The Agent Skills Directory

[PROMPT_INJECTION]: The skill interacts with an indirect prompt injection surface by fetching external data from GitLab.
Ingestion points: Data enters the agent context through glab ci trace (fetching job logs) and glab ci view (fetching pipeline metadata).
Boundary markers: The author has included a clear 'Security Note' in SKILL.md and explicit 'BEGIN/END EXTERNAL CONTENT' markers in scripts/ci-debug.sh to prevent the agent from interpreting fetched data as instructions.
Capability inventory: Capabilities are limited to glab CLI operations for pipeline and job management.
Sanitization: The debug script uses tail to present truncated log data and correctly identifies fetched content as untrusted data.
[EXTERNAL_DOWNLOADS]: The CI templates reference official, well-known container images from Docker Hub, including node:20-alpine, docker:24-cli, aquasec/trivy, and bitnami/kubectl. These are industry-standard tools from established providers.
[COMMAND_EXECUTION]: The provided utility scripts/ci-debug.sh executes the local glab binary to facilitate pipeline debugging. The script uses secure shell practices, such as quoting variables and using structured JSON queries to parse job information.
[DATA_EXFILTRATION]: No evidence of hardcoded credentials or data exfiltration was detected. The skill guides users toward secure practices, such as using GitLab's masked and protected variables for sensitive tokens.

glab-ci