| name | qcode |
| description | Implement planned changes with quality gates. Use when user types 'qcode' OR when ready to execute implementation. Follows qthink rules, coding best practices, and runs npm build and test quality gates. |
QCODE - Implementation Workflow
Purpose
This skill activates when the user types "qcode" to signal that you should implement the discussed plan while following all coding best practices and quality gates.
Instructions
When this skill is invoked, you MUST follow this workflow:
1. Rules
Before and during implementation:
- Do not overengineer your edits
- Do not introduce silent fallback errors
- Follow KISS/YAGNI/DRY principles
- Imports grouped: React/libraries, local modules, CSS
- Functional components with hooks only
- TypeScript interfaces at top of file, no
any
2. Coding Best Practices
- C-1: Follow TDD when appropriate: write failing test → implement → validate
- C-2: Use existing domain vocabulary and patterns for consistency
- C-3: Co-locate CSS with components
- C-4: Use
process.env.PUBLIC_URL for asset paths
- C-5: Maintain terminal aesthetic (dark theme, monospace, prompt-style)
- C-6: HashRouter for all routing (GitHub Pages requirement)
- C-7: Remove unused code — don't comment it out
- C-8: No TODOs — all code must be production-ready
- C-9: Semantic HTML + ARIA attributes for accessibility
3. Quality Gates
After implementation, run these checks:
Step 1: Build Check
npm run build
- This MUST pass without errors
- Validates TypeScript compilation and bundling
Step 2: Test Check
npm test -- --watchAll=false
- All tests MUST pass
- Fix any failing tests before proceeding
4. Mandatory Subagent Reviews (MANDATORY)
After all quality gates pass and before reporting completion, dispatch both code-quality-reviewer and plan-verifier in parallel — a single assistant message containing two Task tool calls, one with subagent_type="code-quality-reviewer" and one with subagent_type="plan-verifier". Do not run them sequentially: they review independent concerns (implementation quality vs plan adherence), so concurrent dispatch is faster and avoids one anchoring on the other's findings.
code-quality-reviewer will:
- Skim the diff against project invariants, coding rules, and the best-practices checklists.
- Surface readability, complexity, security, and architectural concerns.
- Auto-loop a fix-up implementer pass up to 2 times if substantive issues are found.
plan-verifier will:
- Locate the plan (most recent
/Users/randychan/.claude/plans/*.md matching the branch/task, plus the active TodoWrite list).
- Run the four-check rubric (task completeness, scope discipline, test coverage, plan verification section).
- Auto-loop a fix-up implementer pass up to 2 times if gaps are found.
If either agent triggers an auto-loop fix-up pass, re-dispatch both in parallel after each fix-up round so the gates stay synchronised.
qcode is not complete until BOTH subagents return ✅ (or each has exhausted its 2 auto-loop rounds with documented residual gaps).
5. Implementation Checklist
Before marking implementation complete, verify:
Expected Output
After running this skill, you should:
- Have implemented the planned changes
- Pass all quality gates (build, test)
- Have production-ready code with no TODOs
- Confirm all files follow project conventions
- Report any issues encountered during implementation
- Have dispatched
code-quality-reviewer and plan-verifier in parallel and either received ✅ from both or exhausted each agent's 2 auto-loop rounds with documented residual gaps
