Petro Agent — AI skill for Claude Code
Autonomous petrophysical reports from raw LAS well logs - the LLM decides, a golden-tested deterministic engine computes, honesty is enforced.
How to install Petro Agent
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open OilCoder/petro-agent and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Petro Agent does
Autonomous petrophysical reports from raw LAS well logs - the LLM decides, a golden-tested deterministic engine computes, honesty is enforced. Full report on the site.
Alternatives in AI
- WeKnora — Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, an 26.9k ★
- Agentrq — AgentRQ: Human-in-loop realtime conversational task manager for AI Agents 1.1k ★
- Codex Skill — by klaudworks - Enables users to prompt codex from claude code 914 ★
README
petro-agent
**The LLM decides. The engine computes. Nobody lies.**
An autonomous multi-agent system that turns raw well logs (LAS) into complete petrophysical reports — Vsh, porosity, saturation, net pay, uncertainty — **with no human intervention in the process**. Every number comes from deterministic, golden-tested code; the language model only chooses methods, intervals and analyses, and writes prose. When the result isn't credible, the report says so on its front page.
📊 **Full report & interactive site:**
Highlights
- 0 numbers written by the LLM — a claim verifier ties every figure in the prose to a per-well JSON ledger, and rejects the rest (it happened live).
- 198 Kansas wells (Schaben field — chosen because it deceives: two thirds of the column reads like spectacular porosity and is shallow sponge; you can only pass by analyzing, not by computing).
- Engine validated against VOLVE's public interpretation (North Sea, untuned): r = 0.87–0.96; P10–P90 bands honestly recalibrated from overconfident (1.8–35% coverage) to calibrated (88–98%).
- Eight agent versions measured against a declared-contaminated expert baseline — the best final agent (claude-opus-4.8 + thinking) chose 4/4 zones 100% inside producing rock; the yardstick itself was stress-tested with three deterministic cycles.
- Total measured cost of the whole experimental arc: ~$50.
The site
| Page | What it tells |
|---|---|
| Home | The problem, the invariant, methodology, results, verdict |
| Petrophysical engineer | Can an AI read the formation? The rock-focused verdict |
| Software engineer | Architecture, the experimental lever phases, costs, transferable lessons |
| Journal | The evolution, ch |
Related Skills
Golden Thread
Vault-based AI memory for Claude Code, with hook-enforced Core rules
Agent Ready SEO
A Claude Code skill for making a site readable and citable by AI answer engines as well as search crawlers: th
Toon Bend
Byte-for-byte port of the toon CLI (toon_rust: JSON <-> TOON, the token-efficient format for LLM context) to B
Model Comparison
Local open-weight models (Qwen3.5-397B, MiniMax M2.5) vs cloud coding agents (Codex 5.3, Claude Code) on six w
Speedread Skill
Your AI agent reads the codebase. You speedread the report. An agent skill + one-file RSVP reader for markdown
Dev Report Weekly
Local-first desktop app that turns reviewed Git history into auditable weekly reports with Codex, Claude Code,
Related Agents
Reliability Engineer
Python delivery agent for quality gates and test infrastructure. Owns test suite (>=80% coverage enforced), go
Agent Performance Updater
Tier-1 Trade Evaluation. Computes per-agent (hit_rate, sharpe, pnl, n_samples) rows for the agent_performance
Test Architect Agent
You are a testing expert who designs comprehensive test strategies and writes effective tests. You ensure code
