file-path-traversal

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly instructs extracting and reporting sensitive file contents (e.g., /etc/shadow, SSH private keys, /proc/self/environ, wp-config.php) and includes examples of embedding cookie/header values in requests, which requires the agent to handle and output secret values verbatim.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This content provides detailed, actionable offensive techniques for path traversal and LFI-to-RCE (log poisoning, php:// wrappers, encoded payloads, and specific payloads to read /etc/shadow, SSH keys, and execute commands), enabling direct data exfiltration and remote compromise.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). The SKILL.md Core Workflow and Phases (notably Phase 3, Phase 7 and Phase 8) explicitly instruct the agent to fetch and parse responses from arbitrary target URLs using tools like curl, ffuf, and wfuzz (e.g., "curl 'http://target.com/page?file=...'", "ffuf -u 'http://target.com/image?filename=FUZZ'"), meaning the agent will ingest untrusted third‑party web content that can materially influence subsequent actions.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). This skill explicitly instructs an agent how to read sensitive local files, perform LFI-to-RCE techniques (log poisoning, php:// wrappers, /proc/self/environ), and extract or execute commands on a host — enabling compromise of the machine the agent runs on or any targeted server.