The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a process that ingests and analyzes potentially untrusted data from multiple sources, creating a surface for indirect prompt injection.\n
Ingestion points: According to SKILL.md, the agent is instructed to read requirements documents, concept and synchronization specifications, product requirement documents (PRD), and 'staged links' to external content.\n
Boundary markers: The instructions do not define specific delimiters or instructions for the agent to ignore potentially malicious content embedded within the audited files.\n
Capability inventory: The skill is strictly read-only and focused on report generation; it does not explicitly invoke network-sending, file-writing, or code-execution tools.\n
Sanitization: No data sanitization or validation mechanisms are described in the auditing workflow.\n- [EXTERNAL_DOWNLOADS]: The skill references several external resources and academic papers to support its methodology.\n
Evidence: references/sources.md contains links to essenceofsoftware.com, mit.edu, arxiv.org, and github.com/jlifyio/wyx. These links are provided for informational purposes and methodology reference.

concept-audit