Skip to content

CHANGELOG.md

Latest commit

 

History

History
58 lines (43 loc) · 3.38 KB

CHANGELOG.md

File metadata and controls

58 lines (43 loc) · 3.38 KB
Raw

Changelog

All notable changes to Light Skills are documented here.

The project follows a pragmatic release-note style rather than claiming strict semantic-versioning compatibility for every skill prompt. User-facing skill behavior, installation paths, shared contracts, and generated examples should be treated as release-relevant changes.

[Unreleased]

Planned

  • Continue hardening cross-client compatibility for Codex, Claude Code, and OpenCode.
  • Add more runnable examples for research, competition, software, patent-disclosure, and software-copyright workflows.
  • Improve regression checks for skill installation, evidence handling, citation integrity, and reproducible figure generation.
  • Collect public user feedback through GitHub Issues and convert repeatable problems into documented tests.

[1.0.0] - 2026-07-05

Added

  • Public Light Skills repository for research, competitions, and innovation projects.
  • 23 domain-independent skills under skills/, each using SKILL.md as the entry point.
  • Cross-client installation support for:
    • Codex via .agents/skills
    • Claude Code via .claude/skills
    • OpenCode via .opencode/skills
  • scripts/bootstrap_agent_skills.py for installing and checking supported agent-skill layouts.
  • Shared contracts under _shared/ for evidence strength, consistency checks, workflow gates, visual review, and reusable research-process rules.
  • End-to-end research demo under examples/e2e-research-demo/.
  • Reproducible LaTeX paper demo under projects/photocatalytic-dye-kinetics-study/.
  • Research figure gallery demonstrating programmatic scientific figures.
  • Python and R figure workflow expectations.
  • LaTeX typesetting expectations for paper-oriented workflows.
  • MIT license.

Changed

  • Reworked the earlier Light project into a cleaner public skill package.
  • Consolidated overlapping paper-writing, polishing, figure-planning, and figure-drawing flows into stronger user-facing skills.
  • Replaced private-database framing with public, repository-local, inspectable workflow guidance.
  • Expanded the scope from research-only usage to research, competitions, innovation projects, patents, software copyright materials, and frontend/software demos.
  • Clarified that Light Skills supports innovation work by improving search, critique, evidence binding, and validation planning; it does not promise automatic discovery or guaranteed publication.

Integrity and safety boundaries

  • Unknown facts, DOI values, links, software versions, and venue rules must be marked unknown or verified at task time.
  • Paper figures, data figures, and experiment figures must be generated programmatically and remain reproducible.
  • AI-generated marketing visuals must never be presented as scientific evidence.
  • Citation authenticity is a standalone integrity requirement.
  • Private conversations, unpublished research, credentials, email tokens, and local project ledgers must not be committed to the public repository.
  • Patent and software-copyright skills help prepare materials and organize evidence; they do not provide legal advice.

Compatibility notes

  • Windows is the primary development environment for the initial release.
  • Python 3.10+ is expected for repository scripts.
  • Windows users should set PYTHONUTF8=1 before running Python scripts.
  • LaTeX and R are optional environment layers for workflows that need paper compilation or R-based scientific figures.