The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically engineering_analytics_ci_failures and CI logs (tracebacks, error signatures, and test IDs), which are generated by test suite execution. This content is untrusted as it could be influenced by code changes in a PR. However, the skill only uses this data for analytical querying and reporting within a scoped environment. It does not perform high-risk operations like writing to the file system, spawning shell processes, or making arbitrary network requests based on the ingested content.
Ingestion points: engineering_analytics_ci_failures table (error signatures, test IDs) and CI failure logs via engineering-analytics-run-failure-logs.
Boundary markers: The instructions do not explicitly define delimiters for separating the ingested log data from agent instructions.
Capability inventory: Access to posthog:execute-sql and specific CI analysis tools. No capability for file-write or general network egress.
Sanitization: No explicit sanitization of error tracebacks or test identifiers before processing.
[SAFE]: The skill uses localized SQL templates (HogQL) to query internal warehouse views (engineering_analytics_ci_failures, engineering_analytics_ci_job_history). All network-related patterns are limited to interactions with the vendor's own tools or well-known services like GitHub for linking run IDs, which aligns with the skill's intended purpose of CI triage.

investigating-ci-failures