory: every new/changed exported function needs a test
• Table-driven tests preferred for Go code
• Test both happy path and error cases
• Fix issues directly — don't just report them
• Keep iterating until all gates pass