[hsuchifeng]

Summary

Fixes three Dashboard bugs. I found all three by running the real dashboard against a sandbox team repo and driving the page in Chromium.

1. Distribution could not preview or save for many teams. The page sent primaryRole and projects on every preview and save. A team with no manifest/projects.yaml got a 409 (This team repo defines no projects). A team with no roles.yaml got the same error for roles. A workspace with no primary role got a 400 (Expected { primaryRole: string … }), even when only the projects changed. Because Save needs a preview first, these workspaces could change nothing. The page now sends only the halves the team defines and the draft changed. When only additional roles are picked, it asks for a primary role first.
2. Copilot sessions outside every project were in no workspace. workspaceEvents sends a session that matches no project to unassigned. But dashboardWorkspaces listed the Unassigned bucket only when an unowned event had a cwd. Copilot events record no cwd, so such a session never appeared under any option in the picker. The bucket now uses the same per-session rule.
3. Stale pages after a write. After a Distribution save, the Resources page kept showing the old skill and rule delivery until you switched workspace. After a skill exclusion, the Distribution counts stayed stale. Each write now reloads the other page.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Test Plan

• npx tsc --noEmit passes
• npm run lint passes
• npx vitest run src/__tests__/dashboard* passes (298 tests)
• npm run test:e2e -- dashboard-distribution-ui dashboard-distribution dashboard-unified passes (5 tests)
• Added a test: dashboardWorkspaces lists the Unassigned bucket for a cwd-less Copilot session. It fails before the fix and passes after.

Real-CLI record (after npm run build). The sandbox team repo had skills/{frontend,backend}/*, root and namespaced rules, manifest/roles.yaml and manifest/projects.yaml. Setup ran teamai init https://github.com/your-org/your-repo.git --provider git --agent claude --role frontend, then teamai pull, then teamai dashboard --port 4791. I drove the page with headless Chromium:

• Before the fix: POST /api/distribution/preview {"primaryRole":null,…,"projects":["pay"]} returned 400. With projects.yaml removed, a role-only preview returned 409.
• After the fix, with no primary role: ticking only "Additional" and clicking Preview showed "Pick a primary role first.". Ticking project pay gave a preview of skills: +backend. Save showed "Saved". The config on disk then held projects: [pay]. The Resources page showed the new delivery right away (api-kit Delivered).
• After the fix, with projects.yaml removed: picking the frontend primary role, previewing and saving all worked. The config on disk held primaryRole: frontend.
• After the fix, with only a Copilot event (cwd: "") in events.jsonl: /api/workspaces now lists unassigned, and /api/sessions?workspace=unassigned returns the session.
• Skill exclusion (POST /api/resources/skills/exclusions) writes excludedSkills to the workspace's own config, and the effective Distribution then drops the skill.

I covered only the Claude agent and the git provider. I did not run the other agents or providers locally and am leaving them to CI.

Notes for Reviewers

• The Distribution fix is client-only (src/dashboard/client.ts). The server's validation is unchanged.
• One known gap, left out to keep this PR small: the exclusion endpoint and the Distribution save use separate write queues. Two writes to the same workspace config that land at the same moment could still drop one of them.

---

Generated by Claude Code

[github-actions]

Code Erosion Report

Informational — never blocks the merge. scb-check==0.2.0 SlopCodeBench metrics; method & caveats in docs/dev/ci-code-erosion.md.

Metric	Value	Human band	Agent band	Reading
Verbosity*	0.064	0.15	0.33	human
Erosion	0.710	0.31	0.68	agent
Cognitive erosion	0.878	–	–	–

Scanned 386 files / 86114 SLOC · high-CC functions 662/7247

* On TypeScript, verbosity is partial and the bands are Python-calibrated.

Rule hits (TS verbosity layer)

Standalone ast-grep, separate from the number above.

Rule	Hits
unnecessary-else-after-return	53
empty-catch-block	5

58 distinct lines flagged across src/.

[hsuchifeng]

Codex PR Review failed in the openai/codex-action setup step, before any review ran: Failed to read server info from /home/runner/.codex/<run>.json. The Codex server never started on this fork, most likely because the fork has no Codex/OpenAI credentials configured. The failure is not caused by this PR's diff, and no code fix exists for it. I am not re-running it, because a re-run would fail the same way until the action is configured on the fork. Lint & Test is the check that covers this change.

---

Generated by Claude Code