The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external files (agents, skills, commands) belonging to a plugin suite. These files represent untrusted data that could contain hidden instructions or malicious prompts designed to trick the subagents into providing false usability scores or misleading recommendations.
Ingestion points: The skill uses the Read, Glob, and Grep tools to load content from files located in [plugin-path]/agents/*.md, [plugin-path]/skills/*/SKILL.md, and [plugin-path]/commands/*.md.
Boundary markers: The subagent prompts provided in the skill instructions do not utilize delimiters (like triple backticks or XML tags) or explicit instructions to ignore potentially conflicting commands found within the analyzed files.
Capability inventory: The skill has the ability to read arbitrary files within the plugin path and generate actionable tasks via the TodoWrite tool, although the latter requires user confirmation through AskUserQuestion.
Sanitization: There is no evidence of content sanitization or validation performed on the text extracted from plugin files before it is processed by the LLM subagents.

suite-audit