handoff

name	handoff
description	Generate a structured handoff summary for Parallax development sessions.
disable-model-invocation	true

/handoff -- Agent Handoff Summary

Generate a structured handoff summary for Parallax development sessions.

When to Use

Invoke /handoff at the end of a work session, before context gets too long, or when passing work to another agent/human.

Instructions

1. Promote to semantic memory if warranted. If this session produced stable knowledge (implementation details, debugging patterns, gotchas) not already in CLAUDE.md or MEMORY.md, propose additions to MEMORY.md. Session-specific details (what was done, next steps) stay in the session file only.
2. Write the handoff summary to docs/sessions/YYYY-MM-DD_short-description.md (use today's date, append _2 if file exists).
3. Include all sections from the output format below.
4. Be concise but complete -- another agent should continue without re-investigating.

Output Format

# Session: [date] -- [short description]

## What Was Done
- [bulleted list of completed work]

## Key Decisions
- [architectural choices, design tradeoffs, rejected alternatives]

## Current State
- What works: [list]
- What doesn't: [list]
- Test status: [pixi run check result summary]

## Next Steps
1. [prioritized list]

## Open Questions
- [unresolved items needing human input]

## Relevant Files
- [file paths with brief descriptions of changes]

Self-Review Pass

After writing the initial draft, re-read it critically and check for:

1. Unverified claims. Did you assert something works without evidence? Flag it as untested.
2. Missing context. Could a fresh agent pick up from this alone? What would they waste time rediscovering?
3. Gaps in "What doesn't work." "Nothing known" is almost always wrong. Think about what wasn't tested, what was deferred, what assumptions went unchecked.
4. Test intent shifts. If tests were modified, did the meaning of the test change, not just the assertion values?
5. Upstream continuity. Is there prior session context (plans, earlier sessions) that this handoff assumes but doesn't reference?

Update the draft with findings, then present to the user.

Rules

• Memory boundary: MEMORY.md = stable facts learned through experience. Session files = temporal context. CLAUDE.md = rules and conventions. When in doubt, session file -- promote later.
• Include file paths for all code that was touched or is relevant.
• Flag any known issues or risks explicitly.
• If a hypothesis was being tested, include its current status.
• Run pixi run check before writing the summary to capture current state.
• Verify ROADMAP.md and README.md are current before finalizing.