The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from external, untrusted URLs, which creates a vulnerability where malicious instructions embedded in a webpage could influence the agent's output or actions.\n
Ingestion points: The web-fetch integration is used to retrieve content from arbitrary URLs provided by the user (described in SKILL.md).\n
Boundary markers: The instructions in SKILL.md lack explicit delimiters or warnings to the LLM to ignore embedded commands within the fetched content during the summary and insight generation phases.\n
Capability inventory: The skill has the capability to write files to various locations in the local workspace, such as 00-inbox/, 04-projects/, and 05-knowledge/booklets/ (defined in SKILL.md Step 5).\n
Sanitization: There is no mention of sanitization, escaping, or filtering of the fetched content in SKILL.md before it is processed by the agent.

url-dump