The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes diagnostic bash scripts using curl to interact with Glean API endpoints. These commands are localized to the user's specific Glean domain and are used exclusively for health checks and status retrieval during incident response.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes external data from Glean API responses using jq. However, the skill incorporates significant risk mitigations:
Ingestion points: Data enters the context via the diagnostic curl commands in SKILL.md targeting the /search, /getdatasourceconfig, and /getstatus endpoints.
Boundary markers: The instructions explicitly command the agent to "never paste tokens, queries, document titles, or result snippets into incident chat" and to "preserve only error codes and aggregate counts."
Capability inventory: The skill uses Bash(curl:*), Grep, and Read/Write/Edit tools.
Sanitization: The skill mandates the use of "synthetic probe" queries that contain no real employee content and requires redacting all sensitive identifiers and raw content from the final output receipts.

glean-incident-runbook