The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the tool using curl -fsSL https://raw.githubusercontent.com/s0ld13rr/pentestcode/main/install.sh | bash. This pattern executes unverified remote scripts with the user's shell privileges, posing a severe risk of arbitrary code execution from a non-vendor-controlled source.
[COMMAND_EXECUTION]: The agent is designed to autonomously execute high-risk penetration testing tools, including nmap, nuclei, gobuster, and netexec. The 'Auto Mode' allows the agent to plan and execute full engagements, including exploitation and post-exploitation tasks, with minimal user intervention.
[PROMPT_INJECTION]: The skill possesses a broad indirect prompt injection surface as it is designed to ingest and parse untrusted data from various sources (Category 8).
Ingestion points: Reads and parses external files including scan.xml (nmap), nuclei.json, cme.log, gobuster.txt, bloodhound.json, sqlmap.log, and raw HTTP response bodies for XSS analysis.
Boundary markers: The instructions do not define boundary markers or delimiters to separate untrusted tool output from agent instructions.
Capability inventory: The agent has access to powerful CLI tools, credential spraying logic (cred_spray), and shell management capabilities (tunnel_manage).
Sanitization: There is no mention of sanitization or validation of the data parsed from these external tools before it is processed by the agent's LLM components.
[DATA_EXFILTRATION]: The skill explicitly automates the collection of sensitive data, including dumping SAM/LSA secrets, extracting domain admin hashes, and harvesting credentials from discovered services. This data is maintained in a persistent state file (.pentestcode/state.json) and can be reported or potentially transmitted.
[COMMAND_EXECUTION]: The skill documentation suggests modifying system configurations and installing packages using sudo apt install, which involves privilege escalation on the host machine.

pentestcode-ai-pentest-agent