The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's automated fix-and-commit loop (commit.md) and conflict resolution workflow (rebase.md) ingest untrusted data from external sources to influence agent actions.
Ingestion points: In references/workflows/commit.md, the skill reads failing CI job logs via gh run watch or glab ci status. In references/workflows/rebase.md, the skill reads conflict markers directly from project files.
Boundary markers: Absent. The skill does not utilize specific delimiters or instructions to isolate the external logs or conflict data from the agent's primary instructions.
Capability inventory: The skill has the capability to modify local files, execute git commit, git push, and run project-defined task recipes (make, just, npm).
Sanitization: Absent. The workflow relies on the agent's internal reasoning to interpret the untrusted logs and markers without pre-processing or validation.
[COMMAND_EXECUTION]: The skill is designed to discover and execute project-specific task runners (e.g., mise, just, make, npm, pnpm) to perform formatting, linting, and testing during the commit and rebase workflows. While these are standard developer tools, they execute scripts defined within the local repository.
[PRIVILEGE_ESCALATION]: The hooks.md workflow uses chmod +x to enable a git hook. This is the intended behavior for the skill's attribution-stripping feature, but it involves modifying file permissions to allow execution of a script provided within the skill package.

git