zown-gemini-governor
por gtovd
Coding Agents & IDEsResumo
A high-fidelity token management and model stabilization skill.
Descrição
O que esta skill faz
Manages Gemini token usage to prevent TPM rate limit errors during long or complex tasks. Implements context pruning and enforced cool-downs between heavy prompts. Each task step is executed and verified independently under an atomic pipeline model.
Prevents mid-task Gemini TPM errors that would otherwise interrupt long-running pipelines and require manual recovery.
Quando usá-la
- Avoiding 429 rate limit errors during multi-step code generation
- Compacting agent context files before a long engineering session
- Enforcing cool-down pauses between back-to-back heavy prompts
- Syncing agent memory across restarts or context compactions
- Running one-shot Q&A queries without accumulating session context
Exemplo de fluxo de trabalho
User asks: Run a multi-stage refactor of the authentication module
- 1
Check current TPM usage via session_status
- 2
Prune SOUL.md and IDENTITY.md, summarizing history into MEMORY.md under 500 tokens
- 3
Execute stage 1 as a single atomic step and verify output before proceeding
- 4
If TPM usage exceeds 50%, run python3 scripts/cooldown.py 60 before the next stage
- 5
Repeat atomic execution and threshold checks for each remaining stage
All pipeline stages completed without hitting Gemini TPM rate limits
Instalação
clawhub install gtovd/zown-gemini-governorExecute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.