Cc User Autopsy

cc-user-autopsy

An honest, evidence-traceable peer-review for Claude Code users — a skill that analyzes your own `~/.claude/projects/` transcripts and `~/.claude/usage-data/` session metadata, then generates a standalone HTML report combining rule-based scoring with an LLM-written personalized review.

Built because the built-in `/insights` is useful but leans celebratory; this skill leans diagnostic.

Preview

Two sample outputs, both generated from synthetic data (no real user's information):

• `assets/example-output.html` — default self-audit layout (diagnostic letter)
• `assets/example-output-hr.html` — hiring-manager / portfolio layout (--audience hr)

To regenerate them locally, see [Running manually](#running-manually).

Three output modes

The same pipeline can produce three variants depending on who the report is for:

Mode	Flag	Audience	Shows project names	Shows session IDs	Shows profile card
Self audit	(default)	You	Verbatim	Verbatim	No
HR / portfolio	--audience hr + --profile + --public-projects	Recruiters	Allowlisted verbatim; rest bucketed	Hidden	Full letterhead
Showcase	--audience hr + empty public_projects: []	Public readers (blog, demo)	All bucketed into category labels	Hidden	Minimal signature

The skill's Step 0 asks which version(s) to build and walks you through the setup files before running anything. It never silently produces the HR or showcase variant.

What it produces

A standalone HTML file (no remote fonts, no CDN scripts) with:

• Overview — session count, tokens, active days, API-equivalent cost, Task agent / MCP adoption, favorite model, cache hit ratio
• 9-dimension rule-based scores (1–10) with explicit metric evidence:
  1. D1 Delegation — Task agent adoption and good-outcome rate
  2. D2 Root-cause debugging — iterative_refine