Vai al contenuto
TakoAPI
Torna alle skill

token-optimizer

di smartpeopleconnected

Coding Agents & IDEs
5.5kDownload riportati da clawskills.sh
21Stelle
0Visualizzazioni
0 likeAccedi per mettere like a questa skill
—Non ancora valutato
La tua valutazione

Sintesi

Reduce OpenClaw AI costs by 97%.

Descrizione

Cosa fa questa skill

Reduces OpenClaw AI API costs by routing most requests to Claude Haiku, offloading heartbeats to local or alternative providers, enabling prompt caching, and trimming loaded context size. All changes preview before writing, and backups are created automatically before any modification.

Combines model routing, multi-provider heartbeat offloading, prompt caching, and config rollback into a single CLI with dry-run preview, rather than requiring manual edits across multiple config files.

Quando usarla

  • Switching routine coding tasks from Sonnet to Haiku to cut token spend
  • Routing OpenClaw heartbeats to a local Ollama instance instead of paid API
  • Previewing config diffs before applying changes in a CI pipeline
  • Setting daily and monthly spend caps to prevent runaway API bills
  • Rolling back to a previous config after an optimization produces unexpected behavior

Esempio di workflow

INPUT

User asks: reduce my OpenClaw API costs

AGENT
  1. 1

    Run python cli.py analyze to inspect current model routing and heartbeat configuration

  2. 2

    Run python cli.py optimize to generate a diff preview of proposed changes

  3. 3

    Review the diff showing Haiku as default model, caching enabled, and budget limits added

  4. 4

    Run python cli.py optimize --apply to write changes to ~/.openclaw/

  5. 5

    Run python cli.py verify to confirm the new configuration is active

OUTPUT

OpenClaw configured with Haiku as default model, prompt caching enabled, and daily/monthly budget limits set, with a timestamped backup of the previous config saved to ~/.openclaw/backups/

Installazione

clawhub install smartpeopleconnected/token-optimizer

Esegui il comando CLI oppure incolla il prompt in Claude Code, Cursor, Codex o nell'agente di coding che usi.

Link