Search Algorithm

Step 1: Parse arguments

•Extract --after DATE or --project DIR flags if present
•Everything else is the search phrase (keep as literal text, don't interpret as project names)

Step 2: Find matching files

Search strategy (flexible):

•Try exact phrase first: grep -l -i "thread inversion" ...
•If no results, try all words (AND): files containing every word
•Rank by match count

bash

# Search ALL projects (don't filter by assumed project names)
grep -l -i "SEARCH_PHRASE" ~/.dotfiles/.claude/projects/*/*.jsonl 2>/dev/null

Critical rules:

•Do NOT assume words in the search are project names to filter by
•Only use --project flag for explicit project filtering

Step 3: For each matching file, extract:

•Session ID (filename without extension)
•Project (parent directory name)
•Date (file mtime)
•Topic (first non-system user message, truncated to 80 chars)
•Match count (grep -c)

Step 4: Output

•Sort by: match count (desc), then date (desc)
•Limit to top 10 results
•Format:

code

Session: <session-id>
Project: <project-name>
Date: <timestamp>
Topic: "<first user message snippet>..."
Matches: <count>
Resume: claude -r <session-id>

If no matches: "No sessions found matching '<terms>'. Try broader search terms."

Common Search Patterns

•jj change IDs: "ttyomtpu"
•jj bookmarks: "fix-cli-explicit-targets"
•Issue references: "Issue #62"
•Topics: "thread inversion", "fingerprint"

Filters (only when explicitly requested)

•--after DATE: filter by mtime
•--project DIR: limit to specific project directory