mentor

# Mentor — /insights-style report on how you work with agents ## Why this exists Agents now do what juniors used to. But juniors became seniors by getting feedback from seniors. Give agents to juniors and fire the seniors, and nobody gets that feedback loop anymore. This skill *is* that loop: it reads your own local session history and hands back a single report — what you build, where you lose time, and concrete changes to get better. Effectiveness, not token-golf. **The deliverable is the rendered `report.html` file — always.** The eight-section analysis is intermediate: it goes into `analysis.json` and gets rendered, never pasted into chat. A wall of analysis text in the reply is the failure this skill exists to fix. If you cannot run the scripts, say so plainly — do not substitute a text summary. Modeled on Claude Code's built-in `/insights`. ## Agents | Agent | arg | Logs | |---|---|---| | Claude Code | `claude` | `~/.claude/projects/<slug>/<uuid>.jsonl` | | OpenAI Codex | `codex` | `~/.codex/sessions/**/rollout-*.jsonl` (+ `~/.codex/archived_sessions/`) | | both | `both` (default) | union of the two | ## Workflow — build the report 1. **Collect** every session in the window into one stats bundle (deterministic): ```bash uv run --no-project python ~/.claude/skills/mentor/scripts/collect.py \ --agent <claude|codex|both> --days <N> --out /tmp/mentor-stats.json ``` Default `--agent both --days 30`. It writes `stats.json` and prints a compact digest...