← ClaudeAtlas

prolog-verificationlisted

prolog, verification, evidence, invariants, worktrees, brave, hooks
lost-rob0t/skills · ★ 0 · AI & Automation · score 69
Install: claude install-skill lost-rob0t/skills
# Prolog verification ## Goal Verify agent changes against explicit facts and invariants stored under the current worktree's `.prolog/` directory. ## Required state - `.prolog/facts.kb` contains task/control facts and current repository state. - `.prolog/runs/run-<HEAD>.pl` contains machine-recorded observations and Brave evidence for that Git HEAD. - `.prolog/verify.pl` derives completion and owns task-specific PlUnit tests. - `.prolog/result.json` is overwritten by the gate with the checked repository state and result. Use the installed `prolog-verify` command when available. Otherwise run this skill's `scripts/prolog-verify.py` with Python 3. ## Workflow 1. Run `prolog-verify init --task <short-task-id>` in the worktree before recording evidence. 2. Add task-specific requirements and derived invariants to the canonical verification files. Never assert `verified(true)` or equivalent self-certifying facts. 3. Run real tests through `prolog-verify observe -- <command> [args...]`. The helper records the command, exit status, output digest, Git HEAD, and worktree digest in `run-<HEAD>.pl`. 4. When external discovery is required, run `prolog-verify brave --query <query>` so the fixed Brave CLI call and its result are recorded in the same run file. Use `record-brave` only when a compatible Brave tool already wrote its successful result to a file. Do not mark local-only work as research. 5. Run `prolog-verify check`. Treat a missing, stale, timed-out, exceptional, or non-ze