handoff

name: handoff description: Update or create a TPP for engineer handoff when session context is running low. Use when ending a session, handing off work, or capturing progress on complex tasks. allowed-tools: Read Edit Write Glob Grep disable-model-invocation: true

TPP Handoff

We're out of time and need to hand off remaining work to a new engineer (or future session).

Your Task

1. Find or create the TPP - Check doc/todo/ for an active TPP matching current work
2. Re-read key documents - Review TPP-GUIDE.md, TDD.md, SIMPLE-DESIGN.md
3. Update progress - Mark completed tasks, update current phase with completion checklists
4. Add context - Document discoveries, gotchas, and insights from this session
5. Record failures - Include attempted approaches that didn't work, and why
6. Prepare next steps - Clarify what remains and any blockers

TPP Location

• Active work: doc/todo/P{NN}-{desc}.md (e.g., doc/todo/P01-fix-aggregate-null.md)
• Completed: Move to doc/done/{date}-P{NN}-{desc}.md
• Priority: P00 (critical) through P99 (nice-to-have)

Create doc/todo/ and doc/done/ directories if they don't exist.

Style Guide

Follow TPP-GUIDE.md:

• Keep under 400 lines (trim redundancy, clarify, simplify)
• Transfer expertise, not just instructions
• Highlight uncertainties for the next engineer to explore
• Include verification commands for completed work
• Be precise about what tests validate vs. what requires code review

Key Sections to Update

• Goal Definition - Problem, why it matters, success test, key constraints
• Context Research - Patterns found, landmines discovered, node:sqlite behavior
• Tasks - Mark done items, add new discoveries, include completion checklists
• Tribal Knowledge - N-API gotchas, platform-specific issues, learned the hard way

node-sqlite Specific Concerns

When documenting, remember to include:

• API compatibility: Does behavior match node:sqlite exactly?
• Upstream files: Never modify src/upstream/* - note if upstream sync is needed
• N-API patterns: Document traps like ArrayBufferView checking, SQLite callback context, aggregate state
• Platform issues: Windows file locks, Alpine ARM64 slowness, macOS VM timing
• Test coverage: What the test actually validates vs. what's validated by code review

Validation Requirements

Every completed task needs verifiable proof:

• Commands that pass: npm t, npm run lint, etc.
• Code locations: src/sqlite_impl.cpp:234 where implementation exists
• Integration proof: grep commands showing production usage
• Behavior comparison: Test output against node:sqlite

The Goal

The next engineer should be able to continue seamlessly without asking questions—even if the code changed since you wrote the TPP.