The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that ingests data from external, untrusted web servers, creating an attack surface for indirect prompt injection.
Ingestion points: The scripts/fetch_page.py script retrieves raw HTML content, page titles, and HTTP headers from external URLs provided by the user or discovered through scanning.
Boundary markers: Absent. The instructions in SKILL.md do not specify the use of delimiters or provide warnings for the agent to disregard potential instructions embedded in the external content fetched during Step 3.
Capability inventory: The agent is configured to use the nmap utility through the Bash tool for network discovery and has the capability to execute the local scripts/fetch_page.py script. These tools provide the agent with active network reconnaissance capabilities.
Sanitization: Absent. While scripts/fetch_page.py uses regular expressions to extract specific metadata (such as CMS versions and frameworks) and performs basic HTML unescaping, it does not sanitize the text for malicious prompt instructions before returning it to the agent's context.

web-fingerprint