# Changelog

## 0.2.2 — 2026-10-07

- Normalize source-text line endings for stable content hashes while preserving the exact adapter-returned cache bytes.
- Record separate normalized-content and cached-byte hashes, preventing Windows text-mode writes from corrupting existing CRLF content.
- Separate the tested Harness `0.2.0-rc.2` source identity from the reviewed but unverified `0.2.1-alpha.1` API state.

## 0.2.1 — 2026-10-05

- Give competition archive metadata its own pinned citation identity in research task packets.
- Accept `archive_metadata` findings and reject archive/author claim misattribution while preserving unknown metrics and unread links.
- Document the citation contract and synchronize installation commands in all four READMEs.

## 0.2.0 — 2026-10-04

- Added a native DeepSeek Harness plugin with six research tools, including team-plan generation, tested against its official tool runtime.
- Added five provenance-linked competition scenarios and a five-role research workflow with dependency graphs and result validation.
- Scheduled daily cloud integrity/scenario checks and upstream-review maintenance at 09:00 Asia/Shanghai.
- Added an optional Windows daily Codex development runner with isolated worktrees and local execution logs.
- Updated README navigation, architecture visuals and plugin/workflow quick starts; preserved research versus official-score boundaries.

## 0.1.1 — 2026-10-04

- Added a dependency-free npm installer for the complete offline Codex skill.
- Added GitHub Packages publishing and public release tarballs for installation without registry credentials.
- Redesigned the README with a banner, badges, workflow diagram and quick-start navigation.
- Preserved the original archive snapshot and research workflow.

## 0.1.0 — 2026-10-04

- Added a Chinese agent-skill entrypoint for solution research, transfer and maintenance.
- Imported the pinned 717-competition archive with 4,724 original link records.
- Added lexical retrieval, historical rank/type filters, provenance and snapshot differences.
- Added original-body Kaggle/GitHub source caching, with legacy-topic support.
- Added 21 reviewed decision cards from 13 original writeup bodies and one author-code README.
- Added portable installation, evidence checks, tests, CI and review-PR archive maintenance.
- Kept scored execution with the user's chosen competition workflow.