The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for processing untrusted external data, which introduces a surface for indirect prompt injection attacks. 1. Ingestion points: External web content and data ingested through the BrowserOS interface as described in the security hygiene section of SKILL.md. 2. Boundary markers: The protocols do not define explicit delimiters or markers to isolate untrusted content from the agent's core instructions. 3. Capability inventory: The framework allows for terminal command execution using the rtk prefix and file writing to the persistent MemoryStore in Obsidian documentation directories. 4. Sanitization: While the skill explicitly mandates redacting secrets and PII before sharing context, it lacks instructions for sanitizing or filtering malicious instructions embedded in the external data.

rapid-prototyper