Implement code using sonnet model with full main context access
View on GitHubZ-M-Huang/claude-codex
claude-codex
January 24, 2026
Select agents to install to:
npx add-skill https://github.com/Z-M-Huang/claude-codex/blob/main/plugins/claude-codex/skills/implement-sonnet/SKILL.md -a claude-code --skill implement-sonnetInstallation paths:
.claude/skills/implement-sonnet/# Implement with Sonnet
## PREREQUISITE CHECK (MANDATORY)
Before implementing, verify plan reviews are complete by running:
```bash
"${CLAUDE_PLUGIN_ROOT}/scripts/state-manager.sh" validate-plan-reviews
```
**IF THIS COMMAND FAILS** → Stop immediately. Do NOT write any code.
The error message tells you which reviews are missing. Run them first:
- `/review-sonnet`
- `/review-opus`
- `/review-codex`
All must have status: "approved" before you can proceed with implementation.
---
You are implementing code as part of the multi-AI pipeline, using the sonnet model for efficiency while maintaining full context access.
## Your Role
- **Implement the approved plan** in `.task/plan-refined.json`
- **Follow project standards** in `docs/standards.md`
- **Write clean, tested code** following best practices
## Before You Start
1. Read `.task/plan-refined.json` for the implementation plan
2. Read `docs/standards.md` for coding standards
3. If there's previous review feedback, read the relevant review file:
- `.task/review-sonnet.json`
- `.task/review-opus.json`
- `.task/review-codex.json`
## Implementation Process
1. **Understand the plan** - Review all requirements and technical approach
2. **Check existing code** - Read files that will be modified
3. **Implement changes** - Write code following the plan
4. **Add tests** - Create tests for new functionality
5. **Verify** - Run tests and linters if available
## Output
Write implementation results to `.task/impl-result.json`:
```json
{
"status": "completed|failed|needs_clarification",
"summary": "What was implemented",
"files_changed": ["path/to/file.ts"],
"tests_added": ["path/to/test.ts"],
"questions": []
}
```
## Guidelines
- Follow the plan exactly - don't add unrequested features
- Keep changes minimal and focused
- Write tests for all new functionality
- If blocked or unclear, add questions to the output instead of guessing
- Address ALL feedback from previous reviews if present
## After Issues Found: