The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and adhere to policies defined in repository-local files such as .agents/shared/worktree-policy.json, AGENTS.md, CLAUDE.md, and CODEX.md. This represents a surface for indirect prompt injection where an attacker could influence agent behavior or command selection through these files.
Ingestion points: Reads repository-level configuration files including .agents/shared/worktree-policy.json and various agent-specific markdown files.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands within these files are provided.
Capability inventory: The skill utilizes Git CLI tools and can execute shell commands defined in the publication_commands section of the ingested policy files.
Sanitization: The instructions do not specify any sanitization or validation of the content within the ingested policy files beyond checking for malformed JSON.
[COMMAND_EXECUTION]: The skill makes extensive use of standard Git commands (e.g., git status, git diff, git worktree add, git fetch) to inspect the environment, manage worktrees, and verify project state.

worktree-hygiene