the charter blog
charter read introducing-charter # opening introducing-charter.md

Introducing Charter: AI-agent readiness, scored

June 18, 2026 ·2 min read·411 words·Charter

AI coding agents are only as good as the repo they work in. Charter grades any repository 0–100 on how safely an agent can operate in it — offline, deterministic, and with a concrete fix for every gap.

launch
ai-agents
introducing-charter.md — charter 4 sections

An AI coding agent inherits everything about the repository it lands in. The good and the bad. Give it a clear AGENTS.md, pinned tools, and tests it can run, and it behaves like a careful senior engineer. Drop it into a repo with missing context, an unpinned MCP server, and a secret sitting in plain sight, and it will confidently make things worse — and you’ll find out in the diff, not before.

That gap is invisible today. There’s no number for “how ready is this repo for an agent.” So we built one.

What Charter measures

Charter is an offline CLI that scans a repository and returns a deterministic 0–100 readiness score across nine categories — context, secrets, MCP safety, agent config, environment, CI, testing, autonomy, and governance. Eighteen rules, each with a severity that sets its weight, and each paired with the exact thing it checks and how to fix it.

It runs in under two seconds on a 50,000-file repo, makes zero network calls, and never sends your code anywhere. The same check runs in your shell and in CI.

Why deterministic, not an LLM

It would be easy to point a model at a repo and ask “is this agent-ready?” We deliberately don’t. A score you can’t reproduce isn’t a gate — it’s a vibe. Charter’s score is a public formula over a fixed rule set: same repo, same score, every time. You can read the rules, predict the result, and trust the number enough to block a merge on it.

The loop

The workflow is the same three steps whether you’re cleaning up one repo or rolling Charter out across an org:

  1. charter init — scaffold the context files an agent expects.
  2. charter doctor — get an honest baseline, usually 40–60 on a repo that’s never seen this.
  3. charter fix — apply diff-first repairs for the safe rules, then gate it in CI so the score only ratchets up.

Nothing is written without a diff you approve. Secrets and destructive commands are never auto-touched.

The contract

Charter makes ten commitments and shows its work on every one — no network, no LLM in the core, no file deletion, no silent mutation, every finding traceable to a rule and a fix, every release signed. It’s Apache-2.0 and free, forever.

Run brew install use-charter/tap/charter and charter doctor on your repo. The first number is usually humbling. Fixing it is the point.