vibe64 0.1.39 → 0.1.40

package/docs/colleague-usage/ai-onboarding.md

@@ -3,7 +3,7 @@
 Vibe64 works best with a frontier model. AI onboarding is a short full-screen
 flow that explains why and helps you connect one: Claude (recommended), Codex,
 DeepSeek or GLM, or the free option, GLM-4.7 Flash with a free Z.AI account.
-It uses the same sign-in, key and Model routing steps as **AI Accounts**; it
+It uses the same sign-in, key and model steps as **AI Accounts**; it
 never disconnects anything. Vibe64 includes no AI of its own: without a
 connected AI, Colleague, sessions and background helpers (prompt hints, titles)
 do not run, and Colleague shows **Set up your AI**.
@@ -100,10 +100,29 @@ normal AI Accounts setup directly:
   before saving it. This key is a regular Z.AI API key, not a Coding Plan key.
 
 Each choice runs through a program installed on this computer: Claude through
-**Claude Code**, Codex, DeepSeek and GLM through **Codex**, and the free GLM-4.7
-Flash through **opencode**. A card whose program is missing says **Needs
-[program] on this computer** under its buttons before you choose it. If that
-program is missing, Step 2 shows a warning instead of the sign-in or key form:
+**Claude Code**, Codex and GLM through **Codex**, DeepSeek through **Claude
+Code**, and the free GLM-4.7
+Flash through **opencode**. When any of these programs is missing, Vibe64 checks
+each one by running it, and the cards show the result before you choose:
+
+- A card whose program runs here says **Ready on this computer** under its buttons.
+- A card whose program is missing is greyed. Its main button is **Install
+  [program] first**, which opens the install row inside the card: one sentence,
+  the install command with a copy button, **Check again** and **Other ways**.
+  **Get [provider]** (the subscription page) and **Already subscribed? Connect**
+  stay below it as links. The order is install, then subscribe, then sign in.
+  Once **Check again** finds the program, the card returns to full colour. The
+  free GLM option has its own **Install opencode first** button.
+- When the recommended Claude card still needs installing but another card's
+  program runs here, a line above the cards says **Ready on this computer:
+  [program]** with **Use [provider]**, which opens that card's connect step.
+- When no AI program is on this computer at all, every card is greyed with
+  **Install [program] first**, and the heading adds **No AI tool is on this
+  computer yet. Install the one your AI needs first.**
+
+When every program is installed (or the computer's programs are provided for
+you), the cards look as usual with no install rows. If you connect anyway while
+the program is missing, Step 2 shows a warning instead of the sign-in or key form:
 
 - **[Claude Code, Codex or opencode] isn't installed on this computer**, with its
   official one-line install command, a **Copy** button and **Other ways to install**.
@@ -115,11 +134,27 @@ program is missing, Step 2 shows a warning instead of the sign-in or key form:
   the usual sign-in button.
 
 Both warnings also offer **Choose another AI**. See
-[AI accounts](ai-accounts-and-routing.md#install-claude-code-codex-or-opencode-first)
+[AI accounts](ai-accounts.md#install-claude-code-codex-or-opencode-first)
 for the install commands and where Vibe64 looks for these programs.
 
-DeepSeek and the GLM Coding Plan connect through Codex; the free GLM-4.7 Flash
-option connects through OpenCode. Use **Back** (above the header on a
+### While it works
+
+Every wait in Step 2 shows a spinner on the button you pressed and a line
+under it saying what is happening, for example **Starting Claude's sign-in.
+This can take up to a minute.** (until **Continue to Claude** is ready),
+**Checking your code with Claude.** after **Connect Claude**, **Starting the
+sign-in and getting your one-time code.** for Codex, or **Checking the key with
+[provider].** for a key. A wait that runs past about 20 seconds adds **Still
+working. This is taking longer than usual.**; a sign-in also offers **Start
+again** there. After a sign-in is accepted, Step 2 shows **Signed in to
+[provider]** with **Setting up [provider] as your main AI…** until the success
+screen opens (on a redo, **Checking [provider]…**). If that check fails, its
+warning has **Try again**. Waiting for you (open ChatGPT and enter the code)
+shows plain text, not a spinner.
+
+DeepSeek connects inside Claude Code (other tools are under **Advanced**), the
+GLM Coding Plan through Codex, and the free GLM-4.7 Flash option through
+OpenCode. Use **Back** (above the header on a
 desktop, the arrow at the top left on a phone) to return to the provider list.
 Keys are stored privately outside your projects and never shown again.
 
