record-runlisted

# record-run Every number the human-prompting corpus reports today was produced by people who wrote the standard - its own README names this as the corpus's weakest point. The only fix is real adopters' real sessions, and nobody is going to reproduce a run by hand afterward to send it in. So this skill does not ask for that: it assembles what already happened, in the tool the person just used, and asks for one yes or no. A "no" costs the user nothing - the assembled file stays local and nothing is sent. That asymmetry is the entire design. Like every other lifecycle skill, it ships into the adopted repo and runs there: the trigger it serves - the close of an align session - fires wherever that session runs, which is the repo being aligned and not the standards repo (ADR-045). It is not that repo's own tooling, and an align plan does not get to drop it on that reading. **A failed or aborted run is more valuable evidence than a clean one, and this must be said out loud before anything is asked** - a skill that only feels natural to offer after success will only ever collect successes, and the corpus already knows what those look like. ## Steps 1. **When this fires.** At the close of an `align-to-standards` session (wired in at that skill's own step 8) - success, partial, or abandoned mid-run all count. It also runs by hand against any past session that used a shipped skill. It does **not** fire when the intake answered `adopt.evidence` with **send nothing** - say