Remote execution of Nmap scan through SSH persistent connection, returning XML format results
Server has 5 tools with basic schemas and descriptions, but significant gaps in quality. All tools have descriptions (10-100 chars, mostly adequate) and input schemas with type definitions. However, parameter descriptions are minimal or missing context. No output schemas documented. Error handling is present but lacks recovery guidance. Tool naming is clear (verb_noun pattern) but compositions expose security risks (execute_remote_command is dangerously broad). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Baseline: avg tool description ~60 chars (below 194 baseline); params average 1-2 descriptions per tool (well below 100% coverage).
Download a file from the remote Kali Linux server to local machine via SFTP
Execute Nmap scan on remote Kali Linux via SSH and return XML results
Execute arbitrary shell command on remote Kali Linux server via SSH
Read and return the contents of a text file from the remote server
Upload a local file to the remote Kali Linux server via SFTP
execute_remote_command accepts arbitrary shell commands with no validation, sanitization, or confirmation. Agents can be tricked into command injection (e.g., 'rm -rf /'). No dry-run or confirmation step for irreversible operations.
SSH credentials (hostname, username, password, port) hardcoded in SSH_CONFIG dict in main.py. Secrets must never appear in source code or tool parameters. Use environment variables or vault injection.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., execute_nmap_scan returns 'XML results' but structure is undefined.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Parameter descriptions are minimal. E.g., 'additional_args' says 'Additional Nmap arguments (optional, e.g., '-sV', '-O', '-A')' but does not explain what these flags do or when to use them. LLMs cannot reason about valid values.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from destructive writes. upload_file_to_remote and execute_remote_command are marked WRITE/IRREVERSIBLE in metadata but not in tool schema.