Search Learnings

Queries ChromaDB for relevant learnings. Returns tiered results:

• •High confidence (distance < 0.5): Strong match
• •Possibly relevant (0.5-0.7): Review before applying

Usage

# Standard search
python3 search-learnings.py "query" /working/dir

# Peek mode (mid-conversation, excludes seen IDs)
python3 search-learnings.py "query" /working/dir --peek --exclude-ids "id1,id2"

When to Search

Check manifest (~/.projects/learnings/MANIFEST.md) to see what topics exist.

Search when:

• •Task involves a topic in the manifest (authentication, testing, etc.)
• •Error occurs in a manifest topic area
• •User references past work

Skip when:

• •Manifest shows no related topics
• •Trivial changes

Peek when:

• •You shift to a new topic that's in the manifest
• •Stuck on a problem in a manifest topic area

Output

{
  "status": "success",
  "high_confidence": [...],
  "possibly_relevant": [...]
}

Peek mode returns {"status": "found", "learnings": [...]} or {"status": "empty"}.

Key Rules

1. •Check manifest first to see if relevant topic exists
2. •Use topic + context in queries: "authentication JWT refresh" not "implement login"
3. •High confidence = apply directly
4. •Possibly relevant = review first
5. •Empty peek = continue silently