Skillv0.1.0

ClawScan security

PaperBanana · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

SuspiciousMar 11, 2026, 11:40 AM

Verdict: suspicious
Confidence: medium
Model: gpt-5-mini
Summary: The skill's stated purpose (generate academic diagrams) is plausible, but the package is incomplete and the registry metadata disagrees with the runtime instructions about required API keys and dependencies — proceed cautiously.
Guidance: This package appears to implement what it claims, but it is incomplete and has mismatched metadata. Before installing or running: - Do not provide your OpenRouter or Google API key to this package until you verify the source. - Inspect the missing files (agents/, utils/, requirements.txt, configs/) — the run.py imports many modules that are not included in the manifest; request the complete source or the official repository URL. - Be aware run.py will download data from Hugging Face and make network calls to model/image APIs (your input text and any model-generated data will be sent to those services). - If you must try it, run in an isolated environment (VM/container) and use a limited, low-budget API key with billing caps or a test account. - Check configs/model_config.template.yaml and any code in the agents/ modules for unexpected external endpoints or telemetry before supplying secrets. - If the publisher/source cannot be verified or the missing code is not provided, avoid using the skill with sensitive or proprietary text.

Review Dimensions

Purpose & Capability: noteThe described purpose (turn method text into publication-quality diagrams) matches the runtime behavior in run.py (multi-agent pipeline, image generation). However the registry metadata claims no required environment variables or binaries while SKILL.md and run.py clearly expect model provider API keys (OPENROUTER_API_KEY or GOOGLE_API_KEY), Python runtime, and other project files (configs, agents, utils). That mismatch between metadata and actual instructions is incoherent.
Instruction Scope: concernSKILL.md instructs pip installing requirements, setting API keys, and possibly downloading a dataset from Hugging Face. The included run.py will copy model config templates, download datasets via huggingface_hub, import many modules under agents/ and utils/ (PlannerAgent, VisualizerAgent, etc.) which are not present in the packaged files. The code calls external model/image APIs (OpenRouter/Gemini) which implies network calls and transmission of input text to third-party services. There is no instruction or code here that reads unrelated system secrets, but the package is incomplete so the full runtime behavior of the missing modules cannot be verified.
Install Mechanism: concernNo install spec is provided in the registry (instruction-only), but SKILL.md instructs 'uv pip install -r requirements.txt' and run.py assumes presence of additional project files (configs, requirements.txt). Those referenced files are not included in the manifest. The lack of packaged dependencies but expectation of network installs and external packages is a packaging/integrity risk.
Credentials: noteRequesting an API key for an LLM/image provider (OpenRouter or Google/Gemini) is proportionate to a diagram-generation skill. However the registry metadata declared no required env vars while the README and runtime expect credentials — an inconsistency. SKILL.md also suggests storing API keys in configs/model_config.yaml (writing credentials to disk), which users should consider carefully.
Persistence & Privilege: okThe skill does not request always: true and does not assert elevated or persistent system privileges. It will create files under its work_dir (configs, downloaded datasets, output images) but does not modify other skills or system-wide settings in the provided code.