The Agent Skills Directory

[PROMPT_INJECTION]: The skill establishes a workflow that processes untrusted external data via the web_fetch tool. This creates a surface for indirect prompt injection where instructions hidden on external websites could potentially influence the agent's behavior.
Ingestion points: Untrusted data is retrieved from external websites using web_fetch during the deep dive phase (SKILL.md).
Boundary markers: The skill does not provide instructions for using delimiters or negative constraints to isolate external content from system instructions.
Capability inventory: The agent utilizes web_fetch and potentially other specialized research tools to interact with the public internet.
Sanitization: The methodology lacks explicit steps for filtering or sanitizing fetched content before it is synthesized into research findings.
[NO_CODE]: The skill is composed entirely of markdown instructions and YAML configuration. It does not contain executable code, scripts, or package dependencies, which limits the risk of direct execution attacks.

deep-research