The Agent Skills Directory

[SAFE]: The skill implements a secure, read-only environment for architectural research by limiting system command access to a restricted set of tools (git and ls) and providing explicit instructions against file modification.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to its interaction with external data sources.
Ingestion points: Untrusted data can enter the agent context through the WebFetch tool for external URLs and the Read or Grep tools for local files.
Boundary markers: No specific delimiters or "ignore instructions" directives are present in the prompt to separate fetched content from system instructions.
Capability inventory: The skill has access to restricted Bash commands, WebSearch, and the Agent capability to spawn further tasks.
Sanitization: No sanitization or content validation is performed on retrieved data before it is processed by the model.

ask