Skillv1.0.0

ClawScan security

stock-hot-zh-aisa · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

SuspiciousApr 18, 2026, 12:59 PM

Verdict: suspicious
Confidence: medium
Model: gpt-5-mini
Summary: The skill mostly aligns with its stated purpose (it calls an AISA LLM to produce a 'hot assets' report) but there are a few inconsistencies and ambiguous behaviors you should understand before installing.
Guidance: This skill asks only for an AISA API key and runs a Python script that sends a prompt to the AISA model to produce a live market report. Before installing: (1) confirm you trust the AISA endpoint and the model you will be invoking; the script will call the endpoint you configure via AISA_BASE_URL/AISA_MODEL if set; (2) understand that the script does not itself fetch Yahoo Finance/CoinGecko/news — it asks the LLM to do so, which can either (a) call external tools (if the model has browsing/tooling) or (b) hallucinate live numbers if it does not; (3) ensure you install the openai Python package from a trusted source and review the code if you need explicit, auditable data pulls (i.e., replace LLM-driven fetching with direct API calls to financial data providers); (4) do not supply other unrelated credentials — only provide the AISA_API_KEY. If you need guarantees of live-data accuracy, prefer a variant that makes direct API calls to verified financial data providers rather than relying on a model prompt to "fetch" data.

Review Dimensions

Purpose & Capability: noteName/description, required binary (python3), and the required AISA_API_KEY credential are coherent with a skill that calls an external LLM (AISA) to produce market scans. The included script uses the OpenAI-style client pointed at an AISA endpoint, which matches the declared primaryEnv. Expectation mismatch: the prompts instruct the model to "scan Yahoo Finance, CoinGecko, and financial news sources," but the script does not itself fetch web data — it relies on the AISA model to do that via its tooling or knowledge.
Instruction Scope: concernSKILL.md and the script direct the LLM to 'fetch current data' from specific third-party sites. However, the implementation sends only a prompt to AISA and does not programmatically query Yahoo Finance/CoinGecko/news APIs. This creates ambiguity: if the AISA model lacks browsing/tools, it may hallucinate live data; if the model does have tools, it could call external endpoints. The SKILL.md does not clearly document whether browsing/tools are required or used.
Install Mechanism: noteThere is no install spec (instruction-only with an included script). The script comments list a dependency on the openai Python package but no installation steps are provided — user must ensure python and dependencies are installed. Not inherently risky, but requires manual dependency setup which could lead to users running pip installs from instructions elsewhere.
Credentials: noteThe skill requires a single credential AISA_API_KEY (declared as primary). That's reasonable for calling the AISA service. The script also reads optional env vars AISA_BASE_URL and AISA_MODEL (not listed in requires.env). This is a minor inconsistency — they are optional, but you should know the client will use AISA_BASE_URL (default https://api.aisa.one/v1) if present and AISA_MODEL (default 'gpt-4o').
Persistence & Privilege: okalways:false (default) and no attempt to modify other skills or system-wide settings. The skill does not request elevated or persistent platform privileges.