Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
Token Optimizer 1.0.0
v1.2.0✂️ TOKEN节约器 - 工作流程控制器。通过问题预检、路径验证、进度检查、错误快速定位，防止重复无效工作，节约TOKEN消耗。兼容Windows/Mac/Linux/MaxClaw/ClawHub。
⭐ 0· 50·0 current·0 all-time
by@smxtx
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The name/description (workflow pre-checks, path verification, checkpointing) align with the checks shown in SKILL.md (file existence, ports, processes, logs). Requesting no environment variables and no install is consistent with an instruction-only diagnostic helper. However, a user expecting only token-optimization guidance may not expect the skill to run OS-level commands and inspect running processes/ports/files — the operational footprint is broader than the high-level marketing implies.
Instruction Scope
SKILL.md contains explicit runtime code and instructions that execute system commands and read filesystem/process state: the PreChecker sample uses node's child_process.execSync to run netstat/lsof/pgrep/powershell; there are filesystem existence checks and suggested log analysis. These instructions direct the agent (or developer) to inspect ports, processes, files, and environment variables. That scope is coherent for diagnostics but is high-privilege and can reveal sensitive local state. The code concatenates user-supplied values into shell commands (processName, port) without robust sanitization, creating a risk of command injection or unintended execution if inputs are not strictly validated.
Install Mechanism
No install spec and no code files — instruction-only — which means nothing will be written to disk by the skill package itself. This minimizes supply-chain risk, but the instructions still propose executing system commands when used.
Credentials
The skill declares no required environment variables or credentials, which is proportionate. Nevertheless, the guidance includes checks for 'Token/API Key是否有效' and '环境变量检查', implying the agent may be asked to read environment variables or validate credentials at runtime. Because no specific env vars are declared, any access to environment variables would be ad hoc and should be considered sensitive.
Persistence & Privilege
always is false and there is no indication the skill persists configuration or modifies other skills. Autonomous invocation is allowed by default but not combined with any 'always:true' privilege or declared persistent presence.
What to consider before installing
This skill appears to be a local diagnostic/workflow helper and includes code/instructions that run system commands (netstat/lsof/pgrep/powershell), check files, and inspect processes and environment variables. That's plausible for a 'pre-check' tool, but it also means the skill — if executed — can read sensitive local state (open ports, running processes, files, env vars). The SKILL.md sample constructs shell commands by interpolating user-supplied values, which can lead to command injection if inputs aren't strictly validated. Before installing or enabling this skill:
- Treat it as powerful: only enable it for trusted agents in a controlled environment (or sandbox).
- Ask the publisher for provenance (source code repository, homepage, who is 'Matrix Agent') since 'Source' is unknown.
- Request explicit documentation on what exact paths/commands the skill will run and what inputs it will accept; prefer whitelists and input sanitization.
- If you must use it, avoid granting it access to sensitive directories or credentials; test it in a disposable environment first.
Confidence is medium because the behavior is coherent with a diagnostics tool, but the instruction-level use of shell commands and lack of provenance/constraint make the package risky without further clarification.Like a lobster shell, security has layers — review code before you run it.
latestvk972ewrjswebm029mqjfwnezms83mvep
License
MIT-0
Free to use, modify, and redistribute. No attribution required.