hillclimb-verify

name

hillclimb-verify

description

Verify the most recent execution run in a hill-climbing project. Reads `.hillclimb/state.html`, runs the project's verifier script, and updates the dashboard with the result. Use this skill whenever the user finishes an `hillclimb-execute` and wants to know if it improved things. Trigger on "run the hillclimb-verify skill", "verify the last run", "evaluate", "check the result", or after any `hillclimb-execute` completes. Updates the dashboard's `best`, marks the originating idea done/abandoned/open, and reports back.

hillclimb-verify: evaluate the latest run

See hillclimb-execute skill's First principle. False passes are worse than false fails; when in doubt, mark inconclusive with the raw verifier output as notes, never silently coerce malformed output into "pass."

Step 1: Locate state, read it once

STATE_PY="$PWD/.hillclimb/state.py"; STATE_HTML="$PWD/.hillclimb/state.html"
[ -f "$STATE_PY" ] && [ -f "$STATE_HTML" ] || { echo "no .hillclimb/, run the hillclimb-onboard skill first"; exit 1; }
STATE_JSON=$(python3 "$STATE_PY" read "$STATE_HTML")

Use STATE_JSON for verifier.command, the run's plan/summary/actions, the project objective, and the time limit.

Step 2: Pick the run to verify

Walk from the end. Target the run from the back where AND .