Skillv0.1.3

ClawScan security

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

Scanner verdict

BenignMar 21, 2026, 9:56 AM

Verdict: Benign
Confidence: medium
Model: gpt-5-mini
Summary: The skill's declared purpose (P2P sports predictions) matches its requirements and instructions, but there are a few modest inconsistencies and privacy/network risks you should review before installing or using it.
Guidance: This skill appears to be what it says (a CLI/MCP client for agentsports.io), but take these precautions before installing or using it: - Verify the install source: the installer pulls a package from a GitHub repo (elesingp2/agentsports-connect). Inspect that repository and any build/install steps before installing the `asp` binary. - Protect PII and passwords: the skill requires email, password, full name, DOB, and phone for registration. Only provide those if you trust agentsports.io; consider using a throwaway account if you want to test. - Be careful with the MCP server: `asp mcp-serve` can open a network port (example: port 8000). Do not run the server on a public-facing host or without understanding what it exposes. - Check config path usage: the SKILL.md metadata references ~/.asp/ even though the registry summary listed no config paths — expect the client to write configuration/state there. - Review how credentials are stored and transmitted: the docs don't describe secure storage or transport; prefer using saved sessions only if you trust the binary. No static code findings were available (the skill is instruction-only), but absence of findings is not proof of safety. If you plan to use real money, audit the upstream repo and the `asp` binary first and only authorize the skill after you’re satisfied with the source and network behavior.

Review Dimensions

Purpose & Capability: okName/description, required binary `asp`, CLI commands, and MCP tools all align with a skill that controls a sports-prediction client. No unrelated cloud credentials or unrelated binaries are requested.
Instruction Scope: noteRuntime instructions explicitly tell the agent to collect sensitive personal data (email, full name, birth date, phone, password) and to submit it to agentsports.io for registration — this is expected for account creation but is sensitive. The skill also documents running an MCP server (e.g., `asp mcp-serve --port 8000`), which opens a network endpoint the agent could host. The instructions do not attempt to read unrelated system files, but they do instruct handling user passwords and starting network services.
Install Mechanism: noteInstall spec uses a 'uv' installer fetching package from a GitHub repo (git+https://github.com/elesingp2/agentsports-connect.git) and sets PATH. Pulling binaries from a third-party GitHub source is common but higher-risk than a vetted package registry because code is pulled and installed from that repository; you should inspect the repo and binary before trusting it.
Credentials: noteThe skill declares no required environment variables or primary credential, which matches the CLI-oriented design. However, it will handle user-supplied credentials and PII (passwords, DOB, phone) during registration/login; sensitive data handling is inherent to the use case and should be treated carefully (storage, transmission, and reuse policies are not described).
Persistence & Privilege: concernThe skill is not always-enabled, but the SKILL.md metadata references a config path (~/.asp/) even though the registry summary above listed none — this metadata mismatch is an incoherence to note. Also, the MCP server capability lets the agent host an HTTP/streamable endpoint; while not automatically enabled, running it grants network-facing privilege and should be allowed only with user consent and inspection of the server behavior.