claude-contextlisted

# claude-context A ripgrep-backed search tool over `~/.claude` (conversations, project memories, global memories, plans, agents). It resolves the *real* project path for a hit and scores/ranks results, so you don't have to open large files to figure out relevance. Script: `~/.claude/tools/src/claude_context.py` Invoke as: `claude-context <args>` (assume it's on PATH; if not, run `python3 ~/.claude/tools/src/claude_context.py <args>`). ## Core principle: two-phase search, not full-file reads **Never** read a raw `.jsonl` conversation file or a whole memory directory to "look for" something — it burns enormous context for low signal. Instead: 1. **Recall pass** (cheap): search with `--format json` to get ranked candidates (id, score, snippet, path, tags) — a few hundred tokens. 2. **Expand pass** (targeted): once you've identified the right hit, `--expand <id>` to pull the *full* section/content of just that one hit. This mirrors how you'd want a human to work too: shortlist first, deep-dive only on the winner. ## When to use which mode | Situation | Command | |---|---| | "Have I dealt with X before, but I don't know where?" | `claude-context "X" --recall --format json` | | Know the exact phrase, want ranked hits | `claude-context "X" --format json` | | Found a promising hit, want full content | `claude-context --expand <id> --format json` | | Remember the *idea* but not the wording | `claude-context --related <id-or-path> --format json` | | Just want to browse wh