The Agent Skills Directory

[METADATA_POISONING]: The file skill-report.json includes a security_audit section that explicitly claims the skill is safe and dismisses hypothetical security findings as false positives. This represents a self-referential attempt to influence or bypass automated security scanning (Indirect Prompt Injection vector 8e).
[INDIRECT_PROMPT_INJECTION]: The skill defines prompt templates that interpolate external user data (such as role and project details) without security boundaries, which could be used to influence the agent's behavior. * Ingestion points: prompt_templates in skill-report.json. * Boundary markers: No delimiters or instructions to ignore embedded commands are present in the templates. * Capability inventory: The skill manifest in SKILL.md allows the use of Write, Read, Grep, and Glob tools, enabling file system interaction if an injection occurs. * Sanitization: No sanitization or validation of the interpolated variables is defined.

requirements-elicitation