The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository environment, including git logs, file names, and CI check statuses, to generate commit messages and Pull Request descriptions. This creates an indirect prompt injection surface where malicious repository metadata could attempt to influence the agent's behavior or downstream review processes.
Ingestion points: Git diffs and status in validate.sh and commit.sh, git logs in create-pr.sh, and GitHub check outputs in verify.sh.
Boundary markers: The Pull Request body template in create-pr.sh lacks explicit delimiters or instructions for the agent to ignore embedded commands within interpolated content.
Capability inventory: The skill possesses capabilities to execute shell commands (git, gh) and modify repository state.
Sanitization: The implementation relies on shell quoting but does not perform explicit sanitization or filtering of content interpolated into the PR body.
[COMMAND_EXECUTION]: The skill executes a local script as part of its validation phase.
Evidence: validate.sh invokes ./scripts/quality_gate.sh. This execution relies on the existence and security of the external script file within the repository context.

atomic-commit