Back to skill
Skillv1.0.0
ClawScan security
OUA 统一智能评估 · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
BenignApr 26, 2026, 1:53 PM
- Verdict
- benign
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill's files and runtime instructions are consistent with its stated purpose (an LLM benchmarking/assessment framework and report generator); it requires no credentials or installs and the included script performs local scoring/reporting only.
- Guidance
- This package appears coherent for running an offline assessment and generating HTML reports. Before running it: (1) Inspect the Python script (already included) and the test bank to confirm it does only local I/O (it does). (2) Run it in a sandbox or isolated environment (container/VM) if you will execute untrusted code. (3) Do not feed production secrets or sensitive user data into its input JSON — the tool writes report files and could include those values in output. (4) If you intend to rely on results for high-stakes decisions, review the scoring logic and weights to ensure they match your evaluation criteria. (5) If you want extra assurance, verify the referenced GitHub repository and author identity and run static checks (lint, safety scanners) on the code.
Review Dimensions
- Purpose & Capability
- okName/description (OUA unified assessment) align with the included assets: SKILL.md (framework + tests) and scripts/score_test.py (scoring and HTML report generation). The code implements weighting, per-dimension scoring, grade/ quadrant logic and report generation — all expected for a benchmarking tool.
- Instruction Scope
- okSKILL.md provides test definitions, trigger phrases and guidance for running assessments; it does not instruct the agent to read unrelated system files, access secrets, or transmit data externally. The test bank contains prompts that intentionally include harmful examples (to test refusal), but those are test cases, not active exfiltration instructions.
- Install Mechanism
- okThere is no install specification (instruction-only skill with a bundled Python script). No downloads, package installs, or extracted archives are declared. Risk is limited to running the included Python script locally.
- Credentials
- okThe skill declares no required environment variables, no primary credential, and no config paths. The code reads JSON input and writes HTML reports locally; it does not request or use credentials or external tokens.
- Persistence & Privilege
- okalways is false and the skill is user-invocable. It does not request permanent presence or attempt to modify other skills or system-wide agent settings. The script reads/writes local files only (input JSON / output HTML).