The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data such as worker logs, SDK configuration snippets, and Event History JSON files. This creates an attack surface where malicious instructions could be embedded in the data to influence agent behavior.
Ingestion points: The skill explicitly instructs the agent to gather and analyze user-supplied error text, worker logs, SDK connection code (SKILL.md), and Event History JSON files (replay.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when the agent processes this external data, though the skill's philosophy requires explicit user approval for any non-read-only action.
Capability inventory: The agent has access to a wide range of capabilities, including executing temporal and tcld CLI commands (including destructive ones like namespace deletion or batch terminations), and running network diagnostic tools (nc, dig, openssl, nslookup).
Sanitization: The skill does not define specific methods for sanitizing or escaping the content of logs or snippets before they are analyzed or interpolated into operational commands.

temporal-ops