The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands using variables like $TARGET, $H, and $SESSION_COOKIE for fingerprinting and credential validation. The lack of sanitization on these variables, particularly when sourced from files like live-hosts.txt, presents a significant risk for arbitrary command injection if an attacker can manipulate the input data.
[DATA_EXFILTRATION]: The fingerprinting logic uses curl to perform network requests to hosts specified in the $TARGET variable. This functionality can be misused to conduct unauthorized network scanning (SSRF) or to exfiltrate data to an external server by controlling the target list.
[CREDENTIALS_UNSAFE]: Sensitive session cookies and bearer tokens used for greybox testing are passed as command-line arguments to curl. This practice is unsafe as it exposes credentials in cleartext within the system's process table and potential shell history.
[PROMPT_INJECTION]: The skill ingests untrusted data from remote target responses (HTTP headers and body content) and uses grep to identify markers that determine which attack skills to load. This constitutes an indirect prompt injection surface where a malicious target can manipulate the agent's logic and capability loading by serving specific strings (e.g., laravel_session, okta.com). Evidence chain: Ingestion from curl in SKILL.md; No boundary markers; Capability inventory includes curl, grep, and skill loading; No sanitization of target content.

hunt-dispatch