The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains multiple shell scripts designed for automated execution. These scripts perform complex operations including cache header detection, testing for unkeyed header reflection, and generating Proof-of-Concept (PoC) files for SameSite cookie bypass.
[DATA_EXFILTRATION]: The 'Phase 4' procedure targets sensitive URLs to identify Web Cache Deception vulnerabilities. It downloads response content to local temporary files and specifically searches for PII and credentials, such as emails, passwords, and tokens, using keyword filtering.
[EXTERNAL_DOWNLOADS]: The skill uses curl to interact with external web servers provided as user input. It sends multiple requests with various headers (e.g., X-Forwarded-Host) to test for vulnerabilities, which involves downloading data from untrusted remote sources.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes untrusted data from external target responses (via curl) without using boundary markers or sanitization, potentially allowing malicious content from a target server to influence the agent's context. Evidence chain: 1. Ingestion points: Target response data captured via curl in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: Network requests, shell script execution, and local file writing. 4. Sanitization: Absent.

cache-attack