Loading ToolBench
Scans Node, Python, and AI-agent configs for indicators of compromise from npm and PyPI supply-chain attacks.
patient-zero has 14 tools with READ_ONLY and WRITE risk classifications, but critical definition gaps severely limit usability. Most tools lack parameter descriptions entirely (e.g., scanMcp, scanLocalFiles, scanProcesses have empty input schemas {}). Descriptions are present but generic (10-50 chars, well below the 50-200 char LLM-optimized baseline). Output schemas are completely undocumented, LLMs cannot plan downstream calls. The render* tools (renderTerminal, renderJson, renderMarkdown, renderSarif) accept identical 7-parameter inputs but lack descriptions for any parameter, making them indistinguishable to an agent. No error handling guidance, no recovery paths, no validation rules. Tool names are action-verb-based (scan*, render*, load*, install*, detect*) which is good, but the lack of parameter and output documentation makes the tools difficult to use correctly.
Detect which pre-commit ecosystem is in use in cwd
Install patient-zero as a pre-commit hook (husky / lefthook / pre-commit / native)
Load the IoC database from network, cache, or bundled snapshot
Remove the patient-zero managed block from whichever hook system has it
Render scan results as JSON for CI consumption
Render scan results as a Markdown report
Render scan results as a SARIF v2.1.0 file for GitHub Security tab
Render scan results to the terminal with colored output
Four scanner tools (scanMcp, scanLocalFiles, scanProcesses) have empty input schemas {} with no parameter descriptions.
Four render* tools (renderTerminal, renderJson, renderMarkdown, renderSarif) accept identical 7-parameter inputs but have no parameter descriptions. LLMs cannot distinguish when to use each variant or what each parameter controls.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does scanLockfiles return? Is it an array of findings? An object with metadata?). This violates the documented output schema requirement.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
Run the install-interception flow: detect package manager, resolve install tree, scan against IoCs, block on hit
Scan GitHub account for known IoCs
Scan local files for known IoCs
Scan lockfiles for known IoCs
Scan MCP configurations for known IoCs
Scan running processes for known IoCs
Tool descriptions are generic and under 50 characters (e.g., 'Scan MCP configurations for known IoCs' = 42 chars). Current descriptions lack context for agent selection.
No error handling guidance. Tools like runInterceptor (WRITE risk) and installHook (WRITE risk) have no documented error recovery paths, retryability, or confirmation steps. Agents cannot know if a failure is retryable or fatal.