OpenClaw Cost Analyzer

分析 OpenClaw 的 token 使用和成本，识别高消耗场景（长对话、频繁 cron、大 context），给出具体优化建议（模型降级、context 压缩、cron 频率调整）。纯 Node.js，无外部依赖。输出清晰的成本报告。

This skill appears to be a local cost-analysis tool and does not contain network calls, external installs, or credential requests — but review before use: - Verify which session-log path your OpenClaw actually uses. The documentation claims ~/.openclaw/agents/main/agent/sessions/*.jsonl, but the script reads ~/.openclaw/workspace/memory/conversations. If these don't match, the script will find no logs. Update CONFIG.logsDir in scripts/cost_analyzer.js or adjust where your logs are stored. - Inspect the full script yourself (scripts/cost_analyzer.js) and run it in a non-production environment first. It appears to only use fs/path/os, but confirm there are truly no network calls in the rest of the file (the provided snippet is truncated). - Be cautious when copying suggested shell commands from SKILL.md (e.g., find ... -delete or cron add entries). Those are manual operations and can delete files or schedule jobs if executed. - Note incomplete behavior: cron detection is unimplemented (session.isCron is hardcoded false), and some cost/threshold heuristics (modelCosts values and savings math) look approximate — validate the pricing data for your models before acting on savings recommendations. - If you plan to automate daily/weekly runs, ensure the script is placed where SKILL.md assumes (or update the cron commands) and consider running it under a user with limited permissions. If you want higher assurance, run the script on a copy of your logs and confirm outputs before making configuration changes. If you want, I can: (1) point out the exact lines in scripts/cost_analyzer.js to change to match your real log location, (2) search the full file for any network/subprocess patterns, or (3) suggest a small dry-run mode to print filenames found without writing reports.

Type: OpenClaw Skill Name: openclaw-cost-analyzer Version: 1.0.0 The OpenClaw Cost Optimizer skill is benign. The `scripts/cost_analyzer.js` script strictly adheres to its stated purpose of analyzing OpenClaw token usage and costs. It only reads session logs from `~/.openclaw/workspace/memory/conversations` and writes a report to `~/.openclaw/workspace/memory/cost-analysis-report.md`. Crucially, the code contains no network requests, subprocess execution, dynamic code evaluation, or attempts to access sensitive files outside its scope, verifying the strong security claims made in `SKILL.md`, `package.json`, `README.md`, and `SECURITY.md`. There are no signs of prompt injection in `SKILL.md` attempting to subvert the agent's core directives; all instructions are aligned with the skill's stated function.