The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an autonomous diagnostic and remediation loop in Phase 4. It ingests CI build logs via the GitHub CLI and uses the agent's logic to generate and push code fixes to the repository. This represents a potential indirect prompt injection surface where build log content could influence code changes, but it is documented and handled as a core functional requirement of the CI image maintenance task. * Ingestion points: GitHub Actions build logs retrieved during Phase 4. * Boundary markers: The skill instructions focus on functional analysis and do not define specific delimiters to isolate log content from instructions. * Capability inventory: Includes file modification, git commit, and git push operations. * Sanitization: No explicit sanitization of external log data is performed before processing.
[COMMAND_EXECUTION]: The skill executes provided shell scripts to automate version bumping and workflow dispatching. These scripts use tools like sed and gh to manage the repository state and trigger builds. The scripts are local to the skill directory and are consistent with the vendor's intended orchestration.
[EXTERNAL_DOWNLOADS]: The skill verifies binary availability by making HEAD requests to the official Node.js distribution server, which is a well-known and trusted service.

update-ci-images