The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically executes commands based on external configuration. In references/ci-checks.md, the agent is instructed to parse run: or script: blocks from GitHub Actions or GitLab CI files and execute them locally "as-is". It also executes npx commands for linting and spelling checks.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) by processing untrusted data from the repository to drive high-capability actions.
Ingestion points: The skill reads .github/workflows/ci.yaml, .gitlab-ci.yml, and .github/PULL_REQUEST_TEMPLATE.md (references/ci-checks.md, references/pull-request-template.md).
Boundary markers: Absent. There are no delimiters or instructions provided to the agent to treat the content of these files as potentially untrusted or to ignore embedded instructions.
Capability inventory: The skill can execute arbitrary shell commands via the CI check translation logic and perform network operations via gh or glab CLI tools (references/pr-cli.md).
Sanitization: Absent. The skill does not validate or sanitize the scripts extracted from CI configuration files before execution.

create-pr