The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze Zsh scripts, plugin code, and terminal diagnostics which may originate from untrusted sources (e.g., public repositories or attacker-controlled environments).
Ingestion points: Multiple reference files (references/13-completion-authoring.md, references/19-security.md) describe processing repository branch names, command strings, and directory-local files (.env, .zshrc).
Boundary markers: The skill explicitly instructs the agent in SKILL.md and references/19-security.md to identify evaluation boundaries and validate inputs, though these are guidelines for the code the agent produces rather than strict boundaries for the agent's own context.
Capability inventory: The skill utilizes shell command execution via scripts/verify-zsh.zsh and describes the use of subprocesses, file writes, and network operations (references/10-async-concurrency-and-jobs.md, references/19-security.md).
Sanitization: The instructions emphasize sanitizing terminal escapes and percent signs in prompts (references/15-prompts-vcs-and-terminal.md) and blocking sensitive environment variables (references/17-history-directories-and-hooks.md).

zsh-expert