feat: add agent factory with ts-pattern for type-safe instantiation

[arittr]

feat: add agent factory with ts-pattern for type-safe instantiation

Create agent factory using ts-pattern for exhaustive, type-safe agent instantiation, replacing the if/else chain in generator.ts. Update type system to include 'gemini' as a valid agent name.

This task establishes the infrastructure for adding new agents easily by providing a single location to add new agents with compile-time exhaustiveness checking via ts-pattern.

Changes:

• Add AgentName type to include 'claude' | 'codex' | 'gemini'
• Create createAgent() factory function using ts-pattern (≤15 LOC)
• Factory returns correct agent instance for 'claude' and 'codex'
• Factory provides exhaustiveness checking (TypeScript error if case missing)
• Replace generator.ts if/else with single createAgent() call
• Use ts-pattern for default signature generation in generator
• Update CLI schemas to accept 'gemini' as valid agent option
• Add comprehensive factory tests with >80% coverage

All acceptance criteria met:

• ✅ AgentName type includes 'claude' | 'codex' | 'gemini'
• ✅ createAgent() function ≤15 LOC using ts-pattern
• ✅ Factory returns correct agent instance for 'claude' and 'codex'
• ✅ Factory provides exhaustiveness checking (TypeScript error if case missing)
• ✅ Generator.ts replaces if/else with single createAgent() call
• ✅ CLI schemas accept 'gemini' as valid agent option
• ✅ Factory tests achieve 100% coverage
• ✅ All existing agent tests still pass (126 tests passing)
• ✅ bun run check-types passes with no errors
• ✅ bun run lint passes with no violations

🤖 Generated with Claude Code

Co-Authored-By: Claude noreply@anthropic.com

test: add mock cleanup to prevent test isolation issues

Add afterEach hooks to Claude and Codex unit tests to clean up mocks
after each test. Add mock.restore() call at module load in shell integration
tests to clean up any lingering mocks from prior test suites.

This prevents mock pollution that was causing shell integration test
failures when running tests directly with 'bun test' (parallel execution).

The recommended way to run the full test suite is 'bun run test' which
executes unit tests sequentially followed by integration tests, ensuring
proper cleanup between test suites.

Test results:

• bun run test (sequential): ✅ 365 pass, 0 fail
• bun test (parallel): Still has issues due to Bun's parallel test execution

The sequential execution via npm scripts is the intended and supported
approach for running the full test suite.

[arittr]

This change is part of the following stack:

• refactor(agents): extract commit message cleaning logic to shared utilities #58
  • feat: add agent factory with ts-pattern for type-safe instantiation #59 ◀
    • feat: implement Gemini agent with CLI integration #60
      • feat: add Gemini agent support to evaluation system #61
        • refactor: establish SUPPORTED_AGENTS as single source of truth #62
          • docs: update documentation to reflect Gemini agent support #63
            • feat: add agent configuration support to init command #64
              • fix: preserve staged changes after pre-commit hook #65

_{Change managed by git-spice.}