The Agent Skills Directory

[PROMPT_INJECTION]: Vulnerability to indirect prompt injection through unverified peer file ingestion.\n
Ingestion points: The 'Turn N>1 flow' in SKILL.md instructs the agent to read all files matching the pattern turn-{N-1}-*.md in a user-provided directory.\n
Boundary markers: Absent. The instructions do not specify the use of delimiters or warnings to treat peer-generated content as untrusted.\n
Capability inventory: The agent uses directory listing (ls) and file read/write operations to coordinate the debate turns.\n
Sanitization: Absent. No logic is provided to sanitize or validate the content of peer files before the agent processes them for its next turn.

multi-debate