ccoach

Deep, semantic root-cause coaching for how you work with Claude Code or Codex in a specific project. Goes beyond aggregate metrics — reads your own real code (read-only) and, on spiral-flagged sessions, an opt-in token-bounded content digest — to tell you in plain language WHY work churned and the concrete fix, anchored to official native features. Distinct from the entertainment-flavored ccoach-insight report. Claude Code + Codex (symmetric); read-only, local, desensitized.

Generate an enriched HTML report of your local AI coding usage. By DEFAULT it reports the platform you're running in (Claude Code if invoked from Claude Code, Codex if invoked from Codex); the dual-platform comparison is opt-in (ask to "compare" or for "both"). Tokens & models come from `ccoach report --json` (offline local parse); cost is computed from official online prices the agent looks up per model. Use when the user wants a deeper AI-written review of how they use Claude Code or Codex — cost, token, cache, model and active-hours breakdown, habits analysis, project-by-project recommendations, high-token project/session drilldown, or a richer HTML dashboard than the raw text/JSON reports.