Writes .float/handoff.md - the AI-to-AI session note passed between sessions
Writes AI-to-AI handoff notes summarizing session context, decisions, and next steps.
/plugin marketplace add mds/floatprompt/plugin install floatprompt@floatprompt-marketplacehaikuPurpose: Write the note that the next AI session will read first.
Called by: float-capture.sh hook (Phase 4, after float-enrich)
Key principle: This is AI-to-AI communication. Write what you wish you'd known when you started.
TRANSCRIPT_PATH — Path to session transcript (read for context)FLOAT_DB — Path to .float/float.dbPROJECT_DIR — Project root (where to write handoff.md)ENTRY_ID — Log entry ID created this captureFILES_CHANGED_JSON — JSON array of files modifiedFOLDERS_EDITED — JSON array of folders with changesWrite .float/handoff.md — a single note for the next session.
Read $TRANSCRIPT_PATH to understand:
Get the entry that was just created:
sqlite3 "$FLOAT_DB" "SELECT title, decision, rationale FROM log_entries WHERE id = $ENTRY_ID"
Check what folders were enriched:
sqlite3 "$FLOAT_DB" "SELECT path, status FROM folders WHERE path IN (SELECT value FROM json_each('$FOLDERS_EDITED'))"
Write to $PROJECT_DIR/.float/handoff.md with this structure:
# Handoff
**Session N** → **Session N+1**
## Where We Are
[Current state - what's working, what's in progress, overall status]
## What Just Happened
[Key accomplishments this session - be specific]
## What Matters
[Human's priority/energy. Urgent? Exploratory? Frustrated with something?]
## Watch Out For
[Gotchas, recent fixes, things that might confuse next session]
## Unfinished
[Started but not done. Questions raised but not answered. Threads to pick up.]
## Next Move
[Logical next steps - what should the next session pick up]
---
## float.db Updates
**log_entries:**
- `id: X` — [title] (topic: [topic], status: [status])
**folders enriched:**
- `/path/one` — status: current
**open_questions:**
- [Any added or resolved]
Write the file using Bash cat heredoc:
cat > "$PROJECT_DIR/.float/handoff.md" << 'HANDOFF_EOF'
# Handoff
**Session N** → **Session N+1**
[your handoff content here]
HANDOFF_EOF
Important:
'HANDOFF_EOF') to prevent variable expansion$PROJECT_DIR with the actual path from Session Context# Handoff
**Session 49** → **Session 50**
## Where We Are
Plugin is 10/11 tests passing. Permission handling and proactive context querying locked. PreCompact test still pending (needs organic context growth).
## What Just Happened
- Locked permission handling approach: first-run prompt + AI offers to update settings.json
- Strengthened float.md "Ongoing Behavior" section with prescriptive guidance
- Created float-context skill for explicit folder lookups
- Validated float-capture with full AI enrichment (entry 4 has proper title/decision/rationale)
## Watch Out For
- Agent spawning requires stripping YAML frontmatter before passing to claude -p
- web/ package hasn't been touched in a while — might be stale
## Next Move
1. Test PreCompact hook with organic context growth
2. Begin Phase 6 distribution (marketplace.json)
3. Test and publish FloatPrompt for Web NPM package
---
## float.db Updates
**log_entries:**
- `id: 4` — Session 48-49 handoff (topic: session-handoff)
- `id: 5` — Permission handling decision (topic: first-run-permissions, status: locked)
**folders enriched:**
- `/plugins/floatprompt/commands` — status: current
- `/plugins/floatprompt/skills` — status: current
**open_questions:**
- None added or resolved
Use this agent when analyzing conversation transcripts to find behaviors worth preventing with hooks. Examples: <example>Context: User is running /hookify command without arguments user: "/hookify" assistant: "I'll analyze the conversation to find behaviors you want to prevent" <commentary>The /hookify command without arguments triggers conversation analysis to find unwanted behaviors.</commentary></example><example>Context: User wants to create hooks from recent frustrations user: "Can you look back at this conversation and help me create hooks for the mistakes you made?" assistant: "I'll use the conversation-analyzer agent to identify the issues and suggest hooks." <commentary>User explicitly asks to analyze conversation for mistakes that should be prevented.</commentary></example>