cco-anatomy
Generate a compact project map so Claude understands the codebase without opening every file
Share learned file patterns across a team — export an anonymized digest of what's usually waste/useful/co-edited, and import a teammate's so a fresh clone benefits day one
$ npx -y skills add egorfedorov/claude-context-optimizer --skill cco-patterns --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/cco-patternsContext preview
The summary Claude sees to decide when to auto-load this skill.
Share learned file patterns across a team — export an anonymized digest of what's usually waste/useful/co-edited, and import a teammate's so a fresh clone benefits day one
name: cco-patterns description: Share learned file patterns across a team — export an anonymized digest of what's usually waste/useful/co-edited, and import a teammate's so a fresh clone benefits day one license: MIT argument-hint: "[export [--out <file>] | import <file> | show]" allowed-tools: [Bash, Read]
Patterns are learned per-machine, per-project: which files are usually waste, which are usually useful, which get edited together. A teammate's fresh clone starts blind and re-learns the same lessons by wasting the same tokens.
This makes them portable. Run from the plugin root (the directory containing `src/patterns-share.js`).
Parse $ARGUMENTS:
node src/patterns-share.js export
Writes `.cco/patterns.digest.json` in the project (override with `--out`).
The digest contains **relative paths and counts only** — no file contents, no absolute paths, no home directory. Entries that can't be expressed relative to the project root are dropped, and the command reports how many. The file is audited before writing; if anything identifying slipped in, it refuses to write rather than produce a file someone might commit.
Tell the user the digest is safe to commit, and that committing it is the point: teammates get it on clone.
node src/patterns-share.js import .cco/patterns.digest.json
Merges into the current project's patterns. **Imported data is stored as a separate prior — it never touches the user's own counts.**
This matters and is worth saying out loud: `confidence` is computed from how many sessions *this machine* observed a file in. Folding someone else's sessions into that number would make the user's own confidence a lie. So the import fills in only where the user has no observations of their own — the fresh-clone case — and local evidence always wins.
The digest is audited on the way in too, since it came from elsewhere. A file with absolute paths or traversal segments is rejected, not sanitized.
node src/patterns-share.js show
Lists what was imported for this project and how each file is labelled.
Worth offering when:
that every teammate will otherwise rediscover.
offer to import it.
Claude Code plugin that tracks token usage, identifies wasted context, and saves 30-50% on API costs. Heatmaps, ROI reports, budget alerts, efficiency scores, git-aware suggestions — all local, zero config.
Repo: egorfedorov/claude-context-optimizer
Generate a compact project map so Claude understands the codebase without opening every file
Configure token budget limits, auto-compact settings, and view current budget status (model-aware — Claude 5 lineup, Opus 5 default fallback, full 1M context…
Analyze the user's last prompt for clarity, scope and specificity — give a quality score and concrete suggestions to make the next prompt produce better results
View and tune CCO's behavior thresholds — re-read warnings, cache staleness, prompt-coach length bands, and the /cco-pack budget cap