The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from research URLs (via the defuddle tool and BrowserOS), which creates a surface for indirect prompt injection attacks where malicious instructions could be embedded in the retrieved content.
Ingestion points: Research URLs and web content processed via defuddle and BrowserOS.
Boundary markers: The skill does not define specific prompt delimiters for external data, but it includes a high-level instruction to treat external data as hostile.
Capability inventory: The agent has the capability to execute terminal commands via the ExecutionProxy (rtk prefix) and perform file system writes within the Obsidian vault.
Sanitization: The skill provides defensive instructions to redact secrets and PII before sharing context and to treat web content as hostile.
[COMMAND_EXECUTION]: The skill facilitates the execution of terminal commands through an ExecutionProxy interface (using the rtk prefix, e.g., rtk npm test). While this is part of a managed TDD workflow, it represents a capability to interact with the host system's shell.

obsidian-architect