The Agent Skills Directory

[DATA_EXFILTRATION]: The skill follows a strict 'redaction-first' approach, explicitly instructing the agent to strip authorization headers, bearer tokens, API keys, and PII from all logs, evidence, and cURL commands before they are recorded or synced to external platforms.
[COMMAND_EXECUTION]: While the skill guides the execution of API, UI, and DB tests, it limits the agent's autonomy by prohibiting the 'invention' of database queries, credentials, or endpoints, and restricts database interaction to safe, read-only verification queries.
[PROMPT_INJECTION]: The skill processes untrusted test case data from external sources (Notion, Plane, spreadsheets). It mitigates associated risks through a 'Core Rule' that prevents the agent from inventing logic or selectors and a mandatory human review loop (OK/NOK) before final results are published or downstream handoffs are triggered.
[SAFE]: The skill uses managed markers (HTML comments) and idempotency keys to ensure that its interactions with Notion and Plane are traceable, preventing accidental data duplication or unauthorized overwrites.

qa-executor