Scan firmware blobs and filesystem dumps for hardcoded private keys, API tokens, default creds, and weak RSA/ECC material.
Single tool 'keyhound_scan' has a clear, action-oriented name and a substantive description (118 chars), but critical gaps undermine quality. Input schema is present with a 'target' parameter (string type), but lacks description for the parameter itself, LLMs cannot infer whether 'target' expects a file path, directory, or glob pattern. Output schema is not documented; the tool returns JSON but the structure (fields, types, nesting) is opaque to the LLM. No error handling guidance, no pagination support, no parameter constraints (e.g., file size limits, timeout behavior). The tool is READ_ONLY (good security posture), but the definition does not leverage tool annotations to signal this to the MCP client. Baseline: 194 chars avg description (this tool: 118 chars, acceptable); 100% of A+ tools document output schemas (this tool: none visible); 100% of A+ tool params have descriptions (this tool: 0/1).
Scan firmware blobs and filesystem dumps for hardcoded private keys, API tokens, default creds, and weak RSA/ECC material.. Returns JSON findings.
Input parameter 'target' lacks description. LLMs cannot determine whether to pass a file path, directory, glob pattern, or symlink. Baseline: 100% of A+ tools describe every parameter.
Output schema not documented. Tool returns JSON findings but structure (fields, types, array vs object, nested objects) is invisible to LLM. Baseline: 100% of A+ tools document return types.
No error handling guidance. Tool may fail on permission denied, file not found, timeout, or invalid path. LLM has no recovery path (e.g., 'Try a parent directory' or 'Check file permissions').
Tool annotations missing. READ_ONLY risk classification is documented in metadata but not exposed via MCP readOnlyHint. Client cannot optimize caching or safety policies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
No parameter constraints. 'target' accepts any string; no validation for file size limits, path traversal, or timeout behavior. LLM may pass absurd values (e.g., /dev/zero, 100GB file).