Coding Agent Harness — Data skill for Claude Code

Nodaris coding-agent harness

**Your coding agent, with the habits of a senior engineering team.** The harness installs into the coding agent you already use (Claude Code, Codex, Gemini CLI, Cursor or OpenCode) and changes how it works on every task: it plans before it edits, proves its work before it calls it done, keeps secrets and personal data where they belong, asks before anything it cannot undo, and remembers what went wrong last time. It is built for any software: web apps, services, scripts, websites and video. Teams that handle patient or payment data get extra checks on top.

For a one-page tour, with the benchmark results against a plain coding agent, open [`docs/site/index.html`](docs/site/index.html) in a browser.

What changes when you use it

You ask for	Plain agent	With the harness
"Fix the export button"	Edits the first file that looks right and says it is fixed	Finds the cause, writes a failing test, fixes it, runs the checks, and shows you the passing output
"Add sign-in to the admin page"	Writes the feature in one pass	Writes a short spec and threat model first, builds it test-first, and has a separate reviewer check the diff
"Clean up the old build folders"	Runs rm -rf	Moves them to a trash you can restore from
"Deploy it"	Runs the deploy	Stops and asks you to approve that exact command once
A long session with helper agents	Helpers work without your context and spend tokens freely	Helpers get the project's rules and design system, report back at checkpoints, and stay inside a token budget you can see in a live panel
The same mistake next week	Makes it again	Recalls the lesson it recorded, and shares it with your team if you opted in

What it does for you

• Plans, then builds. Each request is routed to the right procedure (bug fix, feature, investigation, security check, release, video) with the steps in order.
• Proves it. A coding turn cannot end until a