The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Git repositories (such as file names, status, and diffs) through its helper scripts (scripts/git-maker-fast.mjs, scripts/repo-status.mjs). This content is interpolated into agent prompts for drafting commit messages and resolving conflicts without clear boundary markers or sanitization.
Ingestion points: Git status, file lists, and diff outputs are read into the agent context in rules/agent-parallelism.md and rules/commit-and-push-policy.md.
Boundary markers: Absent; instructions direct the agent to paste status and diff summaries directly into the prompt.
Capability inventory: The skill possesses the ability to perform git commit and git push via scripts/git-commit.mjs and scripts/git-maker-fast.mjs.
Sanitization: No specific sanitization or escaping of repository-sourced content is described before it is processed by the LLM.
[COMMAND_EXECUTION]: The skill relies on several helper scripts that use Bun.spawn to execute Git commands on the local system. While these operations are scoped to standard Git functionality, they represent a broad execution surface triggered by repository state.
Evidence: scripts/git-commit.mjs, scripts/git-maker-fast.mjs, scripts/git-push.mjs, scripts/repo-discover.mjs, and scripts/repo-status.mjs all use Bun.spawn to interact with the Git CLI.

git-maker