@@ -136,26 +171,35 @@ conversation. If Colleague cannot reach the provider within about 15 seconds
 the card says **Colleague couldn't reach [provider]** with the reason and
 **Retry**, **Check connection** (back to Connect) or **Continue anyway**; the
 check appears only after a reply or **Continue anyway**. Below it **Main AI:
-[provider] · [model]** and **Change it any time in Account settings → AI
+[provider] · [model]** (with the tool when it is not the provider's own, for
+example **DeepSeek, via Claude Code**) and **Change it any time in Account settings → AI
 Accounts**. Otherwise the screen shows
 **Colleague: still using [model]** with **Use [provider] for Colleague**.
 Choose **Start building** (the bottom bar on a phone) to close; Colleague opens
 with its message box ready so you can ask it to build your first page.
 
+The first time you set up your AI (not on a redo, and not in Learn mode) the
+success screen also offers **Take a 5-minute tour** beside **Start building**.
+It starts the installed course's orientation lesson, **Welcome to Vibe64**, in
+Learn mode; **Start building** is the "not now". It is offered only this once;
+**Take the tour** in the account menu starts the same lesson at any time.
+
 In Learn mode Colleague's conversation is your teacher's, so setup sends no
 hello there: the card says **Your teacher uses [provider] from your next
 message**, the check appears at once, and the close button reads **Back to the
 lesson** (or **Back to Learn** with no lesson in progress).
 
 If the provider connected but no model is ready yet, Connect shows **[Provider]
-is connected, but no model is ready for Colleague yet.** with **Review model
-routing**; choose an available Senior there and save.
+is connected, but no model is ready for Colleague yet.** with **Review
+models**; choose an available Senior there and save.
 
 ## Redo with an existing setup
 
-**Your AI setup** starts with a strip of the connected providers' marks. When
-Colleague uses one of them it reads **Currently using · Main AI: [provider]**
-with the only **Keep using [provider]** button; otherwise it reads **Connected
+**Your AI setup** starts with a strip of the connected providers' marks. Main
+AI is the provider behind your Senior model. When it is one of them the strip
+reads **Currently using · Main AI: [provider]**, adding the tool when it is not
+the provider's own (**Main AI: DeepSeek, via Claude Code**), with the only
+**Keep using [provider]** button; otherwise it reads **Connected
 · [providers]** with **Choose a main AI below. We recommend Claude.** Cards you
 use come first. Connected cards have a green outline with a **Connected** badge
 (**Connected · in use** for the main AI), no plan or sign-up offer, and **Manage**; on a phone they
@@ -165,11 +209,41 @@ connect step. Claude's card adds **Upgrade your experience** when you use
 another AI. Choose **Use [provider] as my main AI** (on the other cards **Use as
 main AI**) to switch, or **Get** another provider to add it. When switching
 would replace the model Vibe64 builds with, **Build with [provider]?** offers
-two choices: **Use [provider] for building** applies the suggested routing and
-moves Colleague; **Keep [current] for building** leaves routing as it is.
+two choices: **Use [provider] for building** applies the suggested models in the
+tool the provider is connected through and moves Colleague; **Keep [current]
+for building** leaves the models as they are.
 **Show details** lists each change (Building, Quick tasks) with current and
 suggested models. When nothing is set for building yet, the suggestions apply
-without asking. Nothing is removed from the card itself.
+without asking. Nothing is removed from the card itself. The crown and
+**Recommended** ribbon always stay on Claude's card; they never mean "in use".
+
+Below the cards, **Models** sums up **Senior**, **Junior** and, with a team,
+**Team members**; **Review models** opens Models (see
+[AI model setups](ai-model-setups.md)).
+
+### Team setup card
+
+In a workspace with team members, once Claude is your connected main AI and the
+team setup has no model yet, a card appears above the provider cards: **Let your
+team use this workspace**. Your Claude plan is for you only. Add a DeepSeek key
+(runs inside Claude Code) and team members get Colleague, Fix it with AI and
+helpers. You still approve what they send to main chat.
+
+- **Add a DeepSeek key** opens DeepSeek's connect step, inside Claude Code
+  (titled **For team members · runs inside Claude Code**). When the key is
+  connected, the team setup fills with recommended DeepSeek models.
+- When DeepSeek already works in Claude Code the button reads **Use DeepSeek for
+  team members** and fills the team setup at once.
+- **Not now** hides the card for 7 days in this browser. **Set up team AI** (on
+  the project list) shows it again.
+- Afterwards it reads **Team AI is ready**, for example **Team members use
+  DeepSeek V4 Pro for Colleague and Fix it with AI, and DeepSeek V4.1 Flash for
+  helpers.**, with **Review team models**, which opens Models on **Team
+  members**.
+
+If filling fails the card shows the reason; choose the button again. Opening
+**Set up team AI** when there is no card to offer opens Models on **Team
+members** directly.
 
