review

**documented claims measured against reality:** a scorecard that never flatters you - strictly read-only, and it never flatters the user - punishes hand-wavy conventions and unearned "green ci" claims - produces a graded scorecard saved to file, not a chat reply # Instructions ## Telemetry ```! "${CLAUDE_PLUGIN_ROOT}"/skills/review/review.sh $ARGUMENTS echo "sidecar exit: $?" ``` - `help: requested` → the run was refused before it started; `## Help` below is the whole turn - it already ran, so there is no command to issue - fail (`sidecar exit` > 0) → abort and report: "<raw terminal error>" - success (`sidecar exit` = 0) → continue 1. read the core foundational documents to learn the "claims": - read `README.md`, which always exists and always carries claims worth grading - read `AGENTS.md` IF it exists, for project rules and automation descriptions - host projects ship one, so there it is a required read and its rules are fair game - this repo does not, since `AGENTS.md` is generated and project agnostic here - absent → say so in the scorecard rather than grading claims it never made 2. evaluate the shell telemetry against the documented claims using these dimensions: - **effort vs output:** does the sheer volume of commits/days justify the actual features shipped? does infrastructure/config LOC rival the actual application LOC? - **claim vs reality:** do the docs lie? if `AGENTS.md` claims strict commit types, does the git log reflect that? - **t...