The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous command templates using security tools to demonstrate vulnerability exploitation.
Evidence: Includes commands for ffuf to brute force password reset codes (Path 3) and security answers (Path 8).
Evidence: Includes usage of hashcat for offline JWT secret cracking (Path 6).
Evidence: Includes a bash timing oracle loop utilizing curl to identify valid accounts (Path 7).
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to external sites to retrieve configuration or demonstrate leakage.
Evidence: Path 6 uses curl to fetch JWKS configuration from https://target.com/.well-known/jwks.json.
[DYNAMIC_EXECUTION]: The skill employs inline script execution to perform cryptographic operations.
Evidence: Path 6 uses python3 -c to execute an inline Python string for encoding JWTs.
[DATA_EXFILTRATION]: The skill performs network operations to non-whitelisted external domains as part of its attack demonstrations.
Evidence: Path 1 demonstrates host-header injection using attacker.com.
Evidence: Path 2 demonstrates token leakage via the Referer header to analytics.com.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingests untrusted target data into shell command templates without boundary markers or sanitization.
Ingestion points: The skill instructs the agent to demonstrations attacks against user-specified targets (e.g., target.com) and accounts (e.g., victimB@company.com).
Boundary markers: Absent; there are no instructions for the agent to validate or escape user-provided strings before interpolation.
Capability inventory: The skill utilizes shell execution, network tools (curl, ffuf), and script interpreters (python3).
Sanitization: Absent; the provided command templates rely on direct string interpolation of user-provided values.

hunt-ato