The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's workflow (specifically Step 7) instructs the agent to execute repository-defined verification commands, such as unit tests, linters, or build scripts. This pattern involves the execution of code found within the repository, which is external to the skill's own implementation.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted repository data (e.g., file contents, git diffs) to determine commit scopes and generate commit messages.
Ingestion points: The agent context is populated with output from git status, git diff, and the content of files in the working tree (SKILL.md, Workflow Steps 1 and 3).
Boundary markers: There are no explicit markers or instruction delimiters mentioned in the workflow to separate untrusted repository content from the agent's core instructions.
Capability inventory: The skill has the capability to modify the local file system (via git commit, git add, and git restore) and execute local bash scripts (advance-pipeline-lock.sh) and repository verification commands (Step 7).
Sanitization: The skill relies on the agent performing manual sanity checks (e.g., checking for secrets or accidental logging) rather than automated content sanitization or validation.

commit-changes