The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the analysis and review of external agent skills, which involves reading untrusted content from SKILL.md and reference files.
Ingestion points: The scripts/validate.py script reads the content of SKILL.md from the target directory provided by the user. The skill's main instructions also guide the agent to qualitatively review these external files.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore potentially malicious embedded commands within the analyzed skill data.
Capability inventory: The skill can create new directory structures and template files via scripts/init-skill.sh and read/validate files via scripts/validate.py.
Sanitization: The validation script scripts/validate.py includes a security check that resolves file paths and ensures they do not attempt path traversal outside of the skill's root directory.
[COMMAND_EXECUTION]: The skill requires the execution of local shell and Python scripts as part of the intended development workflow.
Evidence: instructions in SKILL.md for running bash scripts/init-skill.sh and python3 scripts/validate.py.
The init-skill.sh script performs strict regex validation (^[a-z0-9]([a-z0-9-]*[a-z0-9])?$) on the user-supplied skill name to prevent command injection during directory creation or file manipulation.

skill-creator