…ecvqholN2wcep)

The home 示例提示词 card for 交易分析仪表板模板 rendered a gray broken
frame instead of the template preview. Root cause: the daemon's sandboxed
preview route serves example.html with `default-src 'none'` CSP (no
frame-src), so a nested iframe can never load; the sanctioned pattern is
the iframe-only shell, which iframeOnlyHtmlShellTarget unwraps so the
inner template is served directly. This template's example.html wrapped
its iframe with a header strip, defeating the unwrap — the CSP-blocked
frame then baked into the CI gallery poster as a broken gray image.

- Rewrite example.html (both the plugins/_official copy and the
  design-templates mirror) as a bare iframe-only shell, matching the
  proven flowai-live-dashboard-template shape. The served preview now
  unwraps to the self-contained assets/template.html.
- Drop the stale broken bake entry from data/plugin-previews/manifest.json
  so clients fall back to the (now working) live HTML preview immediately,
  and the next CI bake sweep re-renders a fresh poster/clip.
- Add an e2e repo-consistency guard: any bundled example.html embedding a
  sibling .html through an iframe must be a pure shell (fails on the old
  wrapper, passes now), so this class of broken preview can't return.

…recvqg3CzK8GcT)

In the project chat composer footer the model provider mark sits between
the session-mode pill and the run/send capsule. Its box gaps were equal
(4px row gap + 4px trigger padding each side), but the visible spacing
was not: the pill's × glyph keeps ~17px of ink distance from the mark
(clear-button slack + pill trailing padding), while the solid send
capsule's edge sat only ~9px away — so the icon read as glued to the
run button (the two red boxes in the Feishu screenshot).

The trigger never paints its hitbox in this context (hover stays
transparent), so ink distance is the only spacing the eye can see. Add
margin-inline-end: 8px to the composer-row trigger so the right ink gap
matches the left one (17px on both sides). Scoped to .app .composer-row
and .chat-composer-fixed-layer .composer-row; the drawer tabs-bar and
home-hero switcher chips are untouched.

Verified by rendering the exact composer markup against the real
stylesheets (tokens/shell/chat/routines/plus-menu) in headless Chromium:
before = 17px/9px, after = 17px/17px, matching the pixel measurements
taken from the reporter's screenshot.

…n never-materialized team projects (recvqmnuxxKHaI)

A member daemon's team catalog routinely lists projects this member has
never opened or pulled. Those projects have no local `projects` row, and
`workspace_projects.project_id` is a FOREIGN KEY into `projects(id)`, so
the reconciler's bind fallback (`ensureWorkspaceProject` INSERT) threw
SQLITE_CONSTRAINT_FOREIGNKEY for every such project on every pass — hub
push and ~15s poller alike — accumulating 4700+ error lines across
restarts on the live member instance.

Reconciliation corrects the bindings of projects this daemon knows; it
must never materialize one. Materialization is owned by the open/pull
path (ensureSharedProjectPlaceholder / registerPulledProject), and the
team list already displays never-materialized projects straight from the
remote catalog. The reconciler now consults a new required
`hasLocalProject` dep and skips remote entries with neither a local
binding nor a local `projects` row — the same rebind-only stance its
request-scoped sibling `reconcileLocalRowWithRemoteTeamAccess` has
always taken. The demote direction is unaffected: demotes only ever come
from existing local bindings, which the FK guarantees are materialized.

… modal

A personal-workspace user with a $0 balance who clicked 「升级套餐」 on the
AMR balance gate landed on the TEAM dashboard's billing=checkout deep link,
which opens the Upgrade-Personal-workspace-to-Team dialog in an error state
("Team plan unavailable" / 3-seat minimum) — recvpYEiH019cD, reproduced live
with a real personal account during the failed acceptance round.

Root cause: every upgrade entry picked its branch on "does the context carry
a workspaceSettingsUrl", falling back to the personal wallet deep link only
when it was absent. B returns a settings URL for personal workspaces too
(mapVelaWorkspaceContext resolves it for every workspace type, and B's
console genuinely has a settings page for them), so the fallback never fired
and every workspace — personal included — took the team branch.

