The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's exploration phase involves reading untrusted data from the repository, including git remote output, git status, and existing project files such as CONTEXT.md and ADRs. This data is used to determine the scaffolding logic, creating a surface where malicious content in these files could influence the agent's configuration decisions. * Ingestion Points: git remote -v, git status, AGENTS.md, CONTEXT.md, and ADR files. * Boundary Markers: The instructions do not specify explicit delimiters or 'ignore instructions' warnings when reading these files. * Capability Inventory: The skill can execute git add/commit/push, write to the local filesystem (AGENTS.md, docs/agents/, .gitignore), and use the gg CLI to interact with GitHub/GitLab APIs. * Sanitization: No explicit sanitization or filtering of the content read from the repository is performed.
[COMMAND_EXECUTION]: The skill instructions require the agent to execute several shell commands to inspect and modify the repository state. * Evidence: Use of git remote -v, git status --short, git add, git commit, and git push. * Evidence: Extensive use of a specialized gg CLI tool for GitHub and GitLab operations, including gg api for raw API calls and various issue/PR management commands.

vibe-init