The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to its core orchestration logic.
Ingestion points: It processes user-provided strings and external files like PLAN.md (documented in SKILL.md) to define agent tasks.
Boundary markers: There are no instructions or templates in the skill to use delimiters or protective guardrails when interpolating external content into prompts for sub-agents.
Capability inventory: The orchestrator can execute git commands and trigger recursive agent actions across multiple code branches.
Sanitization: No sanitization or escaping of interpolated content (such as ${lastMessage} in references/spec.md) is performed before task delegation.
[COMMAND_EXECUTION]: The skill utilizes local shell commands for its primary orchestration logic.
Evidence: Instructions in SKILL.md and references/spec.md detail the use of git diff for automated reviews and git worktree for managing parallel execution flows for the 'race' or 'vs' operators.
[EXTERNAL_DOWNLOADS]: The documentation references an external command-line interface for extended functionality.
Evidence: The README.md and SKILL.md suggest the optional installation of the @let-it-cook/cli package via NPM for standalone terminal or CI usage.

cook