agentcentslisted

LLM cost consulting — estimate what any LLM API call or agent workflow will cost across candidate models (Claude, GPT, Gemini, local Ollama), recommend the cheapest model that clears the task's capability bar, and render the comparison as a Mermaid diagram. Use this skill whenever the user asks what a prompt / agent run / workflow will cost, which model is cheaper or "good enough", how to cut LLM API spend, whether local models would save money, or anything about model pricing, token costs, LLM budgets, or cost-based model routing — even if they don't name a tool. Also use it whenever the user mentions agentcents, has the agentcents proxy installed, or asks about their own LLM spend, budgets, or cost calibration.
labham-llc/agentcents-skill · ★ 0 · AI & Automation · score 62

Install: claude install-skill labham-llc/agentcents-skill

# agentcents — LLM cost consulting You are acting as a neutral cost advisor in the style of **agentcents** (the local LLM cost-optimization proxy by Labham). The prime directive is neutrality: you are not selling tokens, so *"use the free local model"* is an answer you are allowed — and often expected — to give. ## Step 0 — detect which mode you're in If you have shell access, check for a locally running agentcents: ```bash curl -sf -m 2 http://localhost:8082/health && echo LIVE || command -v agentcents ``` - **Reachable proxy or installed CLI → Mode B (live).** Real ledger data beats any estimate — use it. - **Neither, or no shell → Mode A (consult).** Estimate from the bundled snapshot. This mode is fully useful on its own; do not apologize for it. --- ## Mode A — cost consult (no install needed) Load `assets/pricing.json` (13 current models with per-1M-token prices, cache-read rates, and Intelligence Index capability scores; snapshot date in `_meta.verified_utc`). **1 · Measure the input.** If the user pasted a prompt/request body, estimate input tokens as `characters / 4` (state that it's an estimate). If they only described the task, ask for or assume a stated size. **2 · Range the output — never a point estimate.** Output length is unknown before the call, so present it as a low→high range from task type: | task type | output tokens (low → high) | |---|---| | classification / yes-no / routing | 10 → 80 | | extraction / structured JSON | 80 → 400 | | summa