token-use-optimizer

AI Agent 项目的 Token 消耗审计与系统性优化。提供四层诊断框架、六步优化 SOP、 自动化审计脚本。适用于任何使用 Rules + Memory + Knowledge + Skills 架构的 AI Agent 项目（如 CodeBuddy、Cursor、Windsurf 等）。 This ski...

This skill appears coherent: it bundles a Python auditor and documentation that match its stated goal of auditing Rules/Memory/Knowledge token costs. Before installing or running: - Inspect SKILL.md and scripts/token_audit.py yourself (or have a developer do so). The included script reads files under the target project and prints a report — it does not perform network calls in the provided code, but executing arbitrary scripts can be risky. - Run the script on a copy of your project, not on a production workspace containing secrets. The tool will read repository files (including MEMORY.md) and may output their contents or character counts. - Pay attention to the pre-scan 'system-prompt-override' flag: search SKILL.md for any lines that try to change system prompts or inject agent/system-level instructions. If you find such lines, do not run the skill until they are removed or explained. - Execute the script in a restricted environment (non-root user, ephemeral container) the first time to confirm behavior. - If you rely on enterprise security policy, have your security/dev team review the Python file for any unexpected file writes or subprocess/network usage (the provided code appears to use only local file reads and standard libs). If you want higher confidence, provide the exact SKILL.md frontmatter and the full, untruncated portion of scripts/token_audit.py print/report output for a quick line-by-line check.

Type: OpenClaw Skill Name: token-use-optimizer Version: 1.0.0 The bundle provides a legitimate utility for auditing and optimizing token consumption in AI agent projects. It includes a Python script (scripts/token_audit.py) that performs local file system scanning of rules, memory, and knowledge files to estimate token overhead, and comprehensive documentation (SKILL.md, references/) providing optimization strategies. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found; the script uses only standard libraries and lacks any network or unauthorized persistence capabilities.