Status: Evaluation window (private under 0SxD). This repository is staged for evaluation review. License terms in LICENSE govern; contents may move, change, or be withdrawn. See LICENSE before any use.
This repo is agent-creator-agent. Fork it, rename the clone ce-rd-os, and
you have CE_RD_OS v0.1.0 -- a compound-engineered R&D operating system. The
agent-creator-agent is the pattern; ce-rd-os is the first product it produces.
The fork-and-rename ritual is in INSTALL_BOOTLEG.md.
A setup-first, host-agnostic, model-agnostic scaffold for compound-engineered research and development. Conformant to the AGENTS.md spec (Linux Foundation Agentic AI Foundation) and the Agent Skills spec (Anthropic, agentskills.io).
v0.1.0. First public ship. Self-verifying, human-gated. Apache-2.0 for code,
CC-BY-4.0 for documentation. See LICENSE and LICENSE-DOCS.
CE_RD_OS provides six composable skills that any AGENTS.md-aware coding agent or any text-bundle host (NotebookLM, custom Gem, custom GPT, Claude project) can run as a research and development scaffold.
The six skills:
- system_directive: identity, source hierarchy, 100 percent confidence loop, Trinity rubric reference, output contract.
- setup_intake: captures the irreducible parameters (mission, role, sources, rubric mode, sub-agent profile, memory option, ML backend option, loop bound) before any work begins.
- source_ingestion: detect, normalize, interrogate, score, route incoming sources. Quarantines anything that cannot be conformed.
- trinity_rubric: 9-cell evaluation grid (Pathos, Ethos, Logos by Coverage, Self-contained, Importance) grounded in arXiv:2507.17746 (Rubrics as Rewards). Default mode is boolean-atomic. Optional ML backend hook for arXiv:2507.19457 (GEPA via MLflow) or OpenEvolve.
- workstream_creator: meta-skill. Generates the workstream MD that the user fills with their research bundle, then routes to the other skills.
- publish_bundle: agentskills.io conformant packing, AGENTS.md generation, dual-track output (GitHub repo plus 10-file text zip).
git clone https://github.com/0SxD/ce-rd-os.git
cd ce-rd-os
bash scripts/verify_self.sh
Open the repo in your AI coding environment. The agent reads AGENTS.md
on its own and proceeds through the skills in dependency order.
cd ce-rd-os
bash scripts/pack_text_bundle.sh
Output: dist/ce-rd-os-v0.1.0-text-bundle.zip. Drag and drop into your
host. Then paste the prompt from packets/PROCEED.md to start the
multi-turn setup.
Two commands cover the common case:
bash scripts/verify_self.sh
bash scripts/pack_text_bundle.sh
The first verifies the repo is publish-ready. The second produces the drop-in zip for any text-only host.
This bundle composes (does not copy) from the following Tier-1 sources:
- agents.md, Linux Foundation Agentic AI Foundation, AGENTS.md spec.
- agentskills.io, Anthropic, Agent Skills spec, Apache-2.0 / CC-BY-4.0.
- arXiv:2507.17746, Rubrics as Rewards (RaR), MIT-affiliated authors.
- arXiv:2507.19457, GEPA, plus github.com/gepa-ai/gepa under Apache-2.0.
- arXiv:2510.07743, OpenRubrics.
- github.com/EveryInc/compound-engineering-plugin, MIT, plan-work-review-compound pattern, attribution-only.
- github.com/letta-ai/letta, Apache-2.0, agent runtime architectural reference.
- github.com/mem0ai/mem0, Apache-2.0, memory layer reference.
Full upstream lineage in NOTICES.md.
Code: Apache-2.0. Documentation: CC-BY-4.0. Copyright 2026 0SxD.
This repo is dual-licensed. Apache-2.0 governs code and shell scripts;
CC-BY-4.0 governs documentation, SKILL.md bodies, and references. See
LICENSE and LICENSE-DOCS for full text.
See CONTRIBUTING.md. Pull requests and issues welcome. The repo runs
scripts/verify_self.sh on every push and PR; the gate must pass before
merge.
SemVer 2.0.0 plus Conventional Commits. Annotated tags. Release notes via
gh release create. See CHANGELOG.md.
End README.