Skillv1.0.4

ClawScan security

药物经济学评价技能包 · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignApr 26, 2026, 12:53 AM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: The skill's files, instructions, and requirements are internally consistent with a pharmacoeconomic evaluation toolkit and do not request unrelated credentials or install remote code; only minor documentation/parameter inconsistencies and typos were found.
Guidance: This package appears coherent and implements local pharmacoeconomic calculations. Before installing or running it: 1) Run it in a virtualenv or isolated environment and install only the declared Python libs (numpy, pandas, matplotlib); 2) Review and correct obvious documentation/parameter issues (example: a QALY example sets discount_rate=0.45 which is almost certainly a typo—recommended is ~0.03; example thresholds vary between 100,000 and 120,000 RMB across files); 3) Inspect any omitted/truncated files referenced in docs (e.g., xstpe/data/parameters.py, .codebuddy paths) to ensure they don't contain unexpected behavior; 4) If you will run with real patient data, ensure you follow local data‑privacy controls (do not commit PHI to remote systems); 5) If you plan to integrate this into automated agents, review the code for any future network or file I/O additions—currently there are no network endpoints or credential requests. If you want, I can list the exact lines where the inconsistent numbers/typos appear and suggest concrete fixes.

Review Dimensions

Purpose & Capability: okName/description match the included scripts and documentation: the package provides CEA/CUA/CBA, QALY/ICER calculations, PSA/Monte‑Carlo and budget impact analyses. There are no unexpected environment variables, binaries, or external service credentials requested.
Instruction Scope: noteSKILL.md instructs the agent to use the local scripts (e.g., cost_effectiveness_analysis.py, monte_carlo_simulation.py). That is within scope. Minor issues: inconsistent numeric examples across docs (e.g., SKILL.md example uses threshold=100000 while manifests/examples use 120000), a suspicious-looking discount_rate=0.45 in one QALY snippet (likely a typo vs recommended 0.03), and some documentation paths (e.g., .codebuddy, xstpe/data/parameters.py, Windows path in manifest) that appear to be leftover examples rather than required runtime behavior. These are quality issues but not scope creep or exfiltration instructions.
Install Mechanism: okNo install spec is provided (instruction-only install), and the manifest suggests standard Python dependencies (numpy, pandas, matplotlib). No network/download/install URLs or archive extraction are present in the package metadata.
Credentials: okThe skill requests no environment variables, no credentials, and no config paths. All computations are local and consistent with the stated purpose; there are no requests for unrelated secrets or system access.
Persistence & Privilege: okalways is false and disable-model-invocation is default (agent may call autonomously, which is normal). The package does not declare any special persistent privileges or modify other skills or system-wide settings.