Cc optimizer analyze

Skill Korck-lab/claude-code-optimizer/skills/cc-optimizer-analyze

Low-level analyzer: aggregate THIS project's Claude Code transcripts into token/model/cache statistics (zero model tokens, pure arithmetic). Read-only and project-scoped.From its SKILL.md

Install

npx -y skills add Korck-lab/claude-code-optimizer --skill cc-optimizer-analyze

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.3 KB, as published. Nobody here has run it

/cc-optimizer:analyze

Aggregate this project's session transcripts into statistics, without spending any model tokens. Read-only — it never writes config anywhere.

Scope

Only the current project's transcripts (its root plus its git worktrees) are opened. The machine-wide aggregate is an explicit opt-in and should only be run when the user asks for it.

How to run it

python3 "${CLAUDE_PLUGIN_ROOT}/optimizer/analyze.py" \
  "$HOME/.claude/projects" "$PWD/.claude/cc-optimizer/raw-stats.json" \
  --project-root "$PWD"

Machine-wide (only on explicit request — drop --project-root):

python3 "${CLAUDE_PLUGIN_ROOT}/optimizer/analyze.py" \
  "$HOME/.claude/projects" "$PWD/.claude/cc-optimizer/raw-stats-all.json"

Most of the time /cc-optimizer:run is what you want — it runs this step plus costing, recommendations and the report.

Output: raw-stats.json

{
  "scope": {
    "mode": "project",
    "project_root": "/Users/me/Documents/sources/my-app",
    "slugs": ["-Users-me-Documents-sources-my-app"]
  },
  "projects": {
    "-Users-me-Documents-sources-my-app": {
      "sessions": 150,
      "total_tokens": 8048909762,
      "input_tokens": 156819318,
      "output_tokens": 257451667,
      "cache_creation_input_tokens": 2150345336,
      "cache_read_input_tokens": 30117123696,
      "cache_hit_ratio": 0.9288,
      "by_model": { "claude-opus-4-8": { "msgs": 5000, "input_tokens": 1801208 } }
    }
  }
}

scope states exactly which transcript directories were read — quote it if the user asks what the analysis looked at.

What's measured (from the transcripts)

Total tokens by model
Cache creation vs. read (1h vs 5m ephemeral)
Cache hit ratio
Output-size histogram (identifies expensive models doing trivial work)
Subagent vs. main-thread spending
Base context per session (the fixed per-turn overhead: system prompt, tool schemas, CLAUDE.md, MCP)

What's NOT spent

Zero model API calls — pure Python arithmetic over the message.usage fields already recorded in each transcript