Turn any idea, plan, or codebase into a beautiful interactive diagram. An agent skill for Claude Code, Codex, and more.
HTML agent-skillsai-agentsarchitecture-diagramclaude-codeclaude-skillscodexcoding-agentsdeepseek-harnessdeveloper-toolsdiagramsdiagrams-as-codedsh-pluginflowchartllmmermaidopencodesequence-diagramsoftware-architecturesystem-designvisualization
Merge development base and preserve PR #498 fix
To3akaRin31b3904a659d
diff --cc CHANGELOG.mdindex 2441d432,88f64d7c..b41a4753--- a/CHANGELOG.md+++ b/CHANGELOG.md@@@ -4,11 -4,25 +4,29 @@@ All notable changes are documented here## [Unreleased]++### Fixed++++- **Workflow migration semantic contracts (#497).** Layout planning no longer checks full-topology requirements against a temporary graph with routed edges removed. Migrated documents retain their authored contracts and must pass complete validation before publication.+++ ## [3.0.0] — 2026-09-28++ ### Upgrading from 2.x++ - The Viewer has a redesigned reader layout. Guided/story views and ordinary share-card exports have been removed; use the diagram overview, Route, Lens and canonical exports. Already generated standalone HTML keeps its embedded viewer.+ - Existing schema-v1 diagram inputs remain supported. Lifecycle schema v2 is the recommended path for newly authored lifecycle diagrams; migration is not required to keep rendering v1 inputs.+ - English and Simplified Chinese remain built in. Supply `meta.translations` for other Viewer languages; authored node and message text is separate from Viewer UI translation.+ - Stable release identity is now `v3.0.0`; the previous `2.17.0-dev.1` identity was a development candidate, not a stable release.+### Improved reading+ - **Sequence width review.** `finalize` now reports measured unused right-hand space when fixed participant columns crowd one side of a wide canvas. New unpinned candidates get a bounded `column_fit: "spread"` repair step; explicit fixed and legacy layouts keep their geometry, and the advice adds no validation warning or failure.+ - **Lifecycle schema v2.** New lifecycle diagrams render one row per lane on a shared `col` 0–4 grid, so a state placed under the state it leaves gets a straight transition. The renderer sizes the canvas, widens column gaps for same-row labels, opens row gaps for the tracks their routes need, and routes automatic transitions with a dedicated orthogonal grid router (parallel reciprocal pairs, crossing-minimized tracks, corridor and outer-loop detours). Row titles move to a vertical left gutter, the implied phase rail is replaced by authored transitions, and state text is one step larger. Schema v1 files keep their exact state geometry.+ - **Lifecycle marks and legend.** Start states get a UML initial marker and states without outgoing transitions a double border, both explained in the legend; decision no longer shares failure's red, and success is green. The lifecycle type sigil moves to the top-left like every other mode.+ - **Whole diagram on the first screen.** With notes below the diagram, the adaptive reader now fits the complete diagram and its controls into the viewport down to the renderer's readable text floor, instead of holding primary labels at 12px and scrolling tall diagrams off a laptop screen; zoom restores detail.+ - **Sequence legend matches the drawing.** The legend sits 12px below all timeline content and may wrap into that room (an auto-sized canvas grows for it; a too-short authored showcase canvas fails with the exact height), lifelines stop above the legend instead of running through it, message labels take their line's legend color (gray lines keep readable muted text), and the legend names the emphasis variant "main request" / "主路径请求" and the default variant "message" / "普通消息".+ - **Readable source badges.** The Viewer's source badge is larger, fully opaque, sized to its text, and placed left of the brand mark; lifecycle, workflow, and data-flow label layout now reserve its width, so it no longer covers sigils or labels.- Showcase sequence message names use larger primary type and matching label plates; real browser readability checks now include message labels.-- > Development identity: `v2.17.0-dev.1`. Not a stable release.+ - **Clearer Architecture layouts.** Authoring defaults now classify each relationship (main path, branch or store, return, second entrance, fan-out) before placement, with a measured side length for fan-out ports. A passing `finalize` receipt with crossings adds node-move `hints`, backed by new `routeReview` evidence (`sharedNode` on crossings, `crowdedSides` when a side faces more neighbours than it has ports); the Skill allows one bounded, position-only repair for crossings. Automatic routing steps fan-out siblings into free parallel channels and reaches a blocked row of neighbours through one vertical side instead of wrapping around it.### Added- **Hermes Agent community opt-in.** `integrations/hermes-agent` is a Skill-only directory plugin: it registers the existing Node Archify `SKILL.md` for Hermes. The documented install is `hermes skills install skills-sh/tt-a1i/archify/archify -y` (published GitHub Skill). A checkout symlink remains the local-dev path. Hermes still runs `node bin/archify.mjs`. This is not an official Nous product and is not an agent-switcher target.diff --cc archify.zipindex fb50b539,52c08d7d..db23567bBinary files differ