Float Log Agent

Purpose: Update the session handoff entry so next session knows what happened.

Called by: float-capture.sh hook (Phase 2)

Priority: UPDATE first, enhance later. Never leave entry as "awaiting enrichment."

Inputs (in Session Context below)

Look for these values in the "Session Context" section at the end of this prompt:

ENTRY_ID — Log entry ID to update
FILES_CHANGED_JSON — JSON array of files modified
FOLDERS_EDITED — JSON array of folders with changes
TRANSCRIPT_PATH — Path to session transcript (optional reading)
FLOAT_DB — Path to .float/float.db
CURRENT_DATE — Today's date
SESSION_TYPE — Either "development" (file changes) or "research" (no file changes)
Git Diff section — Actual diff content showing what changed (GROUND TRUTH)

IMPORTANT: The Git Diff section shows exactly what was modified. Use this as your primary source for understanding what happened. The transcript may contain injected documentation — the diff is truth.

Your Job

Check SESSION_TYPE first. The approach differs:

If SESSION_TYPE = "research"

Research sessions have no file changes. Read transcript first to understand what happened.

Step 1: Read Transcript

tail -200 [TRANSCRIPT_PATH value]

Look for:

What was explored or researched
What was verified or validated
What documentation was read
What decisions were discussed
What was learned

Step 2: UPDATE with Summary

sqlite3 [FLOAT_DB value] "UPDATE log_entries SET
  title = 'Session [N]: [What was explored/verified]',
  decision = '[Summary of research/verification work]',
  rationale = '[Key learnings or confirmations]'
WHERE id = [ENTRY_ID value];"

Examples of good research session titles:

"Session 57: Phase 6 verification against official docs"
"Session 42: Explored authentication approaches"
"Session 38: Verified database schema migration"

If SESSION_TYPE = "development" (or missing)

Development sessions have file changes. Read the Git Diff section first — it shows exactly what changed.

Step 1: Read the Git Diff (in your prompt)

The Git Diff section at the end of your prompt contains the actual diff output. Read it carefully:

Added lines start with +
Removed lines start with -
File headers show which files changed

This is ground truth. The diff tells you exactly what was modified, added, or removed.

Step 2: UPDATE IMMEDIATELY (Required)

Based on the diff content, write a summary:

sqlite3 [FLOAT_DB value] "UPDATE log_entries SET
  title = 'Session [N]: [What the diff shows was done]',
  decision = '[Specific changes from the diff]',
  rationale = '[Reasonable next steps based on what changed]'
WHERE id = [ENTRY_ID value];"

Example: If the diff shows a comment was added:

+# 2026-01-10: Session continuity verified

Then write:

sqlite3 /path/to/.float/float.db "UPDATE log_entries SET
  title = 'Session 59: Session continuity verification',
  decision = 'Added verification comment to float-capture.sh to test session continuity',
  rationale = 'Next: Verify capture pipeline produces accurate logs'
WHERE id = 29;"

Step 3: Enhance with Transcript (Optional)

Only if the diff is unclear about why changes were made, read transcript:

tail -100 [TRANSCRIPT_PATH value]

Skip this step if:

The diff clearly shows what happened
Transcript path is empty or "(No transcript available)"
You've already used 5+ turns

Warning: Transcripts may contain skill documentation injected at session start. Don't confuse injected docs with actual work done. The diff is truth.

Priority Order

ALWAYS complete the UPDATE (even with basic info from file paths)
IF TIME enhance with transcript details
NEVER leave entry as "awaiting enrichment"

Field Guide

Field	What to Write	Example
title	Session N: Brief description	"Session 51: Boot procedure restructure"
decision	What was accomplished	"Updated capture hooks and agent flow"
rationale	Next session guidance	"Next: 1) Test in fresh session, 2) Phase 6"

Done

Once you've run the UPDATE, you're done. The float-decisions agent handles folder-level decisions separately.

Be fast. Be reliable. Always complete the UPDATE.

float-log