v0.1.0-alpha.3 — Codex-first, real receiver-owned results

Codex first, using the account you already have

The default first-use example now runs two real Codex sessions. No Pi installation, second subscription, hand-written test project or second terminal is required.

npm install --foreground-scripts --loglevel=info \
  https://github.com/fyaic/threadmesh/releases/download/v0.1.0-alpha.3/fyaic-threadmesh-0.1.0-alpha.3.tgz
npx threadmesh try --live

Requires Node 22+ and an authenticated Codex installation. The package is distributed here, not on the npm registry. Omit --live for instructions without a model call. Existing Codex configuration and normal quota apply.

What actually passed

The website session volunteered a dependency; the brand session chose to share changed product claims. After that message, the runner continued the same website session. Its model updated the landing copy while retaining the user's earlier signup-button decision and the paid price. Unrelated work received no messages.

The exact default installed-package run passed in 272.604 seconds, inside the unchanged 300-second total cap. An earlier diagnostic passed in 184.050 seconds with an extended diagnostic cap. These are maintainer observations, not a success-rate estimate or fixed time promise.

Improvements

• Prefer a newer verified installed macOS desktop Codex runtime over an older PATH installation. No runtime installation, account/model change or silent harness fallback.
• Handle native retry notifications without prematurely failing; retain explicit quota/authentication failures and process cleanup.
• Show runtime choice and actionable old-version errors.
• English/Chinese onboarding now leads with Codex and a concrete before/after case. Pi remains an explicit option.
• Include an experimental native desktop skill using task tools already exposed by the host. This is guidance, not an external adapter or a missing-tool workaround.

Honest limits

The verified example creates two new disposable App Server threads. It does not attach old desktop chats; its runner-triggered continuation does not prove native background wake. The native desktop skill has packaging/tabletop review only, not live existing-task acceptance. Codex first use supports the structured copy case; the API example requires explicit Pi selection. Connection retries and quota limits can still prevent completion.

455 regression tests passed, one optional native test skipped; schema/transition checks and skill/plugin validators passed. The final tarball's runtime sources match the live-tested installed package byte for byte.

Acceptance evidence · First use · 中文说明

Asset SHA-256: b8150ef2b9052b34cd0fb69841992f6a90cc9bc4446bceea172189424ad5ae89

v0.1.0-alpha.2 — Try a real same-agent handoff

First useful collaboration, without writing a harness

threadmesh try preferences --live prepares a small real example, starts two Pi sessions and checks whether a model-selected handoff led to the same receiver's own correct edit. No repository checkout, custom harness, application fixture, workspace paths or second terminal.

Install and try

Requires Node 22+ and installed/authenticated Pi with an available model. Uses your Pi model configuration unless you supply --provider/--model.

npm install --foreground-scripts --loglevel=info https://github.com/fyaic/threadmesh/releases/download/v0.1.0-alpha.2/fyaic-threadmesh-0.1.0-alpha.2.tgz
npx threadmesh try preferences --live

Omit --live for instructions without a model call. Try api --live for pagination. Progress and result checks are visible; missing auth/quota, silence, incorrect work, cancellation and timeouts are not reported as success. Logs/sample files remain private locally; models use your provider as usual.

Verified

• Same-product real copy case: pass in 201.880 seconds, earlier signup label and paid-plan price preserved.
• API case from a separately installed package: pass in 158.371 seconds, same client session's own edit fetched both cursor pages.
• 435 local unit tests passed, 1 optional native skip, 55 schema + 7 transition cases. PR #163 CI passed.
• An additional fresh npm-cache install from the local packaged artifact completed in 18.486 seconds on the maintainer's Mac. This excludes release-asset download, does not reset native prebuild caches, and is not an installation-time guarantee.
• Evidence: https://github.com/fyaic/threadmesh/blob/main/docs/09-reviews/2026-09-07-packaged-first-use.md

Honest boundaries

New disposable Pi sessions, NOT existing desktop chat attachment. Normal model quota and OS permissions apply; this is not a sandbox. One pass per case, not a reliability score or independent first-user success. No simulated fallback. Desktop existing-session adoption remains open.

Addresses setup feedback in #158; does not close its remaining live/busy-receiver checks. English and Chinese first-use documentation updated.

Package SHA-256: 34de3420476b9107be25bf3c18fbc1aa1157bb3499bd9bb6830dada86f555a16.

v0.1.0-alpha.1 — Your first useful agent collaboration

What you can use

• Explicit local workspaces with peer goals, non-consuming inboxes, model-selected advice and portable checkpoints.
• A native Pi extension with opt-in idle follow-up; Codex and Kimi MCP launch configuration; official DeepSeek Harness Cordis MCP integration.
• Three clearly labelled no-model previews: API changes, shared preferences and quota handoff.
• A public continue command to start another harness from the last saved checkpoint.

Start

Requires Node 22+ and a separately installed/authenticated harness for real model work. Install the attached package or use the tagged GitHub source:

npm install github:fyaic/threadmesh#v0.1.0-alpha.1
npx threadmesh preview api

The package is distributed on GitHub, not published to npm. Follow docs/06-guides/first-workspace.md for the tested Pi pair and other integrations.

Verified, with limits

• 395 unit/subtests, 55 schema cases, 7 transition cases; clean packed consumer and GitHub CI/link checks pass.
• Real Pi 0.84.2 / zai glm-5.3 pair: model-selected contact, idle receiver follow-up, actual client update, independently tested cursor pagination, zero unrelated messages. One controlled run, about 205 seconds; not an installation-time or reliability promise.
• Official DeepSeek 0.1.2-rc.1 runtime: four MCP tools and bidirectional delivery pass without a model key. DeepSeek model initiative remains pending credentials.
• Real Pi checkpoint continuation passes from deliberately seeded explicit context. Not a lossless export of a quota-blocked chat.
• New Codex workspace initiative did not pass; Kimi 0.39.1 hit weekly quota. Failures are recorded alongside the success.

Evidence: https://github.com/fyaic/threadmesh/blob/v0.1.0-alpha.1/docs/09-reviews/2026-09-05-first-use-validation.md
Next cross-harness work: #153
First-run reports, including failures: #79

Experimental same-owner local software; no multi-tenant security, arbitrary-host wakeup, quota bypass or production reliability claim.