 ### Manage a provider
 
@@ -182,7 +256,6 @@ characters, with **Default** on the one used by default. Its buttons:
 
 - **Manage sign-in** (Claude or Codex login) or **Replace key** (a provider
   key); **Reconnect** when it stopped working.
-- **Model routing** to choose the models for that tool.
 - **Remove**, which asks for confirmation first.
 
 GLM's Manage lists the free Z.AI connection under **Free option**, with the
@@ -195,7 +268,7 @@ with its close button or Escape; focus returns to **Manage**.
 **Set up later** (next to **Choose my AI** on the first screen, then at the top
 right; on a phone in the bottom bar of the first screen, then below the cards
 and below the connect steps) closes the flow and keeps it from opening by itself in this browser for 24 hours. The
-close button or Escape just closes it. While a sign-in, key check or routing
+close button or Escape just closes it. While a sign-in, key check or model
 save runs, closing and **Back** wait for it to finish.
 
 - **We couldn't check your current AI connections.** Use **Try again**;
@@ -204,7 +277,7 @@ save runs, closing and **Back** wait for it to finish.
   Z.AI overload or insufficient balance); retry the same step.
 - **Manage** stays on this screen: closing it (its close button or Escape)
   returns to the cards with focus on the **Manage** button you used. While a
-  removal, key change or routing save runs, Manage cannot be closed.
+  removal, key change or model save runs, Manage cannot be closed.
 - Someone who may not connect AI sees **Only the Vibe64 owner can connect AI
   accounts.** with **Close**.
 
@@ -226,4 +299,4 @@ motion.
 
 Colleague can explain these screens, compare the providers and offer to open
 the flow. Choosing a plan, paying, signing in, entering a key and approving
-routing remain your steps; never paste a key into chat.
+model changes remain your steps; never paste a key into chat.

package/docs/colleague-usage/ai-personal-use.md

@@ -0,0 +1,41 @@
+# Personal-use AI subscriptions
+
+Some AI connections are personal plans: a Claude subscription (Pro or Max)
+signed in to Claude Code, a ChatGPT subscription signed in to Codex, and the GLM
+Coding Plan. Each provider's terms (Anthropic's, OpenAI's and Z.AI's) allow only
+the person who subscribed to use the plan. Vibe64 therefore lets only the
+workspace owner work with a personal connection. Other people in the workspace
+cannot use it, even in the owner's projects. API-key connections (for example
+DeepSeek, an OpenAI API key or the free Z.AI key) are **Workspace use** instead.
+
+## Where personal use is marked
+
+Personal connections carry an amber **Personal use** chip with a lock icon:
+
+- **Models**: each personal model choice in a role's list shows the chip. While
+  any of the owner's roles uses a personal model, a highlighted note above the
+  roles says **Personal use: only the workspace owner can use this [plan]**. The
+  **Team members** setup lists no personal models (see
+  [AI model setups](ai-model-setups.md)).
+- **AI Accounts**: a provider key editor whose route is personal shows the chip;
+  the GLM Coding Plan form ends with the same note.
+- A host may add the marking on AI setup cards, connection tiles and the success
+  screen after connecting.
+
+## What does this mean?
+
+Select **What does this mean?** beside the chip or note to open **Personal use:
+only the owner can use it**. It names the plan and provider, says that only the
+workspace owner can work with it, and that other people need a Workspace-use
+connection such as an API key billed by use. In Models, the **Team members**
+setup chooses the Workspace-use models they use. With Claude as your main AI,
+the team card **Let your team use this workspace** sets it up with one DeepSeek
+key that runs inside Claude Code (see
+[AI onboarding](ai-onboarding.md#team-setup-card)). Select **Got it** to close it. The explanation is
+the same on desktop and mobile.
+
+## What Colleague can do
+
+Colleague can explain which connections are personal and what members need, and
+can open AI Accounts. Subscribing, signing in and entering keys remain the
+owner's steps in the browser.

package/docs/colleague-usage/auto-workflow.md

@@ -84,9 +84,14 @@ continuation through its existing actions. An explanation or offer alone does no
 authorise another turn. Colleague's watches treat Waiting as needing attention.
 
