The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes techniques for analyzing external web server responses, which is an inherent part of security auditing.
Ingestion points: Analyzes HTTP headers and response bodies from user-provided target URLs (SKILL.md, references/cache-smuggling-techniques.md).
Boundary markers: No explicit instructions provided for delimiting or ignoring embedded instructions in target responses.
Capability inventory: Utilizes the http_request tool and provides curl command examples to interact with targets.
Sanitization: No specific sanitization or escaping of external content is mentioned, relying on the agent's internal safety logic when processing target output.
[COMMAND_EXECUTION]: The skill includes technical command-line examples for curl and references third-party security tools like h2csmuggler.py and smugglefuzz. These are intended for use by a security auditor performing authorized testing.
[EXTERNAL_DOWNLOADS]: Mentions various open-source security tools (e.g., BishopFox's h2csmuggler). These are standard industry tools for the described activities and do not constitute a direct malicious download or execution within the skill's own operations.

cache-poisoning-smuggling