AgentSkillsCN

phase-implement

通过任务分派与质量把控,统筹协调实施阶段的各项任务,助力工程师高效完成各自的工作。

SKILL.md
--- frontmatter
name: phase-implement
description: "Coordinate implementation phase by delegating tasks and ensuring code quality. Use during implementation phase to manage engineer tasks."
mcp_fallback: none
category: phase
phase: Impl

Implementation Coordination Skill

Coordinate implementation phase by breaking down work, delegating to engineers, and maintaining quality standards.

When to Use

  • Starting implementation phase (after Plan completes)
  • Running in parallel with Test and Package phases
  • Need to delegate implementation tasks
  • Managing multiple engineering tasks simultaneously

Quick Reference

bash
# Review GitHub issue for implementation tasks
gh issue view <number> --json body --jq '.body'

# Verify code quality
pixi run mojo test -I . tests/
pixi run mojo build -I . <module>
just pre-commit-all

# Check for warnings (zero-warnings policy)
# Any output = fix before committing

Workflow

  1. Review plan specifications - Understand deliverables and success criteria
  2. Break into tasks - Create granular implementation work items
  3. Delegate by complexity:
    • Senior Engineer: Complex algorithms, SIMD, memory management
    • Engineer: Standard implementations, business logic
    • Junior Engineer: Boilerplate, simple helpers, type definitions
  4. Monitor progress - Check completion status, unblock issues
  5. Code review - Verify quality, standards, tests, documentation
  6. Final checks - Format, lint, test coverage, no warnings

Delegation Matrix

ComplexityTaskEngineer
HighComplex algorithms, SIMD optimizationsSenior
HighMemory management, ownership patternsSenior
MediumStandard functions, business logicStandard
MediumData structures, operationsStandard
LowBoilerplate, type aliasesJunior
LowSimple helpers, constantsJunior

Mojo Implementation Standards

Function definitions:

mojo
# High-performance critical
fn simd_add[dtype: DType](a: Tensor[dtype], b: Tensor[dtype]) -> Tensor[dtype]:
    """SIMD-optimized addition."""
    pass

# Flexible/Python interop
def load_model(path: String) -> PythonObject:
    pass

Memory management:

mojo
# Transfer ownership
fn process(owned data: Tensor) -> Tensor:
    return data^

# Read-only access
fn analyze(borrowed data: Tensor):
    pass

# Mutable access (Mojo v0.26.1+)
fn modify(mut data: Tensor):
    pass

Quality Checklist

Before code review approval:

  • All tests passing
  • 80% test coverage minimum
  • No compiler warnings (zero-warnings policy)
  • pixi run mojo format applied
  • Docstrings complete
  • No TODOs/FIXMEs (or documented)
  • Performance meets requirements
  • Follows Mojo syntax standards (Mojo v0.26.1+)

Phase Dependencies

  • Input from: Plan phase (specifications and deliverables)
  • Parallel with: Test phase (TDD) and Package phase
  • Precedes: Cleanup phase (after parallel phases complete)

Output Location

  • Implementation: /shared/<module>/, /examples/, /tooling/
  • Documentation: GitHub issue comments
  • Issue updates: Track progress, blockers, learnings via gh issue comment

Error Handling

BlockerResolution
Unclear requirementsEscalate to Design for clarification
Performance issuesConsult Performance Specialist
Test failuresDebug with Test Specialist
Missing dependenciesUpdate Plan, communicate status
Compiler warningsFix immediately (zero-warnings policy)

References

  • CLAUDE.md - "Mojo Syntax Standards" (current v0.26.1+ patterns)
  • CLAUDE.md - "Critical Pre-Flight Checklist" (before committing)
  • CLAUDE.md - "Common Mistakes to Avoid" (64+ test failure learnings)
  • notes/review/mojo-test-failure-learnings.md - Real implementation patterns

Key Principle: Implement to make tests pass, not the other way around.