Claude Code / Codex skill that generates a self-contained HTML report debugging what is in a session's context window and how every token is spent — context budget, retained thinking, the dumb-zone cutoff, loaded CLAUDE.md and skills, and full history.
Install it with:
npx skills add Mark-Life/agent-skills -g --skill session-reportThen run /session-report <session-id> to inspect or debug a session's context usage — "what's eating my context", "why did this session get dumb", "is my CLAUDE.md too big". It produces a forensic, zero-dependency HTML report you can just open: peak context versus the available window, dumb-zone detection (degradation past ~40% utilization), thinking-token recovery from output metadata, an inventory of loaded artifacts (CLAUDE.md, skills, opened files), and the full searchable event history with inline dumb-zone markers.
Because it's a script parsing the transcript rather than a prompt, the report is near-instant and costs no tokens — the agent just runs it, it doesn't reason it out. Runs via Node, Bun, or npx tsx and defaults to a 1M-token window for Claude Code (Codex records its actual window); output lands at ./ccx-<id>.html. Worth remembering one caveat it calls out: a peak near 200K on a 1M window is normal (~20% capacity), not evidence of truncation.