AmazingAng/old-coder

An old coder's strategy for the agent era: don't read the code — make it run the gauntlet. Evidence-first development skill for coding agents, inspired by Uncle Bob.

What it solves

It addresses the productivity bottleneck and trust gap when using AI coding agents. Instead of requiring a human to manually review every line of generated code, it shifts the focus to verifying the code's behavior through a rigorous "gauntlet" of automated checks, allowing developers to trust the output based on evidence rather than manual inspection.

How it works

The project provides a set of instructions (skills) that guide a coding agent through a strict workflow:

  1. SPEC: The agent creates a test plan and concrete examples of expected behavior for human approval.
  2. RED/GREEN/REFACTOR: The agent follows a test-driven development cycle.
  3. GAUNTLET: The code is put through a series of checks, including full test suites, linting, complexity analysis, changed-line coverage, mutation testing, property-based tests, and security scans for secrets or risky packages.
  4. EVIDENCE: The agent produces a final report with real numbers and verification results that the human reviews instead of the code.

Who it’s for

Developers using AI coding agents (such as Claude Code, Cursor, Aider, or Codex CLI) who want to increase the reliability of AI-generated code without spending all their time reading it.

Highlights

  • Evidence-Based Trust: Replaces manual code review with a verifiable evidence report.
  • Agent-Agnostic: Works as plain markdown instructions for any agent that follows system prompts or rules.
  • Risk-Scaled Effort: The intensity of the gauntlet checks scales based on the risk level of the task (e.g., higher rigor for financial or concurrency logic).
  • Strict Honesty Rules: Prevents agents from weakening tests to pass or misreporting results, requiring unverified checks to be explicitly labeled.

Related

  • Project
    WILLOSCAR/research-units-pipeline-skills

    A framework for agent-assisted research that turns goals into verifiable, resumable runs with a strict audit trail of evidence and artifacts.

  • Dispatch
    Beyond Smarter Agents: Using Formal Verification Gates for AI Coding Loops

    Explore how shifting from prompt-based constraints to compiler-enforced type systems and formal verification creates 'structural backpressure' that forces AI agents to produce correct code.

  • Project
    QoderAI/better-harness

    A tool for analyzing and improving the workflow around AI coding agents, turning project and session evidence into prioritized improvements for the AI development loop.

  • Project
    Kritt-ai/open-kritt

    An open-source security research platform that orchestrates AI agents to find, validate, and rank vulnerabilities in codebases using focused, parallelized workflows.

  • Project
    astrio-labs/forall

    An AI-powered formal verification tool that automates the creation of specifications and proofs to provide machine-checkable evidence of software correctness.