unpromptedlisted

# Unprompted Reads `~/.claude/projects` and `~/.codex/sessions`, runs a purely statistical analysis, and writes a self-contained HTML report to the user's machine. The insight is comparative, not a score: the same person is measurably a different collaborator on different projects. The report contrasts two poles (project × day), names what is conspicuously absent, and points at moments that contradict each other. On top of the report sits one playful layer: `persona` compresses the same measured rates into a four-letter MBTI-style AI-use type — a parody that says so — and you turn it into a write-up and a bespoke HTML card. ## Run it ```bash node bin/unprompted.mjs scan # pass 1: what's there. No analysis. node bin/unprompted.mjs report # pass 2: analyze + write HTML node bin/unprompted.mjs persona # same corpus, one four-letter parody type ``` `bin/unprompted.mjs` sits next to this file. Node ≥ 20.19, zero dependencies — nothing to install. If a relative path fails, use the absolute skill directory (commonly `~/.claude/skills/unprompted/bin/unprompted.mjs`, or `~/.agents/skills/unprompted/bin/unprompted.mjs` under Codex). ## Workflow 1. **Scan first.** Run `scan` and show the user the project list and counts. They should recognize themselves in the numbers before anything is analyzed — if the row for their main project says 12 utterances, something is wrong with the read, not with them. 2. **Settle the scope before analyzing.** If the user already