The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it is designed to ingest and analyze the content of other skills' SKILL.md files to formulate project conventions.
Ingestion points: The skill reads SKILL.md files from the local catalog and project-specific locations, as well as general project files like README and scripts.
Boundary markers: The instructions do not define boundary markers or explicit 'ignore' instructions for the content being read from external skills.
Capability inventory: The skill can modify project configuration files (AGENTS.md and CLAUDE.md) and can execute or recommend command-line installation of skills using npx skills add.
Sanitization: There is no mention of sanitizing or validating instructions extracted from external SKILL.md files before incorporating them into the project's own instructions.
[EXTERNAL_DOWNLOADS]: The skill's documentation includes instructions for downloading and installing additional components from the author's own repository.
Evidence: references/project-setup.md contains commands such as npx skills add aymericderbois/skills and npx skills add ~/projects/github.com/aymericderbois/skills to expand tool capabilities.
[COMMAND_EXECUTION]: The skill facilitates the execution of system commands for environment configuration.
Evidence: SKILL.md specifies that the agent should install missing skills if authorized, using commands described in the project setup guide.

flow-setup