メニュー
Monitor Amp Free quota and handoff before exhaustion. Use when user says quota-watch, check quota, or monitor usage.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Plan and consume execution metadata for bd/beads issues. Use when Codex inspects, triages, claims, tackles, or delegates beads; assign or read suggested agent type, model, reasoning effort, execution mode, and parallel group using the documented execution_* metadata keys.
Set up a private GitHub Dolt repo for personal beads and configure sync for the current project
Initialize or use beads in stealth mode for a project that does not use beads, with issue data stored in a separate syncable repository. Use when the user says "init with stealth", "stealth beads", "use beads without touching this repo", or asks for external BEADS_DIR setup.
| name | quota-watch |
| description | Monitor Amp Free quota and handoff before exhaustion. Use when user says quota-watch, check quota, or monitor usage. |
Periodically check Amp Free tier quota. Warn when low. Handoff before exhaustion.
Run scripts/check-quota.sh immediately. Output is: remaining total percent.
After approximately every 15 tool calls, run:
.claude/skills/quota-watch/scripts/check-quota.sh
Do NOT mention the check to the user unless a threshold is hit. Keep monitoring silent.
Parse the three numbers from the script output: remaining total percent.
| Percent | Action |
|---|---|
| ≤ 10% | Warn user: ⚠ Quota at {pct}% (${remaining}/${total}). Wrapping up current task. Then finish only the immediate task. |
| ≤ 5% | Stop all new work. Execute Emergency Handoff below immediately. |
When ≤ 5%, execute these steps in order — skip any that don't apply:
git add only files you changed, then git commit -m "wip: quota handoff"git pushbd update <id> --status in_progresshandoff tool with a goal summarizing: what was done, what remains, current branch, any failing tests.Keep the handoff goal under 3 sentences.
Each check costs ~1 tool call (~30 output tokens). At 15-call cadence that's ~7% overhead. This is acceptable for sessions > 50 tool calls. For quick tasks (< 20 calls), checking once on load is sufficient.