Autonomous Phase Execution

Execute one roadmap phase: think → plan → implement → handoff.

Input

Receives from main executor:

• Phase goal (one sentence)
• Success criteria (list of outcomes)
• Project context (language, test command, lint command)
• Previous phase handoff (YAML file path, or "none" for phase 0)

Process

Step 1: Think (Internal - Not Saved)

Evaluate 2-3 architecture approaches internally:

• List components needed
• Consider patterns (MVC, Repository, Factory, etc.)
• Pick simplest that meets requirements
• Decision time: 2 minutes max per choice

Use decision framework:

• Prefer standard patterns over novel
• Prefer simple over complex
• Prefer loose coupling
• Avoid over-engineering

Step 2: Plan (Internal - Not Saved)

Break phase into TDD task pairs:

• Each task: one test file + one impl file
• Sequence tasks logically (dependencies first)
• Estimate: 2-5 minutes per task
• Keep tasks focused (single responsibility)

Step 3: Implement

For each task in sequence:

1. Write test

   • Create or update test file
   • Write test(s) for new functionality
   • Follow language conventions

2. Verify RED

   • Run test command
   • Confirm test FAILS
   • If passes unexpectedly, fix test

3. Write implementation

   • Create or update implementation file
   • Write minimal code to pass test
   • Follow language best practices

4. Verify GREEN

   • Run test command
   • Confirm test PASSES
   • If fails, fix implementation

5. Run quality gates

   • Compile: [language-specific build]
   • Lint: [language-specific lint]
   • All tests: [full test suite]
   • Verify: ALL pass

6. Commit

   • Stage test and impl files

   • Commit with descriptive message:

     feat: [component/feature name]
     
     - What: [what was built]
     - How: [pattern/approach used]
     - Integration: [dependencies]
     - Testing: [what tests cover]
     

Step 4: Verify Success Criteria

After all tasks:

• Check each success criterion from roadmap
• Run integration tests if specified
• Verify all quality gates pass

Step 5: Output Handoff

Create handoff document at

docs/handoffs/phase-[N]-handoff.yml

:

built: [Component1, Component2]
api:
  - Component1.method(args)->return
  - Component2.method(args)->return
patterns: [MVC, Repository]

Include:

• built

  : List of components/modules created (names only)

• api

  : Key public interfaces (function signatures)

• patterns

  : Architectural patterns used

Keep handoff minimal: 50-100 tokens.

Quality Gates (Non-Negotiable)

Before any commit, ALL must pass:

• ✅ Code compiles (no errors)
• ✅ Linter passes (zero issues)
• ✅ All tests pass (zero failures)
• ✅ No regressions (existing tests still pass)

If any gate fails:

• Fix immediately
• Re-run all gates
• Never commit broken code

Language-Specific Commands

Python:

• Test:

  pytest tests/ -v

• Lint:

  ruff check . && mypy .

• Build: N/A (interpreted)

Go:

• Test:

  go test ./... -race -cover

• Lint:

  golangci-lint run ./...

• Build:

  go build ./...

Rust:

• Test:

  cargo test

• Lint:

  cargo clippy --all-targets

• Build:

  cargo build

TypeScript:

• Test:

  npm test

• Lint:

  eslint . && tsc --noEmit

• Build:

  tsc

Java:

• Test:

  mvn test

  or

  gradle test

• Lint:

  mvn checkstyle:check

• Build:

  mvn compile

Decision Framework

When choosing architecture/technology:

1. Simplest solution that meets requirements
2. Well-known patterns (MVC, Repository, Factory)
3. Standard libraries over custom
4. 2 minutes max per decision
5. If unclear: pick most common approach

Avoid:

• Over-engineering
• Premature optimization
• Novel architectures
• Tight coupling
• Magic behavior

Prefer:

• Simple and clear
• Easy to test
• Standard patterns
• Loose coupling
• Explicit behavior

Error Handling

If quality gate fails after 3 fix attempts:

• Document blocker
• Report to main executor
• Do not proceed to next task

If success criteria not met:

• Document gap
• Report to main executor
• Do not output handoff

Output Summary

After completion, report:

• Components created
• Tests added (count and passing)
• Commits created (count and SHAs)
• Handoff location
• Success criteria status
• Quality gate results