v0.0.1 — OuroSSS

OuroSSS — Autonomous Research Agent

Fresh start. Renamed from ouroboros, unforked, reset versioning.

What's inside

Telegram Bot — 19 commands for remote research ops

• Training control: /run, /stop, /logs, /metrics, /ckpt, /completions
• GPU/disk monitoring with auto-alerts (< 1.5T free triggers warning)
• Research logging to Notion (/research log, /eval push)
• Feature dispatch: /feature → GitHub issue → auto-dev agent

Research Pipeline Protocol (OUROBOROS.md)

• 7 stages: Ideation → Bootstrap → Infra → Training → Results → Paper → Release
• Each stage maps to concrete tools (skills, bot commands, CI)
• 10 operational rules (reproducibility, atomicity, ship-or-kill)

Model-Aware Task Dispatch (team/)

• Opus: planning, architecture, review
• Sonnet: implementation (>20 lines)
• Haiku: exploration, search, summarization

Infrastructure

• uv + pyproject.toml + uv.lock for dependency management
• Pre-commit ruff linting (auto-fix on commit)
• Branch protection (PRs required, lint must pass, admin bypass)
• GitHub Actions: lint, release, RTK badge, pipeline stage gates, stale issues, auto-dev dispatch
• RTK hook: 78% token savings (241K tokens saved)

Notion Integration

• Experiment Log DB (eval checkpoints, benchmarks, accuracy)
• Research Log DB (findings, milestones, progress — collaborator-facing)
• Bot commands for both: /eval, /research

Website Visibility Gate

• Public site (fr0do.github.io) auto-hides projects with private GitHub repos
• Client-side GitHub API check, fail-open

Issue Routing

• Infra → Fr0do/ourosss (GitHub issues)
• Research → project repos (Fr0do/s_cot, etc.) + Notion