Agent-Safety-Set

The three skills that run one production fleet's safety layer: pre-send validation, human-approval gates, and boot-context drift detection, priced as the set.

Each skill covers one failure surface, but an agent that writes email, mutates records, and loads its own context files has all three. Gate the sends and an unapproved CRM write still slips through; gate the writes and a corrupted memory file still changes what the agent believes at every boot. The set closes the loop: what leaves the agent is validated, what the agent changes is human-approved, and what the agent becomes is watched. They are also built to interoperate; the integrity skill's re-seed cards carry the approval skill's typed fallback, so approval still lands even when Telegram buttons fail to render.

For operators running agents on real business data who want the safety layer as one decision instead of three. The checks grew out of documented production failures: each package ships its own SECURITY.md disclosure, provenance documentation, and runnable acceptance tests, 63 collected cases across the set. Individually $167.31; as the set, $96.69.

The full audit story behind all three, including the part where the audit caught these packages themselves, is at https://www.tradeplier.com/audit-story/