The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies and documents an attack surface where untrusted content from GitLab (such as job logs, commit messages, or issue bodies) could contain malicious instructions designed to influence the agent.
Ingestion points: The scripts/ci-debug.sh script and various debugging workflows in SKILL.md retrieve external data using glab ci trace and glab ci view.
Boundary markers: The skill provides clear instructions in SKILL.md under a "Security Note: Untrusted Content" header and includes warning comments in scripts/ci-debug.sh to treat fetched output as data only.
Capability inventory: The skill possesses the ability to manage pipelines, retry jobs, and download artifacts, which could be abused if the agent followed instructions embedded in logs.
Sanitization: The skill relies on providing explicit guidance to the agent to disregard embedded instructions rather than using programmatic filtering or sanitization of the log content.

glab-ci