Crew Recall

Overview:

~/.agent/tools/journal-search.py summary

Repo patterns:

~/.agent/tools/journal-search.py auto-recall "$REPO" --top 5

Semantic search (preferred for questions):

~/.agent/tools/journal-search.py query "<search text>" --top 5

Structured search (fallback for metadata filters):

rg "<pattern>" ~/.agent/journal/ --context 10

Metadata filters via ripgrep:

• Type: rg "^\*\*Type:\*\* pattern" ~/.agent/journal/
• Tag: rg "#agent-memory" ~/.agent/journal/
• Repo: rg "^\*\*Repo:\*\* kibana" ~/.agent/journal/
• Date: search only files named YYYY-MM.md where YYYY-MM >= DATE

Step 3: Present results

• Summary queries: show the overview as-is, highlight interesting stats.
• Specific queries: for each match, show date, summary, and the most relevant field (Key idea, Rule, What I learned, Decisions made).
• Questions ("how does X work?"): synthesize an answer from matching entries, don't just list them.

If many results (>5), group by type and summarize themes. If nothing matches, say so and suggest broadening.

Rules

• Always use ~/.agent/tools/journal-search.py commands — don't read journal files sequentially
• Read ~/.agent/PROFILE.md for context on what the user cares about
• Present results concisely; the user wants answers, not raw entries
• When inside a repo, bias toward that repo's entries unless the query is clearly cross-repo