The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external reconnaissance files without sanitization or clear boundary markers.
Ingestion points: recon/$TARGET/api-endpoints.txt and recon/$TARGET/live-hosts.txt via shell while read loops and awk commands.
Boundary markers: Absent; the content of these files is treated directly as command arguments.
Capability inventory: Shell execution of curl and httpx with arbitrary URL inputs and custom headers.
Sanitization: None; the skill assumes the input files contain only valid hostnames or URLs.
[DATA_EXFILTRATION]: The skill provides templates and instructions for exfiltrating sensitive authenticated data (including PII and CSRF tokens) to external domains. While it suggests well-known security services such as oastify.com for verification, these patterns demonstrate the capability to move data from a target system to an attacker-controlled listener.
[COMMAND_EXECUTION]: Automated reconnaissance logic is implemented through shell commands. The agent is instructed to use curl to probe endpoints with custom Origin headers and user-provided session cookies to detect reflected headers.
[EXTERNAL_DOWNLOADS]: Recommends the installation of external security packages such as corsy (via pip) and the use of the nuclei scanning engine for automated vulnerability detection.

hunt-cors