The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill handles production incident data and system logs which represent untrusted external input. * Ingestion points: Error logs and incident descriptions processed during production incident management in SKILL.md. * Boundary markers: The skill includes high-level guidance for the agent to treat log data as untrusted text, although it lacks formal programmatic delimiters. * Capability inventory: The skill logic suggests the agent may perform file-write operations for generating post-mortems and coordinate containment strategies like system rollbacks. * Sanitization: The instructions explicitly warn the agent to ignore instructions embedded in logs and recommend favoring containment over writing new code in production.

incident-response