This server has critical definition quality gaps. Only 2 tools are defined. The 'run' tool has a destructive pattern (shell execution) with a TODO comment acknowledging missing input sanitization, a major security and quality issue. The 'tools' tool has no input parameters, which is correct, but both tools have minimal descriptions that lack context for LLM selection. Parameter descriptions are present but sparse. No output schemas are documented. The server includes webshell PHP samples in the codebase, which is concerning from a security perspective. Overall, definition quality is poor and would not pass production code review.
Tools (2)
rundestructivesource verified37/100
Run a shell command and return output as a JSON string.
toolsread onlysource verified37/100
List available malware analysis and utility tools.
Critical: 'run' tool executes arbitrary shell commands with shell=True and no input sanitization. Code contains TODO comment: '# TODO: sanitize the input'. This is a command injection vulnerability.
Tool descriptions are under 50 characters and lack context about when/why to call them. 'run' description (80 chars) does not warn about destructive nature or explain that it executes arbitrary commands. 'tools' description (45 chars) is minimal and does not explain the discovery purpose.
Output schemas are not documented. The 'run' tool returns JSON with 'command', 'stdout', 'stderr', 'exit_code' fields, but LLMs have no declared schema to understand this structure beforehand. The 'tools' tool returns a dict with tools_dict, but the nested structure is undocumented.
Add comprehensive input validation to the 'run' tool. Reject commands containing dangerous patterns (rm, kill, shutdown, dd, mkfs, etc.). Implement a whitelist of allowed binaries if possible, or at least a blacklist of dangerous commands. Remove the TODO comment and actually sanitize.
Expand 'run' description from 80 chars to 150-200 chars. Add warnings: 'WARNING: This tool executes arbitrary shell commands. Only safe for trusted use. Commands run in cwd=<DATA_FOLDER>. Execution is NOT sandboxed.' This tells LLMs the destructive nature upfront.
Implement error classification. Instead of generic 'exit_code=-1', distinguish between timeout (retryable), permission denied (user-fixable), command not found (guidance to use 'tools' tool), and unexpected error (fatal). Return actionable messages like 'Command not found. Run tools() to see available binaries.'
Add a 'dry_run' parameter to the 'run' tool. When true, return what would execute without actually running it. This lets agents plan before committing to destructive operations.
Implement permission checks and audit logging. Log every execution to a secure audit trail: timestamp, caller ID, command executed, result, exit code. Gate execution behind a permission check that the agent/user has 'execute:shell' scope.
Score history
Overall score trend
↑ 1 points across a rubric change (v1 → v2)
30/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
F
30
2026-07-28+
v2
2026-03-09
F
29
-
v1
No error handling guidance. The 'run' tool catches generic Exception and returns exit_code=-1 with stderr. LLMs have no context to know whether to retry, what caused the failure, or how to recover.
No destructive operation protection. The 'run' tool can execute 'rm -rf /', 'kill -9', or any destructive command without confirmation, dry-run, or permission check.
Dangerous tool exposed to untrusted agents. Combining an unvalidated shell execution tool with an LLM that can be prompt-injected creates an attack surface. No rate limiting, permission checks, or audit trail.
PHP webshell samples included in repository (/data/webshell.php, /data/webshell1.php). These are malware examples and should not be committed. This raises concerns about the server's intended use and raises red flags for security review.
Parameter naming lacks type suffixes. The 'command' parameter in 'run' is clear, but lacks description detail about shell escape rules, length limits, or forbidden patterns.
run
Remove or .gitignore webshell PHP samples from the repository. These are red flags for malware/hacking tools and should not be included in production code. If they are test fixtures, document them clearly as test-only and exclude from released packages.
Add rate limiting. An agent in a retry loop could spawn hundreds of shell commands. Limit to 10 calls/minute per agent. Return clear rate-limit-exceeded error.
Document the 'tools' tool description better. Expand from 45 chars to ~120 chars: 'Discover available analysis tools installed on this system. Returns tool name, description, usage, and path. Call this first to see what binaries are available before using run() to execute them.'
Consider splitting 'run' into narrowly scoped tools: 'analyze_file', 'search_strings', 'identify_filetype', 'scan_with_yara', each wrapping a specific safe command. This reduces surface area and makes the LLM intent clearer.