The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document content during its assessment phase without using specific boundary markers or 'ignore embedded instructions' directives. This creates a surface where a document containing malicious instructions could potentially influence the agent's behavior, scoring, or automated fixes.
Ingestion points: SKILL.md logic reads external text via DOCUMENT = read target document.
Boundary markers: Absent; the skill does not specify delimiters to isolate the document content from the instructional prompt.
Capability inventory: The skill instructs the agent to 'Apply auto-fixes directly to document' (found in SKILL.md), which implies a file-writing or modification capability.
Sanitization: Absent; there is no evidence of filtering or escaping external content before interpolation.

document-refinement