Testing rules
Check a rule without starting a Claude Code session.
explain
unbound-paws explain 'gh pr create --fill' # a Bash commandunbound-paws explain --tool Edit src/app/page.tsx # a file editunbound-paws explain --tool mcp__slack__send_message '{}'unbound-paws explain --event session-start compact # a session eventecho '<hook event json>' | unbound-paws explain - # a full hook eventIt prints the rules in scope, marks the ones that apply with →, and shows the decision. It writes nothing to the audit log. With --tool, the text goes in the field that tool's rules match against:
| --tool | Text fills |
|---|---|
| Bash (the default) | command |
| Read, Write, Edit | file_path |
| NotebookEdit | notebook_path |
| Grep, Glob | pattern |
| WebFetch | url |
| WebSearch | query |
| Task, Agent | prompt |
event : PreToolUsetool : Bashinput : gh pr create --fill.paws : /Users/you/.pawsrules: humanizer-commit event=pre-commit action=require-skill→ humanizer-pr event=pre-tool action=require-skilldecision:DENY — This action requires the "humanizer" skill. Invoke it with the Skill tool, then retry this exact command.explain has no transcript. A skill never counts as loaded, agent-count never holds and no-edit-matching always does. A line starting with ! is a problem to fix, such as a skipped file, an unknown key, a digest it couldn't read, or two rules with the same name. Test a Bash rule against the command and against the same command inside echo '...'. Only the first should show →.
validate
unbound-paws validate [root] checks a checkout's catalog/ rules and packs/ manifests strictly, including unknown keys and pinned hashes. It's for catalog contributors. For your own rules, use explain.