The Agent Skills Directory

[DYNAMIC_EXECUTION]: During the pre-flight phase, the skill identifies and executes quality gate commands (e.g., npx lefthook, pnpm test, uv run ruff) as defined in the repository's docs/git-workflow.md file. This behavior allows for the execution of arbitrary shell commands based on the local content of the repository.
[INDIRECT_PROMPT_INJECTION]: The skill determines its branching policies, commit formats, and execution logic by reading untrusted repository files such as docs/git-workflow.md, AGENTS.md, and CONTRIBUTING.md. This ingestion surface could be exploited if an attacker provides a repository with malicious configuration files meant to influence the agent's behavior.
Ingestion points: Reads files including docs/git-workflow.md, AGENTS.md, CONTRIBUTING.md, and .github/pull_request_template.md at runtime.
Boundary markers: None identified; external file content is integrated into the agent's task context without explicit isolation.
Capability inventory: Capability to perform file staging, git commits, branch pushes, and Pull Request creation via GitHub CLI (gh).
Sanitization: The skill does not implement validation or sanitization for commands extracted from repository documentation before they are passed to the shell for execution.
[COMMAND_EXECUTION]: The skill frequently invokes Git and GitHub CLI command-line utilities. To mitigate risk, it implements an 'Approval Gate' that requires the user to explicitly confirm a delivery recap before any persistent or remote-facing actions (such as git commit, git push, or gh pr create) are executed.
[DATA_EXFILTRATION]: The skill performs authorized network operations by pushing code and creating Pull Requests on the official GitHub platform. It includes a safety audit that scans for sensitive files, such as environment variables (.env) and private keys (.pem, .key), ensuring they are flagged and excluded from the staging area to prevent accidental credential exposure.

git-ship-acrazie