Skip to content

egohygiene/reflector

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

339 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

reflector

Reflective synchronization systems for recursive AI-assisted software engineering.

Python License REUSE Status Research

Build Paper GitHub Pages REUSE Compliance arXiv Ready Hugging Face Ready DOI

reflector publication preview

What is reflector?

reflector is an open-source research repository and recursive engineering platform for studying how AI-assisted software systems can stay aligned, auditable, and governable as they iterate. The repository combines:

  • a research manuscript in paper/
  • a synchronization-oriented CLI in reflector/
  • deterministic publication workflows in scripts/ and .github/workflows/
  • specification contracts in specs/
  • a GitHub Pages publication surface in docs/

If you want the short version: reflector treats recursive development as a systems problem, not just a prompting problem.

Why it exists

Recursive AI workflows can move faster than humans can inspect them. reflector exists to make those workflows easier to reason about by introducing bounded execution, explicit checkpoints, inspectable metadata, and publication-grade traceability.

Core concepts

Concept Meaning in reflector
Recursive Drift Failure modes that emerge when recursive optimization loops move away from human intent.
Reflective Auditing Continuous validation of state, artifacts, and workflow contracts against declared invariants.
Synchronization Checkpoints Explicit handoff boundaries where human review or deterministic validation can gate progress.
Human Governance The repository treats human oversight as a first-class systems constraint, not an afterthought.
Mixed-Initiative Systems Humans, automation, and AI agents collaborate through bounded, reviewable execution layers.

How reflector works

semantic content and metadata
            ↓
specifications and synchronization contracts
            ↓
audits, validation, and milestone checkpoints
            ↓
publication orchestration and deployment
            ↓
repeatable artifacts for research, review, and release

The repository architecture is intentionally layered:

  • Content — manuscript sections and semantic paper structure in paper/sections/
  • Metadata — canonical publication and repository metadata in paper/macros/ and metadata/
  • Style — renderer-facing presentation assets in paper/styles/
  • Orchestration — build, audit, and deployment logic in scripts/, reflector/, and .github/workflows/

Quick start

Recommended local workflow

reflector's canonical developer workflow uses task with uv:

task setup
task doctor
task test
task examples

Direct Python fallback

If you do not use task or uv, you can still install the repository directly:

python -m pip install -e '.[dev,huggingface]'
reflector --version
reflector huggingface --check-sdk

Build the paper

./scripts/build-paper.sh paper

To publish a locally built PDF into docs/:

./scripts/build-paper.sh paper --publish

Build the visual magazine artifact

task magazine:doctor
task magazine:build
task magazine:build:print

See magazine/README.md for the full workflow and output paths.

Repository map

Need Start here
Canonical onboarding 00-README.md
Repository architecture docs/architecture-overview.md
Publication architecture docs/publication-architecture.md
Workflow overview docs/workflows.md
Toolchain requirements docs/toolchain.md
Paper overview paper/README.md
Magazine overview magazine/README.md
Publication specs specs/publication/
Research notes docs/research/
Roadmap ROADMAP.md

Research paper and publication links

Citation

Use the version DOI when citing a specific archived release:

10.5281/zenodo.20477044

CITATION.cff is the canonical machine-readable citation source for GitHub and downstream tooling.

Publication

Hugging Face readiness

reflector is not publishing to Hugging Face yet, but it is now structured to make that step straightforward later:

  • canonical repository metadata already includes a Hugging Face integration surface in metadata/repository.yaml
  • the CLI exposes reflector huggingface for scaffold inspection
  • recommended publication and mirroring strategy is documented in docs/huggingface.md
  • a future-facing card scaffold is available in README_HF.md

Contributing and project governance

Long-term vision

reflector aims to become a durable reference architecture for recursive engineering systems: publication-aware, specification-driven, synchronization-first, and legible to both humans and automation. The near-term roadmap is to complete the manuscript, harden the audit and release paths, and prepare clean distribution surfaces for arXiv, GitHub Pages, and future Hugging Face publication.

About

🪞 Reflective synchronization architectures for recursive AI-assisted software engineering.

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors