The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to audit and verify other agent skills, which involves reading and processing potentially untrusted Markdown and script files.
Ingestion points: workflows/audit-skill.md and workflows/verify-skill.md instruct the agent to read the entrypoint and all supporting resources of a target skill directory.
Boundary markers: The skill includes guidance on clear instructions and validation in references/be-clear-and-direct.md, but it does not implement specific delimiters for the external data it processes during audits.
Capability inventory: The skill has the capability to write files (scripts/init_skill.py), package files (scripts/package_skill.py), and execute scripts via uv run.
Sanitization: scripts/quick_validate.py uses yaml.safe_load() for parsing frontmatter, and scripts/package_skill.py rejects symbolic links during packaging to prevent path traversal or unauthorized data leakage.
[PRIVILEGE_ESCALATION]: The initializer script programmatically modifies file permissions to make generated scripts executable.
Evidence: scripts/init_skill.py uses chmod(0o755) on the example.py script created during skill initialization. This is an intended function of the skill's primary purpose to scaffold new executable resources.
[COMMAND_EXECUTION]: The skill provides instructions for executing local Python scripts to perform management tasks.
Evidence: SKILL.md contains bash blocks for running init_skill.py, quick_validate.py, and package_skill.py using the uv run command.
[EXTERNAL_DOWNLOADS]: The management scripts utilize a well-known library for YAML processing.
Evidence: scripts/package_skill.py and scripts/quick_validate.py specify PyYAML==6.0.3 in their PEP 723 dependency blocks.

create-agent-skills