The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell script blocks utilizing curl, awk, grep, and jq to perform active network testing against user-specified targets. These scripts are intended to be executed by the agent to validate security flaws.
[DATA_EXFILTRATION]: The methodology involves capturing session cookies and replaying them to test for invalidation gaps. While the scripts are designed to send data only to the specified $TARGET domain, the handling of sensitive session tokens in temporary files (mktemp) and variables is a necessary part of the security testing workflow.
[PROMPT_INJECTION]: The skill uses instructional language such as "IMPORTANT" and "CRITICAL" to emphasize testing priorities. This is used in a benign, pedagogical context rather than an attempt to override the agent's safety guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze HTTP response data from external, potentially untrusted web targets.
Ingestion points: HTTP response headers and bodies retrieved via curl from the user-defined $TARGET variable in all testing phases (SKILL.md).
Boundary markers: The instructions recommend using "body-diff" and "negative controls" to validate findings, but do not provide explicit prompt delimiters to isolate the untrusted external data from the agent's instructions.
Capability inventory: The agent has access to network operations (curl) and shell processing utilities (jq, awk, base64) to evaluate the ingested data.
Sanitization: The skill employs structured parsing using jq for JSON responses and awk for cookie jars, which provides some protection, but lacks comprehensive sanitization for raw HTML or text response bodies.

hunt-session