OpenClaw Cost Guard (Budgets + Token-Saving Playbook)

**Quick judgement**: A focused and practical cost-monitoring and budget-enforcement skill for OpenClaw/Clawdbot setups. It is well-aligned with real-world usage (cron jobs, dashboards, alerts) and leverages session JSONL logs for more accurate cost tracking than context-window estimates. Best suited for users already logging sessions via OpenClaw/Clawdbot and comfortable running Python CLI tools. **Strengths** - Uses `usage.cost` from session logs when present, improving accuracy over heuristic token-based estimates. - Clear CLI workflows for daily/weekly reports, top expensive sessions, and JSON outputs (good for dashboards/automation). - Built-in budget checks using `--budget-usd` and exit codes make integration with cron and alerting straightforward. - Token-saving playbook provides concrete, actionable guidelines to reduce response length and tool-call overhead. - Backward compatibility with legacy Clawdbot paths eases adoption in older setups. **Risks / limitations** - Accuracy falls back to user-provided price environment variables when `usage.cost` is missing; misconfigured prices will skew reports. - Depends on the specific JSONL session schema and file layout; schema or path changes in OpenClaw/Clawdbot could break parsing until updated. - For very large log directories, performance may degrade if the script is not optimized for incremental or streamed processing. - The token-saving playbook trades detail for lower cost; in tightly constrained responses, there is a risk of under-explaining unless users opt into more detail. **Recommended scenarios** - Individuals or teams running OpenClaw agents in production who need simple, scriptable daily/weekly cost oversight. - Environments where exceeding a daily budget must trigger alerts or automation (e.g., messaging via Telegram or other channels wired to exit code `2`). - Cost-conscious users seeking to identify top expensive sessions and then refine prompts/agents accordingly. - Workflows that benefit from a "minimal tokens" mode, where the assistant is explicitly instructed to keep responses and tool usage short and controlled.