Research OS Studio v2.0.0

Research OS Studio v2.0.0

Version 2.0.0 replaces the conversational entry with skill-layer orchestration, adds explicit authorization gates, and lands bounded acceptance evidence across three hosts.

What changed

• Route-only entry. A single Research OS entry routes to the three top-level workflows: Idea Discovery, Validation, and Paper Writing and Improvement. Workflows stop after their authorized scope. Nothing advances to the next stage without explicit authorization.
• Authorization gates. Default compute policy is no longer treated as an execution grant. Batch size, density, and side-effect boundaries require explicit pre-authorization (see docs/adr/0006-gate-density-batch-preauthorization.md).
• Explicit invocation contract. Entry and host mapping are specified in docs/adr/0007-explicit-invocation-and-research-os-entry.md.
• Domain glossary. CONTEXT.md became GLOSSARY.md, with tightened terms for Workflow, Internal Skill, Claim, and Evidence.
• Host-native invocation docs. Logical /name examples are now separated from native forms: pi /skill:name, Claude Code /name, Codex installed-skill naming.
• Maintainable diagrams. README architecture now uses workflow-v2.svg with companion Mermaid. Independent entries, authorization gates, and candidate manuscripts replace unconditional stage transitions and publication-ready implications.

Breaking changes

• The old ask-research-os skill is removed. Skills that were previously hidden now require host-native commands (see above).
• Workflows no longer promise unconditional continuation or a publication-ready manuscript. Leaf approval gates remain active.
• Default compute policy does not authorize execution. Grants come from explicit user authorization per batch.

Verification

Evidence is archived under docs/acceptance/. Scope and limits are documented in docs/v2-release-acceptance.md.

• Static checker: 39 skills PASS (U38/M1). Unit and script suites: 74 tests PASS.
• Three-host regression (pi, Claude Code, Codex): 39 × 270 install files byte-identical across hosts; 79-round host acceptance with recorded failures and retries; gate-combination suites on the merged tip.
• Security static scan across the release packages: 898/898 payload hashes match, no real credential candidates found.
• These are bounded behavior checks, not certification of scientific correctness on real projects.

Known notes

• The evidence archive exceeds GitHub's recommended 50 MB push size. No archive rewrite or LFS migration was performed.
• Some host-specific limitations were disclosed during acceptance and are listed in docs/acceptance/v2-merged-host-smoke.md.

v1.0.0

research-os-studio

1.0.0

Major Changes

• 0d6b42f Thanks @toRolex! - Initial major release of Research OS Studio:

  • 39 self-contained agent skills across General, Idea Cycle, Validation Cycle, and Writing Cycle
  • Human-in-the-loop workflows with bounded execution and zero unauthorized cross-stage progression
  • Direct installation via standard Skills CLI (npx skills add toRolex/Research-OS-Studio --all)
  • Comprehensive static validation and integrity check scripts
  • Documentation and guides in English and Chinese