Highlights
Code lessons alongside your source (#753) — by @06FallingLight90
The code viewer now includes per-file explanations aligned with the guided tour. Tour-covered files get detailed sections; clicking a section highlights its source lines. Other files receive an overview. Lessons are generated when the tour is built or rebuilt, published after the graph is saved, and pruned when files leave the graph. Partial updates preserve existing lessons; a tour rebuild refreshes their line ranges.
Devin support and a stdio MCP server (#726) — by @isaiahfrantz
Devin can load the plugin's analysis pipeline and post-commit hooks. The new MCP server exposes six tools for existing graphs: status, search, explanation, change-impact analysis, onboarding, and dashboard startup. Each tool includes graph freshness information, and dashboard processes are stopped when the server exits. Setup instructions are in the MCP README.
Incremental analysis and symbol preservation
- Deterministic incremental preparation reconciles file changes, ignore rules, structural fingerprints, and imports. Updates preserve surviving architecture and tour content when those stages do not need to run (#682).
- Source-evidence checks protect existing symbols during incremental updates, and failed finalization restores the saved graph and fingerprints (#683, #753).
- Fingerprints match methods by their enclosing owner and name, keeping same-named methods in different classes distinct (#766) — by @thejesh23.
- Baseline batch merging, Python relative imports, Makefile target detection, and
tested_byrelationships for Swift, Rust, Ruby, and PHP receive correctness fixes (#564, #728, #725, #689).
Graph freshness and automatic updates
- The dashboard and graph-consuming skills report when saved analysis is stale (#558) — by @jiang4wqy.
- PostToolUse hooks read tool input from stdin and deliver additional context in the supported response format (#598, #609) — by @amitvijapur and @KumamuKuma.
- Commits containing only generated graph artifacts no longer trigger an auto-update loop (#634) — by @SulimanAbdulrazzaq.
- Unreadable files are reported separately from files without parser support (#686) — by @dajiaohuang.
Dashboard and large-repository support
- Force-layout computation runs in a Web Worker (#576) — by @KumamuKuma.
- Graph-processing improvements cover tour generation, parsing, community detection, filtering, and embedding search (#490) — by @DashBot-0001.
- A reproducible benchmark records deterministic analysis timings, memory, coverage, and integrity results for large repositories (#587) — by @KumamuKuma.
- Markdown files render in the code viewer, and graph-loading failures surface their HTTP errors (#577).
/understand --excludeaccepts user-supplied ignore patterns (#373) — by @yiziff. Starter ignore rules now cover additional Python, Rust, Ruby, and Swift tests and fixtures (#579, #582, #597, #600) — by @thejesh23.
Languages, installation, and documentation
- Vietnamese dashboard, skill, and README support (#709) — by @rgb-vgx.
- Kimi installation uses
.kimi-code/skillson Unix and Windows (#685, #687) — by @FumasterLin and @dajiaohuang. - Windows installation supports PowerShell 5.1 (#185) — by @Bortlesboat.
- Updated Codex and VS Code Copilot installation guidance, contributor test commands, translated READMEs, and onboarding output saved as
UA_ONBOARDING.md(#566, #756, #718, #567, #606, #637).
Standalone viewer
The attached understand-anything-viewer.tgz includes the dashboard from this release. Open an already analyzed project with:
npx https://github.com/Egonex-AI/Understand-Anything/releases/download/v2.9.8/understand-anything-viewer.tgz /path/to/analyzed/projectThe viewer requires Node.js 18 or newer. Workspace builds and the MCP server require Node.js 22 or newer and pnpm 10 or newer. Existing .ua/ and legacy .understand-anything/ data directories remain supported.
Full Changelog: v2.9.0...v2.9.8