The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external SKILL.md files provided by users. A maliciously crafted skill file could attempt to influence the agent's behavior during the validation process through embedded natural language instructions.
Ingestion points: Untrusted content is read from SKILL.md files identified by the user or found via file globbing.
Boundary markers: The skill does not employ explicit boundary markers or directives to the agent to ignore embedded instructions in the content being analyzed.
Capability inventory: The skill possesses Write, Edit, and multiple Bash tool configurations, which could be misused if an injection is successful.
Sanitization: Validation logic relies on YAML parsing without additional sanitization of the natural language components for adversarial instructions.
[COMMAND_EXECUTION]: The skill uses the Bash tool to run various local scripts and utilities, including a Python-based schema validator and the JRig behavioral evaluation suite.
[EXTERNAL_DOWNLOADS]: Setup instructions for the JRig evaluation tool involve using pnpm install to download dependencies from the public Node package registry.
[PERSISTENCE]: The skill documentation recommends creating a symbolic link to the j-rig binary in a directory on the user's path (~/.local/bin), allowing the tool to persist and be available in future terminal sessions.
[DYNAMIC_EXECUTION]: As part of its static analysis routine, the skill generates and executes a short Python script using python3 -c to extract specific configuration data from the YAML frontmatter of files being validated.

validate-skillmd