Repolode
Submit
AmazingAng

AmazingAng/old-coder

An evidence-first skill for coding agents that makes them run test plans, mutation tests and checks, so reviewers approve a spec and read a report instead of the code.

View on GitHub
Stars
753
Forks
62
Open issues
4
Last commit
2 months ago

What is it?

Old Coder is a plain-markdown skill for coding agents such as Claude Code, Codex CLI and Cursor. It addresses the problem of trusting code you have not read: before coding, the agent writes a test plan with concrete examples of required and forbidden behavior, and you approve it. After coding, the agent runs a gauntlet of checks and produces an evidence report you can rerun with one command.

The gauntlet covers the full test suite, types, lint and complexity, changed-line coverage, mutation testing, property-based tests, real execution outside the test harness, dependency and secret scanning, and test-suite stability. Additional domain layers such as concurrency, UI, API compatibility and performance are selected per task based on risk. The skill also forbids weakening tests to pass, and requires unverified items to be labeled as such.

The repository also includes an optional companion skill, old-coder-api, for HTTP/JSON API design and review, and a demo rate limiter built under the workflow.

Key features

  • 01Requires the agent to write a test plan with concrete examples that you approve before coding.
  • 02Runs a gauntlet of checks including mutation testing and property-based tests.
  • 03Measures changed-line coverage so every new line is exercised by a test.
  • 04Labels unverified results as unverified rather than passing.
  • 05Scales the depth of checks to the risk of the change.
  • 06Ships an evidence report that can be regenerated with one command.

Use cases

  • Building features that touch payments, logins, user data or concurrency where code review alone is not enough.
  • Working with coding agents while keeping human review focused on specs and test evidence.
  • Adding an HTTP/JSON API with compatibility, authorization and idempotency checks via the companion skill.
  • Adopting a verification-first workflow in Claude Code, Codex CLI or Cursor.

Topics

#ai-agents#coding-agent#testing#mutation-testing#property-based-testing#tdd#code-quality#agent-skills#ai-coding#claude-code

FAQ

What is old-coder?
An evidence-first skill for coding agents that makes them run test plans, mutation tests and checks, so reviewers approve a spec and read a report instead of the code. Old Coder is a plain-markdown skill for coding agents such as Claude Code, Codex CLI and Cursor. It addresses the problem of trusting code you have not read: before coding, the agent writes a test plan with concrete examples of required and forbidden behavior, and you approve it. After coding, the agent runs a gauntlet of checks and produces an evidence report you can rerun with one command.
Is old-coder free to use? What license does it have?
old-coder is open source under the MIT license. Check the license terms before using it in commercial products.
Is old-coder still maintained?
Updated within 6 months. The last commit to old-coder was on Aug 18, 2026.
What can I use old-coder for?
Building features that touch payments, logins, user data or concurrency where code review alone is not enough. Working with coding agents while keeping human review focused on specs and test evidence. Adding an HTTP/JSON API with compatibility, authorization and idempotency checks via the companion skill. Adopting a verification-first workflow in Claude Code, Codex CLI or Cursor.
Where can I get old-coder?
The source code and installation instructions are on GitHub: https://github.com/AmazingAng/old-coder

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects