[carbogninalberto]

Problem

Round 10 S7 (plan 2B, UI part). PR #100 added workspace skills to the backend (/assistant-skills, load_skill, propose_skill), but Settings → Assistant still had a disabled Skills tab placeholder, and chat showed proposed and loaded skills only as raw tool JSON.

Stacked: this branch is based on carbninai/r10-s6-skills, so until #98 and #100 merge the diff also contains S4 and S6. This PR's own change is the last commit only.

Change

• Settings → Assistant → Skills (?tab=skills, &skill=<id> deep-links one skill):
  • A list showing each skill's scope (workspace or team), status, version in effect, pending draft, author kind, update time and an enabled Switch. It can be filtered by scope.
  • A create dialog with a scope picker. The picker uses listTeams, so private teams appear only if the user can see them.
  • An editor dialog with a Markdown Textarea and a preview tab using the existing sanitizing Markdown renderer. Saves are "Save as draft" or "Save and apply" for an active skill, and "Save draft" for a draft skill.
  • A review dialog that shows a draft as a line diff against the version in effect, or the full text for a new draft skill, then activates it.
  • Version history with show text, restore and per-version activate.
  • Delete behind the shared AlertDialog ConfirmDialog.
  • Members see the same views with no write controls. The backend still enforces every write.
• Chat (ToolCalls.svelte):
  • A completed propose_skill call renders as ProposedSkillCard: name, description, full body through the sanitizing renderer (no images), a "Draft — not active until activated" badge, the draft and active version, and "Review in Settings" deep-linking to the skill. The skill_id must be a UUID.
  • A completed load_skill call renders as one "Used skill X" line.
  • Both tools have labels.
• The tab state lives in the URL.
• skill-diff.ts is a small LCS line diff. Above 4M line pairs it falls back to "all removed, then all added".
• All new strings are in en, it, zh-CN and zh-TW.
• Docs: BE/docs/conversations.md (Skills → Settings UI) and TECHNICAL.md.

No migration and no backend change.

Verification

• npm run check: 0 errors. The 1 warning is pre-existing, in [workspaceSlug]/+layout.svelte.
• npm run test:unit: 110 passed, including the 4 new skill-diff tests.
• Playwright on port 4197, run with --workers=1 on assistant-skills.test.ts, assistant-settings.test.ts and assistant.test.ts: 36 passed. The new tests cover:
  • the admin flow: list, filter, enable, create a skill for a private team with preview and script sanitization, deep-link URL, review diff and activate, save as draft, show and restore a version, delete with confirmation;
  • members' read-only view;
  • a deep link to a draft skill, with no horizontal scroll at 390 px;
  • the propose_skill card (sanitized body, deep link) and the load_skill line.
• The new tests failed before the implementation (4 failed).
• node AGENTS/TOOLS/check-harness.mjs passed; git diff --check is clean.
• The secret scan is clean on this PR's commit. The outgoing range from main flags three test fixtures in the S4/S6 Go tests, which are not part of this change.

Limits

• Editing an active skill that has a pending draft starts from the draft's text.
• The diff covers whole lines only; name and description changes are shown as before → after.
• The editor is a plain Textarea: the project's RichEditor produces HTML, not Markdown.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T13:36:07.149689Z	895a89f	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.