The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains shell scripts that use curl for API connectivity testing and kubectl for retrieving service logs. These are standard diagnostic tools for incident response and are explicitly allowed in the skill configuration.
[DATA_EXFILTRATION]: The triage and authentication verification scripts communicate with https://api.exa.ai/search. This is the official endpoint for the Exa service, and the use of the EXA_API_KEY is consistent with the skill's purpose of managing Exa-related outages.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources during triage.
Ingestion points: Processes the response from the Exa API (saved to /tmp/exa-triage.json) and output from kubectl logs commands.
Boundary markers: The instructions explicitly remind the operator to use "redacted diagnostic access" and "preserve minimal redacted evidence," which limits the flow of raw, potentially malicious external data into the agent context.
Capability inventory: Shell execution (Bash) and file reading.
Sanitization: API responses are piped through python3 -m json.tool for structural validation and grep is used to filter log output, reducing the surface area for injection attacks.

exa-incident-runbook