The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it reads metadata (such as name, author, and version) and audit findings from user-supplied VRM files and incorporates them into the agent's report context.
Ingestion points: Metadata and findings are extracted from processed VRM files within scripts/vrm-tool.mjs and returned to the agent.
Boundary markers: Absent; the audit findings and metadata are interpolated directly into the markdown report structure defined in the SKILL.md instructions.
Capability inventory: The skill allows the agent to write files to the local file system via the optimize command in scripts/vrm-tool.mjs.
Sanitization: The script treats metadata as literal strings and does not appear to perform sanitization or filtering before the data is presented to the agent.

avatar-check