Open Source · MIT · Agentic Operating Doctrine

OpenTenet

Your AI coding assistant is capable — but it's also amnesiac, overconfident, and stateless. It forgets everything between sessions, says "done" without checking, and hallucinates paths and functions that don't exist. OpenTenet fixes that with an operating doctrine your AI reads at every session start: a constitution of hard rules, a six-phase algorithm for real work, and a verification doctrine that will not let it claim success without proof.

View on GitHub → ← All Tools

How It Works

One Disciplined Loop for Every Non-Trivial Task

OpenTenet runs the same six-phase loop on every meaningful piece of work. Each phase has a job, and the loop cannot skip ahead: intent is captured before anything changes, risks are surfaced before execution, and nothing is marked complete without evidence. When a session ends, hard-won lessons are routed into a knowledge layer the next session reads — so the workspace gets sharper the longer you use it.

OBSERVE THINK PLAN EXECUTE VERIFY LEARN capture intent premortem map deliverables do the work prove it route lessons the workspace compounds — each loop feeds the next
The six-phase algorithm: OBSERVE → THINK → PLAN → EXECUTE → VERIFY → LEARN

The Moat

It Cannot Say "Done" Without a Tool-Verified Probe

Most AI failures aren't wrong answers — they're confident ones. An assistant tells you the tests pass, the file was written, the endpoint returns 200. OpenTenet makes that impossible to fake: every acceptance criterion is a single, binary probe (a grep, a test run, an HTTP check), and the criterion cannot be marked complete until the probe actually runs and its output is captured. "Looks good" is not a probe. That one rule is the difference between a system prompt with good intentions and an operating doctrine.

AI claims: "done" criterion marked pending Required probe grep · test · curl · read-back PASS ✓ evidence FAIL ✗ back to work
Every criterion maps to one binary probe. No probe, no "done."

Portable by Design

One Doctrine. Every Major AI CLI.

The discipline lives in plain markdown files, not in a plugin bound to one vendor. OpenTenet ships pre-wired config files for six assistants, all pointing at the same doctrine — so you can switch models or tools without rebuilding your operating rules. Your memory, your knowledge layer, and your constitution travel with you.

OpenTenet doctrine constitution · algorithm · verification · memory Claude Code Cursor Codex Gemini CLI Aider Copilot
Switch tools, switch models — the rules stay.

What's in the Box

Not Just a Prompt. Real Engineering Underneath.

Constitution

Hard NEVER / ALWAYS / BEFORE rules that fire before every action — the non-negotiable floor.

ISC Doctrine

Ideal-State Criteria: every goal split into binary, tool-verifiable checks. No fuzzy "acceptance."

Verification Table

Twelve artifact types each mapped to a required proof shape — file, command, endpoint, schema, and more.

Git-Native Memory

Decisions, learnings, and knowledge accumulate in version-controlled files future sessions read.

Optional Agentic Runner

A stdlib-only runner executes the six phases with real, workspace-confined tools and swappable backends.

Secret-Scan Pre-Commit

A real hook scans every staged commit for 16 credential patterns and validates frontmatter. Not a vibe.

# 30-second install
git clone https://github.com/MAGI-Systems-AI/OpenTenet my-workspace
cd my-workspace && ./bootstrap.sh

Give your AI an operating doctrine.

Free. MIT licensed. Zero dependencies. Works with every major AI CLI.

View OpenTenet on GitHub →