"API throttling prevention and recovery for AI agents. Apply on all agent-assisted work on the Griot repo to stay within LLM provider rate limits."
Install
npx skillscat add donartkins/griot/throttling-prevention Install via the SkillsCat registry.
SKILL.md
Throttling Prevention & Recovery
Recovery (on "request was throttled")
- PAUSE — stop making tool calls immediately.
- WAIT —
sleep 5minimum. - BATCH — combine remaining small operations into fewer, larger ones.
- VERIFY — confirm the last operation completed before continuing.
Prevention (mandatory)
- Batch file operations: read/write multiple related files in single calls.
- Use glob +
find/grepin single shell commands instead of many tool calls. - Consolidate writes: plan all changes, then execute in batches.
- Insert deliberate 2–3 s pauses roughly every 5 calls on long tasks.
- Check state before reading a file to avoid redundant reads.
- Move rapid-iteration work into executable scripts (single shell call).
- Checkpoint progress: write incremental state so recovery doesn't require re-reading everything.
Rules
- Never make 10+ rapid sequential tool calls without a pause.
- Never retry immediately after throttling.
- Never read the same file twice in quick succession.
- Never give up after one throttle error — implement recovery and continue.