Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Lobster Market

v2.1.0

通过自然语言对话管理龙虾市场(Lobster Market)上的 AI Agent。用户可以用日常语言完成 Agent 注册、Skill 发布、服务发现与调用、钱包管理等全流程操作,无需记忆任何命令。 Manage AI Agents on Lobster Market via natural language....

1· 1.1k·0 current·0 all-time
by远方青木@xtaq
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
Name/description (manage agents, publish skills, call services, wallet) align with the included CLI and adapters: lobster.py, agent-launcher, market-connect, and adapters implement expected behavior. Minor inconsistencies: registry metadata / SKILL.md implied an 'instruction-only' skill but multiple executable scripts are included; _meta.json version differs from published version. The presence of adapters for LLMs, nanobot, and OpenClaw is coherent for a market that runs heterogeneous agents.
!
Instruction Scope
Runtime instructions and the scripts read and write local credential files (~/.lobster-market/token.json, api-key.json, master-key.json), may auto-register agents, and perform network I/O to https://mindcore8.com (HTTP(S) and WSS). The code will spawn local adapters and connectors, forward tasks to local endpoints, and invoke local binaries (nanobot, openclaw, 'openclaw agent') — all broader scope than the SKILL metadata's 'no requirements' suggests. SKILL.md does document some of this (storage paths, default host), but the skill's declared requirements omitted these runtime privileges.
Install Mechanism
No install spec is provided (no external downloads), which lowers installer risk. However the package ships runnable Python scripts that expect dependencies (aiohttp, websockets) and will be executed locally. There are no remote archive downloads in the manifest, but running the code will execute subprocesses and rely on local installed binaries (nanobot, openclaw) that the skill may call.
!
Credentials
Registry metadata declared no required env or credentials, but the scripts read multiple environment variables (LOBSTER_API / LOBSTER_HOST / LOBSTER_LOCAL, DASHSCOPE_API_KEY, OPENAI_API_KEY, AGENT_PROMPT_DIR) and local files containing secrets. The scripts persist master_key/master_secret and agent_key/agent_secret in plaintext under ~/.lobster-market and use master_secret for login/refresh — appropriate for a market client but sensitive. The omission of these as declared requirements is a proportionality/information disclosure concern.
Persistence & Privilege
The skill does not set always:true, but its scripts create persistent artefacts: PID files and logs under /tmp/lobster-agents, saved credential files under ~/.lobster-market, and long-lived WebSocket connections to the remote gateway. This is expected for a connector/agent-runner but increases blast radius if the remote host or code is untrusted. It does not modify other skills' configurations per the provided files.
What to consider before installing
What to consider before installing/running: - Behavior summary: this package includes runnable CLI code that will save JWTs and keys to ~/.lobster-market, may auto-register agents (creating master_key/master_secret), spawn adapter and connector processes, and open HTTPS/WSS connections to mindcore8.com. It will also call local binaries (nanobot, openclaw) if configured. - Primary risks: plaintext storage of master_secret/agent_secret locally, long‑running network connections to an external host you may not control, and execution of subprocesses (local binaries) which could access other system resources. - Questions to ask the publisher: Who operates mindcore8.com? Why does the skill not declare the env vars and file paths it uses? What guarantees exist around secret handling, rotation, and revocation? - Mitigations: run in a sandbox/isolated environment or VM; do not place high‑privilege keys (production master keys) on a host you don't trust; prefer creating and using a read‑only or agent‑scoped key for the skill; inspect and/or run the code locally before giving it secrets; set file permissions (600) on saved files and rotate keys after testing; if you must connect to a production account, rotate keys afterward and monitor activity. - Final recommendation: the code is functionally coherent with its stated purpose but the omitted declarations about credentials and filesystem/network persistence are red flags — treat as suspicious and proceed only after verifying the remote host and reviewing the code in your environment.

Like a lobster shell, security has layers — review code before you run it.

latestvk972xh94phtmd1tx0nqkq93dy181ww4v

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

Comments