Bring Thomas’s shared compute design into the community Mesh page

[sandro-sq]

Shared compute design above the community mesh

Members can now see the Share compute toggle, Thomas Petersen’s original animated bee, hardware / AI memory / model cards, and shared-compute activity together above the existing community-agent and community-mesh content.

This PR targets jimmy/mesh-share-wip, the source branch of #580—not main.

• Replaces the earlier sharing presentation with the supplied design, using existing app controls and responsive design tokens.
• Preserves the existing Share On/Off, Auto / Reset to Auto, model selection, recovery, consent, identity and community boundaries.
• Keeps Thomas’s shared-compute agent labels and per-agent Start/Stop controls in the normal agent inventory.
• Unknown completed-request and serving-node counts show —, not invented zeroes. Token totals are routing-observed session completions, not a claim of local contribution.

Screenshot

Synthetic IPC fixture with sample hardware/model/usage, not live serving evidence:

Authorship and integration

feat/community-compute-plugin at b4a910e7 remains an ancestor through merge 590a4ab2. Thomas’s six original commits, identities, hashes and Signed-off-by trailers are preserved; there is no squash, cherry-pick or rewritten attribution. The older parallel compute/runner/wallet stack is not enabled.

The original four-design renderer is embedded in an allow-scripts sandbox with parent-fed allowlisted telemetry. The existing native owner supplies worker epochs and nullable SDK usage; no new runtime, dependency, permission, IPC command or persistence authority is introduced. See the integration record.

Verification

Checked at bc29ec27 (pre-commit hooks applied no source changes):

• pnpm check: Biome, TypeScript, design TypeScript and design-system checks pass.
• pnpm build: passes (existing chunk-size advisory remains).
• Focused Vitest: 157 passed, including Mesh page/model/activity, donor renderer, agent cards and both inventory render paths.
• Existing sharing browser journey: Chromium and WebKit pass, covering sandbox canvas pixels, telemetry, keyboard switching, Share On/Off, desktop/narrow geometry and light/dark modes.
• cargo fmt --all --check: passes.
• cargo test -p buzz-mesh-compute --features mesh --locked: 56 passed / 1 ignored.
• cargo test -p buzz-foundation --features mesh --lib --locked: 365 passed / 8 ignored.
• Mesh-enabled Clippy with --all-targets -- -D warnings: passes.
• Mandatory pre-push hooks: 224 test files / 3,573 tests pass, design-system checks and workspace Clippy pass; remote head verified.
• Hosted DCO Check passes at bc29ec27; broad GitHub CI is running.
• Independent source review: 9/10, no recommended blockers. Optional unknown-counter wording suggestions are deferred.

No browser cases were added or removed; the existing sharing journey retains its model/Share assertions and adds the browser-only iframe/canvas boundary. Colocated tests cover activity unknown counters and the telemetry DTO; Rust tests cover usage projection and launch epochs. Existing PR CI will provide broad validation; no workflow or test-cost policy is changed.

Remaining acceptance

Draft until hosted checks are inspected and native/live acceptance is confirmed. The human approved the screenshot preview, but this does not attest to testing the running app. Real community inference, packaged quit/relaunch, native webview acceptance, and cross-platform packaging remain deferred. No buzz-review-completed attestation is made.

[sandro-sq]

🤖 Independent agent review of the feature diff against jimmy/mesh-share-wip: 9/10; no recommended blockers. Reviewed Mesh vision/design, native worker-epoch sampling and shutdown fences, the sandboxed iframe/message boundary, existing Share On/Off consent/lease enforcement, and the regression-test seams. Thomas’s original branch remains merge ancestry with unchanged commits and sign-offs. Optional unknown-counter/status wording suggestions are deferred; unknown serving-node/request counts intentionally remain —. Source review and synthetic tests do not establish packaged/native live inference acceptance.