 Auto always uses a separate Senior review after implementation, even when Senior
-implemented it or both roles use the same model. The **Deslop** switch adds
-behavior-preserving cleanup to Senior review; turning it off does not disable
-review. Unfinished checks continue in Senior review; substantial necessary rework
+implemented it or both roles use the same model. The **Deslop during Senior
+review** switch in the chat-mode menu adds behavior-preserving cleanup to Senior
+review. It is on for new chats. Turning it off opens **Turn off Deslop?**: without
+Deslop, code quality suffers and the project can become hard to change and may
+crash and burn over time, unless you ask the AI to “Deslop” in the chat yourself.
+Skip it only for quick throwaway projects. **Cancel** keeps Deslop on; **Turn off
+Deslop** saves it off for this chat. Turning it off does not disable review.
+Chats created before Deslop became the default keep their saved setting. Unfinished checks continue in Senior review; substantial necessary rework
 returns to Junior and comes back for another review. A changed product requirement
 needs your decision. Planning-only work waits for your request to implement it.
 

package/docs/colleague-usage/codex-context-recovery.md

@@ -51,7 +51,8 @@ message is needed to enable its tools or context window. Each provider still
 requires its own connection in **AI Accounts**.
 
 When supported, switching from a compacted GPT conversation to DeepSeek or GLM
-recovers readable earlier messages automatically. You do not need to paste a
+recovers readable earlier messages automatically. DeepSeek V4.1 Flash, DeepSeek
+V4 Pro and GLM 5.3 support this, so each can be a Codex Senior or Junior. You do not need to paste a
 second copy. Saved native history stays intact; unsupported recovery produces
 the error described below instead of silently omitting earlier work.
 

package/docs/colleague-usage/coding-agent-goals.md

@@ -29,14 +29,17 @@ during a pending delivery may require cancelling that request and sending it aga
 These steps apply on desktop and mobile. Account sign-in and credential entry
 remain your actions; Colleague can explain the recovery and offer supported help.
 
-An unfinished goal stays with its current orchestrator. Open **Chat mode** while
+An unfinished goal stays with its current tool. Open **Chat mode** while
 it is running or paused and choose **Senior**, **Junior**, or **Custom**. A model
-or direct-mode change within the same orchestrator applies to the next turn,
+or direct-mode change within the same tool applies to the next turn,
 including automatic goal continuations. The current turn keeps its settings;
 the goal keeps its objective, budget and progress. You do not need **Pause goal**,
 **Cancel goal** or a reload to save settings.
-Choosing another **Orchestrator** or **Auto** saves the preference for after the
-goal finishes; it does not transfer or cancel the current goal. Ordinary messages still use the current goal’s orchestrator while that preference is deferred. If a message is rejected before sending, its delivery check also uses the current goal’s orchestrator. These steps apply
+Choosing **Auto**, a model in another tool, or moving the main AI to another tool
+in AI setup applies after the goal finishes; it does not transfer or cancel the
+current goal. Ordinary messages still use the current goal’s tool and the AI setup
+saved when it started. If a message is rejected before sending, its delivery check
+also uses the current goal’s tool. These steps apply
 on desktop and mobile, including goals created by the assistant. Colleague can
 explain or offer a supported change and needs your request before performing it.
 

package/docs/colleague-usage/coding-agents.md

@@ -153,17 +153,18 @@ added. Use an open session before uploading again. Colleague can explain the
 notice; it cannot override the closing check. The same rule applies on desktop
 and mobile.
 
-Open **Chat mode** at any point to choose **Senior**, **Junior**, an
-**Orchestrator**, or **Custom** → **Apply**. Saving settings does not stop the
+Open **Chat mode** at any point to choose **Auto**, **Senior**, **Junior**, or
+**Custom** → **Apply**. Senior and Junior come from the AI setup and always run
+in the main AI's tool. Saving settings does not stop the
 current turn. The next turn uses the new settings; see [Goals and model
 changes](coding-agent-goals.md) for automatic continuations and an unfinished
-goal's retained orchestrator. If saving fails, the control reports the error;
+goal's retained tool. If saving fails, the control reports the error;
 retry after resolving it. Colleague can explain or perform a supported selection
 change when you explicitly ask.
 
 To change the model for an active Update merge repair, open **Chat mode** in
 that repair's composer, then **Choose model**. In **Custom AI**, choose a
-**Model** and select **Apply**. The repair keeps **Senior**, its history and its
+**Model** and select **Apply**. The repair keeps **Fix it with AI**, its history and its
 current turn. The next turn uses the selected model; Main chat's model is
 unchanged. On compact screens, reveal chat first; the controls are the same as
 on desktop.
@@ -203,7 +204,9 @@ Prompt suggestions wait for five seconds without draft changes, and start no
 more than once every thirty seconds in the current view. Refocusing does not
 regenerate dismissed suggestions. Changing the draft can request new suggestions
 after those limits; unchanged input does not poll. You can keep writing and send
-without waiting for suggestions.
+without waiting for suggestions. Suggestions read the project's Blueprint and its
+subsystems map, so after a starter is applied they propose changes to the app it
+built instead of building it again.
 
