autoclaw-builder - SKILL.md Agent Skill

name: autoclaw-builder description: Implement focused code changes in PANaCEa. Sprint-based execution with mandatory verification gates. Use for all implementation work after scouting and architecture. mode: builder

Builder Mode — PANaCEa Implementation

Purpose

Execute focused code changes with verification at every sprint boundary.

When to Use

Implementing features, fixes, refactors
After Scout and Architect have defined the plan
Any code change regardless of size

Workflow: AUDIT → EXECUTE → VERIFY → LOG

AUDIT  → read every file you'll touch, understand patterns
EXECUTE → make changes in 1-4 file sprints
VERIFY → npm test + typecheck after EVERY sprint
LOG    → update .autoclaw/task-ledger.md

Sprint Rules

1-4 files per sprint
Each sprint independently testable
Verify (test + typecheck) between sprints
Direct edits for ≤4 files, sub-agents for 5+
NEVER skip verification

Verification Gates (MANDATORY)

npm test                                              # Full suite
NODE_OPTIONS="--max-old-space-size=4096" npx tsc --noEmit  # Typecheck
npm run build                                         # Production build

Code Rules

Audit imports before editing
Match existing patterns (naming, structure, error handling)
Use existing components/hooks/services — don't duplicate
Remove dead code and unused imports
Add tests for new behavior
Handle loading, empty, error states for UI changes
Verify auth/ownership for API changes

After Each Sprint

✅ Sprint N: {what changed}
   Files: {list}
   Tests: {passing}/{total}
   Build: pass/fail

Pre-Flight

cd /Users/aaronullger/GitHub/StudyPANaCEa
npm test                    # Baseline: all passing?
npm run typecheck           # Uses tsconfig.production.json

Coordination

Receives from: Architect (design), Debugger (bug triage), Orchestrator (task routing)
Hands off to: Reviewer (after every sprint), QA (final verification)
Sub-agents: spawn for parallel work on >4 files — verify ALL output before accepting

Common Pitfalls

Mixing refactors with features: Keep sprints focused on one concern
Skipping audit: Import errors are the #1 waste — read every file first
Typecheck OOM: Always use npm run typecheck (not raw tsc)
Sub-agent quality: Sub-agents produce wrong import paths and missing null safety — verify build + tests before accepting

Never

Mix refactors with features
Change public API behavior unintentionally
Skip safePrismaDisconnect in Edge functions
Use process.env in Edge
Commit without verification