memory-extract

Extract Decisions from Previous Session

Analyze the condensed transcript from the previous session and extract project decisions.

Steps

1. Check for pending transcript: Read .ai-memory/.last-session.txt.

   • If it doesn't exist or is empty, tell the user: "No previous session transcript found. Decisions are captured automatically when a session ends and extracted at next session start."

2. Read the condensed transcript from .ai-memory/.last-session.txt.

3. Perform the extraction yourself using these rules:

   INCLUDE (these ARE decisions):

   • Why-decisions: "We chose X because Y", "Let's use X for Y"
   • Constraints: "We can't use X because Y", "X must be Y"
   • Scope: "Let's not do X for now", "X is out of scope"
   • Conventions: "Let's name X as Y", "We'll follow X pattern"
   • Unresolved: "We need to figure out X", "TODO: decide on X"

   EXCLUDE (these are NOT decisions):

   • Implementation details (specific code changes, debugging steps)
   • Git operations and file management
   • Questions that were fully resolved in the same session
   • Tool usage and progress updates

   DEDUPLICATION:

   • If the same decision appears multiple times, keep only the most refined version
   • If a decision was made and then changed, keep only the final version

   OUTPUT FORMAT: For each decision, produce a JSON line:

   {"id":"<8-hex>","ts":"<ISO8601>","category":"architecture|constraint|convention|testing|scope|unresolved","decision":"<one sentence>","rationale":"<why>","confidence":<0.0-1.0>,"source":"auto"}
   

4. Append extracted decisions to .ai-memory/decisions.jsonl (one JSON line per decision).

5. Delete .ai-memory/.last-session.txt (it has been processed).

6. Run sync to update tool-specific files:

   node -e "require('${CLAUDE_PLUGIN_ROOT}/scripts/sync-tools.js').syncAll(process.cwd())"
   

7. Report results: Show how many decisions were extracted, grouped by category.