The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local project environment, including directory structures and the contents of configuration files (e.g., package.json, pyproject.toml), to generate documentation. While this creates an indirect injection surface where a malicious project could influence the AI's behavior via generated instructions, the skill is focused on standard documentation tasks and does not grant excessive privileges.
Ingestion points: SKILL.md describes scanning the project root and reading engineering config files.
Boundary markers: None explicitly used for interpolating project metadata into the AGENTS.md template.
Capability inventory: File system write access (AGENTS.md, AGENTS.LOCAL.md, .gitignore) and invocation of the yy-detect-terminal skill.
Sanitization: No explicit sanitization of project metadata is mentioned before writing to documentation.
[COMMAND_EXECUTION]: The skill automates the identification of terminal capabilities by triggering an auxiliary skill, yy-detect-terminal. This is a functional requirement for environment configuration and is managed within the agent's tool execution framework.
[SAFE]: The skill body contains explicit security-focused instructions. It mandates the inclusion of a rule in the generated documentation that forbids the use of compound shell operators (e.g., &&, ||, ;) to prevent potential command execution in unintended directories (mitigating "naked repository" style attacks). It also prioritizes using a Node.js process over shell wrappers for reading configuration files to reduce command injection surface.

yy-create-agents