The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection (Category 8) by processing untrusted data generated during CI processes.
Ingestion points: The agent reads build logs (build-log.txt), job metadata (finished.json), and test results (junit_*.xml) from external storage buckets.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the content of these logs as untrusted data or to ignore instructions embedded within them.
Capability inventory: The skill utilizes potentially powerful capabilities including shell command execution (gh, gcloud), local file system writes (mktemp, gcloud storage cp), and network operations (WebFetch).
Sanitization: No sanitization or validation is performed on the content of the logs or artifacts before they are processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill fetches configuration, logs, and metadata from remote sources.
Evidence: Downloads artifacts from storage.googleapis.com/test-platform-results/ and gcsweb-ci.apps.ci.l2s4.p1.openshiftapps.com/gcs/. These domains represent the official CI infrastructure for the OpenShift project.
[COMMAND_EXECUTION]: The skill relies on shell commands to interact with external services and manage local data.
Evidence: Uses the gh CLI for GitHub API interactions and gcloud storage or gsutil for downloading large log files and directories to temporary local storage.

investigate-ci-failure