The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill performs legitimate QA testing activities using tools explicitly defined in its configuration.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from external websites via browser snapshots and console logs. 1. Ingestion points: agent-browser snapshot and agent-browser console commands in SKILL.md. 2. Boundary markers: No explicit delimiters or instructions to ignore website-provided text are present. 3. Capability inventory: The agent has access to Bash (specifically agent-browser, mkdir, and cp) allowing it to interact with the web and manage local files. 4. Sanitization: No evidence of data sanitization or validation of the ingested website content was found. While the surface exists, the risk is inherent to the skill's primary purpose and no active exploitation was detected.

dogfood