Tools reference
check_action leads. It's the differentiator: every other tool lets an agent look policy up, this
one intercepts an action before the agent takes it. search_rules/get_rule are the human-approved
retrieval surface it's built on (status == "approved" only, git-native, PR-merged); the last two
compile those same rules into downloadable SKILL.md files (what gnt pull fetches).
check_action
Code
Check a described action against your org's approved rules BEFORE the agent takes it. Returns a
verdict (allowed, blocked, or needs_human) with the governing rule(s) cited and a one-line
reason. Conservative by design: no covering rule, a retrieval failure, or an unclear call all return
needs_human, never a guessed allowed or blocked.
Code
search_rules
Code
Semantic search over this org's approved rules. Only rules with status == "approved" are ever
returned; draft, in-review, rejected, and deprecated rules never reach this tool. Each hit includes a
provenance footer (who approved it, when, and its source citations).
Code
get_rule
Code
Fetch a single approved rule by id, scoped to this org. Refuses (returns an error, not another org's
data) for cross-tenant ids and for anything not in status == "approved".
Code
list_skill_packs
Code
List this org's compiled skill packs, newest first, with each one's id, version, and creation time.
Code
get_skill_pack
Code
Fetch a compiled skill pack's manifest and file list by id: paths and sha256 hashes, not full file content.
Code