The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill employs a highly dangerous execution pattern in SKILL.md by downloading an external shell script and piping it directly into a shell interpreter (curl -L ... | sh). This allows for unverified code execution from a remote source.\n- [COMMAND_EXECUTION]: The script scripts/linpeas_runner.py uses subprocess.Popen to execute the downloaded linpeas.sh script. While it uses a list for arguments, the underlying action executes a downloaded script from an untrusted external source.\n- [EXTERNAL_DOWNLOADS]: The skill fetches executable content from a source flagged as malicious by reputation scanners (https://github.com/peass-ng/PEASS-ng/releases/latest/download/linpeas.sh) using urllib.request.urlretrieve.\n- [DATA_EXFILTRATION]: The skill contains logic and instructions to search for and expose highly sensitive files, including SSH private keys (id_rsa, id_ecdsa), cloud credentials (.aws/credentials), and password hashes (/etc/shadow). This represents a high-severity data exposure risk.\n- [PROMPT_INJECTION]: Indirect prompt injection surface detected. 1. Ingestion points: scripts/linpeas_runner.py processes raw stdout from a shell script running on an external target host. 2. Boundary markers: None identified; output is processed directly after ANSI stripping. 3. Capability inventory: Subprocess execution and network requests. 4. Sanitization: None; the output is parsed via regex but not sanitized for embedded instructions that could manipulate agent behavior.\n- [COMMAND_EXECUTION]: The file references/privesc_vectors.md provides explicit payloads for privilege escalation, such as SUID binary exploitation (find . -exec /bin/bash -p \;) and sudo misconfiguration abuse, which are designed to bypass system security controls.

privesc-linpeas