The Agent Skills Directory

[PROMPT_INJECTION] (MEDIUM): Indirect Prompt Injection via source code comments. The skill is highly vulnerable to instructions embedded in the data it processes. * Ingestion points: The skill reads external source code files (e.g., src/utils/math.ts) to perform audits. * Boundary markers: No delimiters or 'ignore embedded instruction' warnings are present to distinguish code comments from agent instructions. * Capability inventory: The skill possesses file-read capabilities and generates structured output based on untrusted content. * Sanitization: There is no evidence of sanitization or filtering for the content within JSDoc blocks or comments.
[COMMAND_EXECUTION] (LOW): The skill documentation in README.md describes logic to 'Verify Current Structure' and 'Load Documentation References' with 'fallback discovery'. This implies the agent is instructed to perform recursive file system navigation and read operations on the local environment to find configuration files like SKILL.md.

accelint-ts-documentation