Habits prompt-injection guard

Each tool targets a completely different action: checking text (check_prompt), obtaining credentials (get_free_key), and submitting feedback labels (teach_guard). The descriptions explicitly cross-reference each other (teach_guard needs a key from get_free_key; check_prompt results are what you'd teach), making boundaries unambiguous.

All three names follow a consistent snake_case verb_noun pattern (check_prompt, get_free_key, teach_guard) with clear, concrete verbs. No mixing of conventions or vague generic verbs.

Three tools is at the low end but appropriate for a narrowly scoped prompt-injection guard service, and each tool earns its place. A slightly richer surface (e.g. batch checking or key management) would justify a larger set, but nothing here is filler.

The core lifecycle is covered: detect (check_prompt), onboard/key (get_free_key), and improve the guard (teach_guard). Minor gaps exist—no way to retrieve a past request_id's details, no batch check, and no key rotation/revocation—but agents can work around these.