Skip to content

About

No description, website, or topics provided.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

1 watching

Forks

Latest commit

 

History

193 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FlameMirror Autonomous Agent

FlameMirror is an autonomous code generation and validation engine designed to coordinate large language model (LLM) assistance, deterministic validation, and continuous learning. The system iterates through a five stage loop that mirrors real-world developer workflows:

  1. Plan — Build a prioritized execution strategy.
  2. Generate — Create code artifacts locally or through GraphformicCoder.
  3. Test — Run the project test-suite through the Crucible sandbox.
  4. Validate — Perform static checks before committing.
  5. Commit — Stage and record the successful work unit.

The repository couples this loop with a fuzzy guidance engine and optional ML backends so that experimentation can proceed safely without sacrificing repeatability.

Quickstart

pip install -e .
python run_agent.py --workspace ./demo-workspace --enable-ml --enable-fuzzy

The agent defaults to a dry-run Git workflow so it will never mutate your repository history unless you explicitly opt-in.

Features

  • Autonomous agent loop with injectable sub-systems for generation, validation, testing, and version control
  • GraphformicCoder Codex bridge with secure sandbox execution
  • Crucible reinforcement harness with automatic failure logging
  • Fuzzy guidance rules that dynamically reorder plan priorities
  • Documentation and notebooks showcasing the full system pipeline
  • Continuous learning hooks with checkpoint management
  • Benchmark tracking utilities for SWE-bench, BigCodeBench, HumanEval, and reliability gates (see docs/benchmarking.md)

Repository Layout

.
├── src/flamemirror/          # Core package
├── docs/                     # High level architecture references
├── examples/                 # Jupyter notebooks demonstrating workflows
├── tests/                    # Unit tests for each subsystem
├── training_problems/        # Seed tasks + automatic failure logs
├── checkpoints/              # Saved Crucible checkpoints
└── .github/workflows/ci.yml  # Ruff, mypy, pytest and coverage (>=90%)

Development

Install the optional developer dependencies to run linting and typing locally:

pip install -e .[dev]
ruff check src/flamemirror tests
mypy src/flamemirror
pytest

Validation with GitHub CLI

The GitHub Actions workflow mirrors the local validation gates (notebook, tests, linting, typing, security, Docker build). Trigger it directly from the command line with the validate make target:

make validate

This command requires an authenticated gh CLI session. It will dispatch the ci.yml workflow for the current commit and stream job results until the run is complete.

License

Released under the MIT License. See LICENSE for details.

About

No description, website, or topics provided.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Used by

Contributors

Languages