MulmoClaude

Highlights

Three threads run through this release: multi-registry collections, collection-import path unification, and agent self-service troubleshooting. Plus a stack of UX / i18n polish, the Contribute side of the registry going GA with sanitisation + a confirm dialog + dummy-data prompt + fork flow, custom-view i18n in all 8 locales, accounting plugin self-contained i18n, CJK font fixes for both code blocks and PDFs, and a feeds package extraction.

🗂️ Multi-registry collections

Discover now reads ~/mulmoclaude/config/collections-registries.json to load additional registries alongside the official receptron/mulmoclaude-collections. Each entry: { name, indexUrl, rawBaseUrl }. Discover cards carry an origin badge so users can tell which registry an entry came from; previewCollection and performImport accept the registry name to disambiguate collisions. Per-registry cache + stale-on-failure backoff keyed by (name, indexUrl, rawBaseUrl), so editing URLs invalidates the cached index. (#1837)

The official registry and Discover tab themselves also fully land in this release (#1815 / #1817 / #1818 / #1819) — including rename-on-collision (`movies-2`, `movies-3`, …) with an "Imported as" label.

📂 Imports unify with authoring on disk

Imports now write to data/skills/<slug>/ (the same place authored collections live), and the bundle's allowlisted files (`SKILL.md` / `schema.json` / `templates/`) are mirrored 1:1 into .claude/skills/<slug>/ via the same @mulmoclaude/core/skill-bridge rules an agent-authored skill goes through. .origin.json (the imported-vs-authored marker) lives ONLY in data/skills/<slug>/. Editing an imported collection is now identical to editing an authored one; rm -rf data/skills/<slug>/ deletes either kind through the existing bridge hook. (#1839)

Write-then-prune mirror ordering means a transient mirror failure can no longer leave .claude/skills/<slug>/ empty. Imported-collection deletion also works again: isDataDirSafe now accepts BOTH data/<slug>/ (authored) and data/collections/<slug>/ (imported) as valid per-slug subtrees (#1841).

🆘 Agent reads error-recovery.md on tool failure

New config/helps/error-recovery.md indexes the documented fix for common tool failures: gh/git/SSH inside the sandbox, Marp PDF, registry import, build/workspace, plugin runtime. The system prompt now points the agent at it BEFORE asking the user a clarifying question on a failed tool call. CLAUDE.md rule mandates appending new diagnostics to that file (so know-how that lives only in a PR description or docs/ stays accessible to the sandboxed agent). (#1846)

🤝 Contribute (registry export) goes GA

The Contribute icon on each Installed-tab collection card seeds a new chat with a sanitised template prompt; a confirm dialog wraps the icon so a stray click doesn't launch the agent. The agent generates 3-5 synthetic dummy records based on schema.json (privacy-safe by default — never copies real user records), builds the contribution bundle, runs build-index.mjs + validate.mjs, and opens a PR to the registry after user confirmation. Plus an export bundle generator with strict input validation. (#1825 / #1828 / #1830 / #1832 / #1835 / #1845)

🇯🇵 CJK fonts

• Code blocks (chat UI + 11 plugin Views + JSON editor): Windows CJK monospace fonts (MS Gothic, BIZ UDGothic) added to the monospace stack. Japanese inside code blocks now renders correctly on Windows (was tofu). (#1829)
• PDF rendering: Hiragino / Yu Gothic / Meiryo / Noto Sans CJK JP fallback added to MARKDOWN_CSS body + Marp inline style so PDF exports of Japanese decks render the glyphs. (#1826)

🧩 Other notable changes

• Custom-view i18n (#1842) — sandboxed views pick up the host locale via <meta name="mulmoclaude:locale"> and document.documentElement.lang.
• Accounting plugin self-contained i18n (#1838) — @mulmoclaude/accounting-plugin ships its own 8-locale i18n inside the package.
• Feeds package extraction (#1840 + #1843) — feeds engine + schema + paths moved to dedicated subpaths of @mulmoclaude/core. Shareable feed refresh registration so the same feed can be triggered from multiple surfaces.
• Field-driven spawn from a record field (#1820) — collection schema `spawn.every` accepts `{ fromField: "interval" }`.
• Imported-collection custom-view fallback (#1836) — view reader falls back to data/skills/<slug>/views/ for project-source collections.
• Custom view help — default fields hint (#1834) — the addView prompt clarifies referenced fields must exist on the schema.
• README refresh (#1849) — the npm-shown README catches up with the last year of features; /publish-mulmoclaude skill gains a step to verify the README each release.

Cascade publishes

• @mulmoclaude/core 0.2.7 → 0.2.12
• @mulmoclaude/collection-plugin 0.5.11 → 0.5.16
• @mulmoclaude/html-plugin 0.2.4 → 0.2.5
• @mulmoclaude/markdown-plugin 0.1.6 → 0.1.7
• Initial publishes: @mulmoclaude/[email protected], @mulmoclaude/[email protected], @mulmoclaude/[email protected]

📦 npm: `npx mulmoclaude@latest` — or `[email protected]`

Full changelog: https://github.com/receptron/mulmoclaude/compare/v0.9.0...v0.9.1