The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The references/exploit-chains-template.md file contains illustrative examples of remote code execution, including a PHP webshell payload (<?php system($_GET["cmd"]); ?>) and instructions for its deployment via XML-RPC. These are documented for reporting purposes and are not executed by the skill itself.
[DATA_EXFILTRATION]: The provided reporting template includes a JavaScript proof-of-concept for exfiltrating sensitive data via CORS vulnerabilities to an external placeholder domain (attacker.com).
[COMMAND_EXECUTION]: The skill documentation provides multiple curl command examples demonstrating interaction with various target endpoints (e.g., XML-RPC, WooCommerce REST API) to aid in documenting verified findings.
[PROMPT_INJECTION]: The skill defines a process for ingesting and normalizing "evidence" and finding data from other validation skills, which constitutes an indirect prompt injection surface.
Ingestion points: Vulnerability findings and evidence from "owning validation skills" are processed and documented within the SKILL.md workflow.
Boundary markers: The procedure requires manual evidence confirmation and normalization, but does not provide specific programmatic delimiters or sanitization instructions for the ingested content.
Capability inventory: The skill uses basic shell commands (mkdir, touch) to create and manage output documentation directories.
Sanitization: No explicit sanitization or filtering of external evidence content is described.

cross-attack-chains