Survival Manager

自主调整运行模式，管理资金与任务，需用户授权高风险操作，保障系统稳定与财务健康的生存管理工具。

This skill is plausible for local financial/agent management but has a few unclear points you should verify before installing: - Telegram & notifications: SKILL.md mentions Telegram notifications but the skill does not declare or request any Telegram API token or explain how the 'message' tool is configured. Ask the author how notifications are delivered and where credentials are stored; do not provide tokens unless you trust the integration. - Model invocation & costs: The skill references qwen3.5-* models and per-call costs. Confirm whether model calls go through your OpenClaw account (and billing) or require external API keys. Understand and accept potential usage costs before enabling autonomous behavior. - File writes and authorization flow: The scripts will write and update local files (finance logs, survival-config.json, authorization-pending.md). Ensure these files and their paths are acceptable and that the authorization flow actually pauses high-risk operations the way you expect. - OS mismatch: Provided scripts are PowerShell (.ps1) but the skill has no OS restriction. If you run on non-Windows hosts, verify compatibility or convert scripts. - Test in a sandbox: Run the skill in an isolated environment first to confirm where network calls go, what files are created/modified, and how notifications are sent. Inspect survival-config.json and scripts line-by-line. If the author can clarify how Telegram and model invocations are configured (what credentials are needed and where they are stored), and if you confirm the skill will not receive or require hidden secrets, the inconsistencies become minor. Without that clarification, proceed cautiously.

Type: OpenClaw Skill Name: survival-manager Version: 1.0.0 The skill is designed for local agent management, focusing on financial tracking and operational adjustments based on a 'survival level' system. It explicitly states '无外部 API 调用' (No external API calls) and '无数据离开机器' (No data leaves the machine) in SKILL.md, and its PowerShell scripts perform only local file operations (reading/writing JSON/Markdown files). High-risk actions like file writes and sending external messages are clearly documented as requiring user authorization, indicating a design focused on transparency and user control rather than malicious intent. There are no signs of data exfiltration, obfuscation, or prompt injection attempts to bypass security measures.