doctor

cclens doctor — usage health check

Run cclens over the user's Claude Code transcripts and config, then present the one-screen health check. Everything runs locally and read-only over ~/.claude; nothing is sent anywhere.

1. Resolve the binary and the store

• Binary: use cclens if command -v cclens finds it. Otherwise run every command below through Nix instead: nix run github:lambdalisue/cclens -- <subcommand …>. If neither cclens nor nix is available, stop and tell the user how to install it (nix profile install github:lambdalisue/cclens, or cargo install --git https://github.com/lambdalisue/cclens).

• Store: if ./cclens.db exists, use it and omit --db. Otherwise use the per-user store so no file is dropped into the current project:

  DB="${XDG_CACHE_HOME:-$HOME/.cache}/cclens/cclens.db"
  mkdir -p "$(dirname "$DB")"
  

  and pass --db "$DB" to every command below.

2. Report

cclens doctor --db "$DB"

doctor refreshes the store automatically (incremental; fast when current) and prints a freshness line on stderr. If the user asked about one layer, add --scope global or --scope project:<slug>. (doctor has a --format json for machine use, but present the default text form.)

3. Present

Relay the report's findings in the conversation language, keeping its structure — it is split into a global section (fix in ~/.claude) and per-project sections (fix in each project's own config) precisely because those are different tasks; do not merge them back together. Preserve the caveats it prints (estimated tokens, evaluation windows); they exist so numbers are not over-trusted. Do not pad it with generic advice.

Close by naming the follow-ups: /cclens:optimize to investigate and fix the findings, /cclens:query for any specific slice the report does not show.