Changelog

All notable changes to the diagram-kit plugin will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[0.3.1] - 2026-10-10

Diagram Kit 0.3.1 is the icon patch included in the Marketplace 8.1.1 release candidate; publication is tracked separately from this version entry.

Changed

• Prepare the next patch with the approved D brand icon: package editable SVG and 1024×1024 PNG, and point listing logo and composerIcon to the same ./assets/icon.png. Public skills and behavior remain unchanged; local acceptance and publication are tracked separately in the brand RUNBOOK.

[0.3.0] - 2026-10-09

Changed

• Use portable root plugin.json and skills/ for ChatGPT desktop and local Codex, keeping arch-diagram as the only public skill. OpenAI display information moves to extensions.com.openai, and obsolete Claude/Codex host wrappers retire; Claude support stops.
• Keep fact tracing, installed-resource location and the actual Python validator; repository development skills are separate in .agents/skills.
• Apply 0.2.0 → 0.3.0 under the owner's authorization to prepare versions before desktop acceptance. The plugin remains in its 0.x phase; the version-rule exception is recorded in the migration ADR. Publication and acceptance status are tracked in the readiness record. Upgrade and resource-validation steps are in the client guide.

[0.2.0] - 2026-07-22

Added

• Codex native wrapper — .codex-plugin/plugin.json + skills/arch-diagram/agents/openai.yaml, so the plugin installs and the skill discovers on Codex as well as Claude Code.

Changed

• 0.1.0 → 0.2.0 (backward-compatible minor) — host-neutral runtime: arch-diagram resolves the bundled scripts/validate_diagram.py from the current SKILL.md locator (realpath-contained), not ${CLAUDE_PLUGIN_ROOT}, passing the absolute script path + each .md as separate quoted arguments.
• Dual-host invocation — Claude /diagram-kit:arch-diagram, Codex $diagram-kit:arch-diagram; internal routing uses bare qualified skill names, not host-specific slashes.
• Interpreter probe order aligned to python3 → python → py -3.

Fixed

• validate_diagram.py UTF-8 output on non-UTF-8 locales — the linter prints CJK (diagram names, rule messages, the 共扫描 N 张图 tally) and the scanned file path; on a Windows pipe/console whose active code page is not UTF-8 (e.g. cp1252), that raised UnicodeEncodeError and aborted the lint. The CLI entry point now reconfigures stdout/stderr to UTF-8, so linting a diagram whose path or content contains non-ASCII characters works on any platform locale.

[0.1.0] - 2026-06-05

Added

• Initial release of diagram-kit — architecture / UML diagramming plugin, functionally orthogonal to learn-kit. Single skill arch-diagram.
• arch-diagram skill (/diagram-kit:arch-diagram <target>) — turns a codebase / system's source facts into evidence-bound Mermaid diagrams across 7 types: context / container / component / code (C4 structural L1–L4) + sequence / state-machine (behavior) + deployment (physical), for any domain (docker / python / postgreSQL / claude-code-plugin / ...). 5-step fact-first flow: scope + domain auto-detect → acquire facts via the L0–L3 ladder (L0 声明扫描 → L1 结构推断 → L2 命名归类 → L3 HITL 补缺) → pick high-value types via the §5.1 applicability matrix → draft Mermaid (§5 edge semantics + §6 naming) → validate-fix-repeat. 铁律: every node/edge traces to file:行号 evidence, never fabricated. Frontmatter name + description + allowed-tools (Read / Glob / Grep / Bash / Write / AskUserQuestion); no MCP / no network.
• 9-file references/ bundle (progressive disclosure, one level deep) — architecture-methodology.md (4+1 视图 / C4 / §4.1 Mermaid 语法 / §5 边语义) + domain-acquisition.md (bridge + §2 六类画像 + §3 L0–L3 阶梯 + §5.1 全局适用性矩阵 + §6 命名唯一事实源) + 7 per-type drawing prompts. Domain-agnostic; introduced verbatim from a self-contained vault corpus.
• scripts/validate_diagram.py — pure-stdlib (Python 3.7+, no pip) Mermaid linter. Generalized from a PostgreSQL-specific version: removed the PG-only role↔shape map (ROLE-03), widened the slug regex to the general baseline (struct-l[1234] | dyn | phys, per domain-acquisition §6), and added the classDiagram naming gate. Validates generated diagrams (NAME-01/02 / MM-01 / CLS-01·02·03 / LEG-01 / TXT-01·03 / STATE-01·02 / SEQ-02·03·04). Interpreter detection (python → python3 → py -3) + graceful degradation when no Python.

Notes

• Output convention: generated diagrams use ```text fences (NOT ```mermaid) so they display as source — preserving %% Name / %% Slug metadata, team line-style legend, and dual-labels — rather than auto-rendering.
• Initial version 0.1.0 (not 1.0.0): signals "functionality not yet settled" — the classDiagram structural lint gap (only the naming gate runs on code diagrams; full structure lint is a 0.x → 1.0.0 item) + marketplace-domain dogfood pending + awaiting field evaluation. Mirrors learn-kit's own 0.1.0 origin.
• Validator dual-source fork: this generalized validator now evolves independently from the PostgreSQL handbook's validate_diagram.py; PG-specific tightenings will not flow back.
• Design, the "why a second plugin" reconciliation (marketplace's first 1 → 2 plugin count), and the validator-generalization decisions are recorded in marketplace-layer docs/adr/[ADR]_Diagram_Kit_Addition.md.