The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from various observability sources, which could potentially contain malicious instructions targeting the AI agent.
Ingestion points: The agent reads alert annotations (such as runbook URLs), Prometheus metric labels, and Loki log entries into its context to provide analysis (found in SKILL.md Step 4 and throughout references/alert-investigation-patterns.md).
Boundary markers: There are no explicit instructions to wrap fetched runbook content or log data in delimiters or to ignore embedded instructions within that data.
Capability inventory: The skill has access to the gcx tool for infrastructure queries and uses gh and curl for network operations.
Sanitization: The skill does not specify any validation or sanitization steps for the content of runbooks or logs before processing them.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading external documentation from URLs provided within alert metadata.
Evidence: The file references/alert-investigation-patterns.md includes instructions to fetch runbooks using curl and the GitHub CLI (gh api). While these are standard tools and the primary targets are typically trusted repositories, the URLs are derived dynamically from alert annotations, which is an external data source.
[COMMAND_EXECUTION]: The skill relies on executing shell commands via the gcx, jq, gh, and curl utilities to perform its primary function of alert investigation.
Evidence: Numerous examples in SKILL.md and references/alert-investigation-patterns.md demonstrate the use of these tools for listing rules, querying metrics, and fetching logs.

investigate-alert