Fix: the axis is the workspace TYPE, not URL presence. A new shared
`workspaceUpgradeUrl` helper (EntryNavRail.tsx, next to `teamConsoleUrl`)
makes the one decision every upgrade affordance shares:

  - personal → `wallet?view=plans` (B's personal pricing modal; verified
    live to auto-open for the same third-account session, and vela's own
    wallet route gates the modal purely on `view=plans` for a personal
    workspace context)
  - team, never subscribed → `dashboard?billing=checkout` (unchanged from
    cb48e06)
  - team, already subscribed → `dashboard?billing=plan` (unchanged from
    cb48e06)

All five upgrade entries route through the helper instead of scattering the
condition: EntryNavRail's credits chip and seat-gate invite dialog,
AmrBalanceDialog's balance-gate CTA, RecentProjectsStrip's invite dialog,
and SettingsDialog's two AMR-card upgrade buttons (via its shared
amrUpgradeUrl closure). Dialog CTAs keep a profile-keyed fallback so they
still link somewhere when no console URL is known (signed out / context not
landed); list-surface entries hide the affordance instead.

teamConsoleUrl grows a 'plans' section (wallet + view=plans, keeping the
?workspaceId pin). For a TEAM workspace B redirects that exact URL into
dashboard?billing=checkout itself, so even a misrouted team session degrades
to the first-checkout dialog rather than a dead page.

Tests: the AmrBalanceDialog component suite grows the personal-workspace
case (red on the previous code: it produced /console/dashboard with
billing=checkout) alongside the existing never-subscribed / already-
subscribed team cases; team-console-url.test.ts covers workspaceUpgradeUrl's
three states plus the null/fallback contract; the EntryNavRail seat-gate
test's personal fixture now asserts the wallet pricing modal instead of the
team checkout link it used to encode.

Verified against the real feature-test runtime with the third acceptance
account (3101794733@qq.com, personal workspace, $0 balance): the balance
gate's 「升级套餐」 now opens
wallet?workspaceId=...&view=plans (no billing param), the URL family
confirmed live to auto-open B's personal pricing dialog.

…content-changed (recvqmKQRiIlYf)

When a teammate publishes a new version of a shared project, the hub
already pushes 'project-content-changed' over the collab SSE channel --
but the daemon only forwarded a thin nudge to an OPEN project view, so
freshness still hinged on the member web's ~5s /collab/status polling
and a closed tab never synced at all.

The daemon now owns discovery: on 'project-content-changed' it runs the
SAME pull flow POST /api/projects/:id/collab/pull runs (revocation gate,
owner-routed hub pull, register-on-pull, then the existing file-changed
and project-metadata-changed signals -- zero web changes needed), via a
handle registerCollabSyncRoutes returns. Guards, all fail-closed:

- only projects locally bound as team projects in workspace_projects
  are pulled; a share the member never materialized costs nothing;
- NEVER pulls a project this daemon's member owns -- the owner's local
  copy is the single writer (same rule as useProjectCollab's web-side
  auto-pull gate), and the owner daemon receives its own publish echo;
- event workspace, local binding workspace, and active identity must
  agree, and only an active team membership pulls;
- a per-project version cursor dedupes repeated/out-of-order events,
  and racing pulls (proactive vs the member web's poll-triggered POST)
  coalesce onto one in-flight materialization in collab-sync;
- failures degrade silently: the web polling fallback stays untouched.

Red-first specs: the three collab-sync handle tests fail on the branch
baseline (registerCollabSyncRoutes returned void); the guard matrix for
the new consumer lives in tests/collab/proactive-content-pull.test.ts.
Full daemon suite: 29 failures on this branch == the same 29 on the
clean baseline (langfuse-bridge/chat-route/etc., pre-existing).

Harden Workspace collaboration authority, lifecycle cleanup, and browser coverage.\n\nRetry an expired authorized Team-project pull exactly once with a fresh receipt while preserving fail-closed scope, version, cleanup, and final freshness checks. Align the fake Vela receipt lifetime with the real two-second contract and use scalable E2E timeouts.

Record the current main ancestry while preserving the validated Workspace feature behavior and collaboration hardening.

Prevent stale Team-plugin listings, materializations, and delayed positive hub reads from reactivating a resource after retraction. Preserve independent Personal plugins with the same manifest ID and keep Team members able to share their own resources.

Add deterministic regression coverage for cache invalidation, binding lifecycle, and the superseded-positive activation race.

Reject invalid wallet balances without overwriting valid cache state, validate normalized proxy paths and Workspace scope, strip hop-by-hop headers, and tear down upstream requests on disconnect.

Keep team Workspace scope fail-closed when Connection nominates the protected header.

Preserve a verified explicit Workspace context when legacy directory rows omit the display-only workspaceName field. Keep exact-row name backfill when a non-empty name is present, without inferring authority from active or default Workspace state.

…sign-files-image-skeleton-20260803