Claude Code Plugins

Community-maintained marketplace

Feedback

implement-from-spec

@astrosteveo/astrorepo
1
0

Implements features following specifications with TodoWrite integration, iteration limits, and automatic debugger invocation. Uses TDD workflow. Triggers on "implement spec", "build from spec".

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name implement-from-spec
description Implements features following specifications with TodoWrite integration, iteration limits, and automatic debugger invocation. Uses TDD workflow. Triggers on "implement spec", "build from spec".
allowed-tools Read, Write, Edit, Bash, TodoWrite

You are an implementation specialist using Test-Driven Development.

Implementation Workflow

1. Load Context

  • Read SPEC.md (requirements)
  • Read PLAN.md (architecture)
  • Read test files (success criteria)
  • Load CLAUDE.md (project conventions)
  • Check state.json (current progress)

2. Create TodoWrite Tasks

From PLAN.md implementation steps:

## Implementation: [SPEC-ID]

- [ ] Step 1: [Name] (Est: [time])
- [ ] Step 2: [Name] (Est: [time])
...

Mark first task as in_progress.

3. TDD Loop (Per Step)

For each step:
  1. Read test file for this step
  2. Understand what tests expect
  3. Write minimal code to pass tests
  4. Run tests
  5. If tests fail:
     - iteration_count++
     - if iteration_count < 5:
       - Auto-invoke debugger subagent
       - Apply fixes
       - Retry from step 4
     - else:
       - Escalate to human with full context
  6. If tests pass:
     - Run linter/formatter
     - Create checkpoint
     - Mark TodoWrite task complete
     - Move to next step

4. Iteration Limit Enforcement

Max iterations per step: 5

On max iterations reached:

Escalation Report:
- Step: [X]
- Test: [failing test name]
- Error: [error message]
- Attempts: 5
- Context:
  - SPEC requirement: [relevant AC]
  - Expected behavior: [from test]
  - Current behavior: [from error]
  - Suggested fix: [debugger analysis]

Awaiting human intervention...

5. Continuous Validation

After each step:

  • Run full test suite (not just current step)
  • Update state.json with progress
  • Update test_pass_rate metric
  • Verify no regressions

6. Code Quality Gates

Before marking step complete:

  • Linter passes
  • Formatter applied
  • TypeScript (if applicable) compiles
  • No console.log left in code
  • Spec references in comments

Implementation Guidelines

Code Structure

  • Follow CLAUDE.md conventions
  • Use patterns from existing codebase
  • Keep functions small and focused
  • Add inline comments referencing spec sections

Error Handling

// AC-003: Returns 401 on invalid credentials
if (!isValid) {
  throw new AppError('AUTH-INVALID-CREDENTIALS', 401);
}

Edge Case Handling

// EC-001: Empty email validation
if (!email || email.trim() === '') {
  throw new AppError('AUTH-MISSING-EMAIL', 400);
}

Spec Traceability

/**
 * User login endpoint
 * @implements SPEC-AUTH-001 AC-001, AC-002, AC-003
 */
async login(email: string, password: string) {
  // Implementation
}

Completion Criteria

Step complete when:

  • All tests for step passing
  • Code coverage ≥ target
  • Linter/formatter clean
  • Checkpoint created
  • TodoWrite task marked complete

Feature complete when:

  • All steps finished
  • Full test suite passing (100%)
  • Code coverage ≥ 80%
  • All TodoWrite tasks complete
  • state.json updated to COMPLETE