The Agent Skills Directory

Execution of Shell Commands: The skill and its associated validation scripts execute git and python3 commands to manage implementation workflows. This involves the use of subprocess calls to inspect repository state, manage worktrees, and verify commits. These operations are standard for development automation tools and are implemented using best practices, such as passing commands as argument lists to avoid shell injection vulnerabilities.
Data Ingestion Surface: The skill processes information from Git repositories, including pull request metadata, commit histories, and repository-relative paths. This creates a potential surface for indirect prompt injection if external repository data contains instructions designed to influence the agent's behavior. The skill mitigates this by defining strict implementation boundaries, requiring explicit user authorization for certain actions, and incorporating human review checkpoints before final commits.
Secure File Handling: The provided validation script includes defensive programming checks, such as verifying that manifest files are regular files rather than special device files or FIFOs, and ensuring that all paths are normalized and repository-relative. These measures help prevent the skill from interacting with unintended or potentially malicious file system structures.

implementation-kickoff