The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill directs the agent to ingest and analyze untrusted data from external targets (such as HTTP response headers, website bodies, and JavaScript files). This creates an attack surface where a malicious target could attempt to influence the agent's behavior via content injection.
Ingestion points: Analysis of target response headers, URL patterns, and JavaScript patterns located in SKILL.md.
Boundary markers: The skill advocates for 'Marker Discipline' using unique random alphanumeric strings and 'Body-Diff Rule' for differential analysis, which serve as logical validation steps for the researcher but do not act as hard boundaries for the agent's processing.
Capability inventory: The skill documentation includes commands for curl, ruby, dig, and grep (SKILL.md).
Sanitization: There is no evidence of programmatic sanitization or escaping of the external content before the agent processes it.
[DYNAMIC_EXECUTION]: The skill provides examples of executing short, inline scripts using ruby -e and python3 -c to perform timing-based detection and test for Regular Expression Denial of Service (ReDoS) vulnerabilities. These are low-risk executions of static test strings from known templates used for local environment validation.

hunt-misc