# CodeFRAME

> Regression memory for coding agents that records fixed bugs as requirements, locks them to affected code, and checks them before every commit.

CodeFRAME is an Apache-2.0 open-source CLI and Claude Code plugin (proof9) from Frank Bria that gives coding agents memory of bugs that were already fixed. A defect is recorded as a requirement, scoped to the code it affects, and checked before each commit and before merge. It is in public beta (0.9.8).

## What It Is

CodeFRAME is a command-line tool (`cf`) for coding-agent workflows, centered on PROOF9, a regression ledger. Each captured bug becomes a requirement with a named test that proves it stays fixed. The check fails if that test is deleted, skipped or has its body rewritten, and releasing a requirement takes a recorded waiver with a name, a reason and an expiry date.

The same CLI covers the work before and after the code: Socratic PRD discovery, task decomposition with dependencies, delegating tasks to Claude Code, Codex, OpenCode, Kilocode or a built-in agent, and opening pull requests that carry a proof report.

## How the proof loop works

`cf proof capture` records a defect and generates obligations and draft test stubs. `cf proof run` checks the requirements touched by uncommitted changes, or all non-waived requirements with `--full`. `cf proof ratchet` locks in currently passing tests as requirements. Nine gates exist (unit, contract, e2e, visual, a11y, perf, sec, demo, manual); unit, contract and sec run automatically, while users implement the generated stubs for five others. Evidence is checksummed and bound to the run and gate, and a run that verified nothing is reported as vacuous rather than passing.

## In Claude Code and at merge

The proof9 plugin adds `/proof` to capture a bug or lock in passing tests, checks the ledger before every commit Claude makes (warning by default, with an opt-in block mode), and shows failing requirements above the prompt. `cf pr merge` refuses to merge while an in-scope requirement is open; overrides require a reason and are audited.

## Current Status and limits

The project documents public beta status, and on-disk `.codeframe/` formats may change between betas. It states it does not judge whether a test is meaningful, fingerprints only the named test's own body, does not automate post-merge glitch capture, and is not enterprise-ready (SQLite only, no SSO, RBAC or multi-tenancy). It needs Python 3.11+, git, and an LLM provider key (Anthropic by default; OpenAI, Ollama and other compatible endpoints are supported).

## Features
- PROOF9 regression ledger that turns bugs into locked requirements
- Pre-commit check that detects deleted, skipped or rewritten tests
- Nine verification gates with checksummed evidence
- Waivers with name, reason and expiry
- Socratic PRD discovery and stress testing
- Task decomposition with dependency graph
- Agent-agnostic execution via Claude Code, Codex, OpenCode, Kilocode or built-in agent
- Batch execution with git worktree isolation
- Human-in-the-loop blockers
- Pull requests with proof report and merge gate
- proof9 Claude Code plugin
- TUI dashboard
- Multi-provider LLM support
- Opt-in telemetry

## Integrations
Claude Code, Codex, OpenCode, Kilocode, Anthropic, OpenAI, Ollama, vLLM, LM Studio, GitHub, pytest, ruff, bandit

## Platforms
WEB, API, CLI

## Pricing
Open Source, Free tier available

## Version
0.9.8

## Links
- Website: https://codeframe.sh
- Documentation: https://github.com/frankbria/codeframe/blob/main/docs/CLI_GUIDE.md
- Repository: https://github.com/frankbria/codeframe
- EveryDev.ai: https://www.everydev.ai/tools/codeframe
