know what you spend before you spend it
Open Source
Claude Code / Codex CLI / OpenClaw

cost prediction
for ai agents

What is Costea

A set of AI agent skills that track, analyze, and estimate token consumption across Claude Code, Codex CLI, and OpenClaw. Estimate costs before execution, review spending after.

Install

ln -s /path/to/costea/skills/costea ~/.claude/skills/costea
ln -s /path/to/costea/skills/costeamigo ~/.claude/skills/costeamigo

Skills

/costea

Estimates the token cost of a task before running it, then asks for your confirmation.

/costeamigo

Generates a multi-dimensional report of your historical token spending across all platforms.

Usage

  • costea refactor the auth module
  • costeamigo all
  • costeamigo claude
  • costeamigo codex
  • costeamigo openclaw
How It WorksPlatforms

Costea scans session history from all supported platforms, builds a task index using pure jq, then uses LLM reasoning to match your new task against historical patterns and estimate token usage, cost, and runtime.

Session JSONL (3 platforms) ↓ build-index.sh ~/.costea/task-index.json ↓ ┌───┴───┐ ↓ ↓ /costea /costeamigo

COSTEA

Agent Cost Receipt

2026-03-25 14:32:07

Task

Refactor the auth module

Input tokens12,400
Output tokens5,800
Tool calls14
Similar tasks matched3
Est. runtime~2 min

Provider Estimates

Claude Sonnet 4$0.38
GPT-4o$0.54
Gemini 2.5 Pro$0.29
Estimated Total$0.38

best price: Claude Sonnet 4

Confidence96%

Proceed? [Y/N]

POWERED BY /COSTEA SKILL

THANK YOU FOR BEING COST-CONSCIOUS

Requires jq and session history from at least one supported platform.·View on GitHub