 An unconfirmed message shows small red error text and **Retry** inside
 its own bubble. The notice belongs to

package/docs/colleague-usage/colleague-watches.md

@@ -14,6 +14,14 @@ one-shot watch ends after its first delivered report. Ask “Tell me when the wo
 finishes” for a completion watch instead. Watching starts only on your request,
 checks your current project access and does not stop or resend the coding work.
 
+Colleague can also watch more than conversations, wherever your Vibe64 offers
+it, for example a project's deployment on a hosted Vibe64: ask “Tell me when my
+app is live” or “Tell me if anything changes in that deployment”. Ask Colleague
+what it can watch here; it lists what this installation offers. An app that is
+already live is reported at once. Every reading checks your current access to
+that project; a watch only reports and never deploys, restarts or changes
+anything.
+
 Use the eye-and-count **Watches and assignments** button beside Previous
 conversations to inspect ongoing work. It opens a separate view; nothing is
 listed below chat. **Resume** restarts a paused watch, and **Cancel** removes its
@@ -21,10 +29,13 @@ watch without stopping the agent. Close this view to return to the same chat and
 draft. On phone, use **Close watches and assignments** in its header. Ask
 Colleague to list, cancel or resume watches if you prefer.
 
-Colleague does not watch lessons. In the Learning environment your own step
-actions (an answer, **Done**, or an automatic check from your App press or a
-code step's **Check now**) record step results directly; see
-[The Learning environment (L)](learning-environment.md).
+Colleague never watches lessons or decides steps. In the Learning environment
+your own step actions (an answer, **Done**, or an automatic check from your App
+press or a code step's **Check now**) record step results directly; see
+[The Learning environment (L)](learning-environment.md). In Learn mode your
+teacher can watch your lesson's practice project for you (its agent, or its
+deployment where offered) and nothing else; that watch ends with the lesson
+attempt. Watches in Develop stay in Develop's conversation.
 
 You can ask “Which projects do I have?” from any page, including AI Accounts;
 no open project is required. Colleague checks the projects available to you.

package/docs/colleague-usage/colleague.md

@@ -70,12 +70,17 @@ If no usable model is connected yet, Colleague shows a warning as soon as you
 open it (or after your first request) that nothing is sent and your draft is
 kept. Choose **Set up your AI** to open the guided AI setup, which recommends a
 provider, opens its sign-up page and connects it, then switches Colleague to that
-AI's Senior model. Or choose **Open AI Accounts**: hosted Vibe64
-opens Management → AI Accounts; standalone Vibe64 opens its existing Account
-settings on **AI Accounts**. Connect an AI and configure an available Senior
+AI's Senior model. Or choose **Open AI Accounts** to open the AI Accounts
+settings. Connect an AI and configure an available Senior
 model, then return and press **Send**. Opening setup does not send the draft or
 promise a response from an unavailable model. Credential entry and provider
 consent remain your actions.
+Colleague starts on the Colleague model of your setup in **Models** (see
+[AI model setups](ai-model-setups.md)): the owner's follows Senior unless the
+owner chose another; team members start on the team setup's Colleague model.
+Each person can pick their own model from Colleague's model menu, which lists
+only models they may use (team members never see personal plans). Your pick is
+yours alone and does not change Models or anyone else's Colleague.
 Model changes keep saved messages and replies, including between native and API.
 A changed engine, provider, model, agent or effort prepares the next conversation;
 the same selection retains it. Applying starts no model work.
@@ -179,12 +184,23 @@ If no progress sentence is available when the first action starts, Colleague say
 including when you reopen the conversation, until the answer replaces it.
 Background monitoring does not speak an interactive lookup acknowledgement.
 Its completed notification can still be spoken when sound is on. A saved notice,
-“An update from your watched conversations.”, appears before the completed answer,
+“An update from what you asked Colleague to watch.”, appears before the completed answer,
 including after reopening. Internal watch instructions and tool arguments stay hidden.
 
-For response watches, Learning observation and recorded action outcomes, read
+For response watches, watching more than conversations (such as a
+deployment), Learning observation and recorded action outcomes, read
 [Colleague watches and operations](colleague-watches.md).
 
+## Colleague outside projects
+
+Colleague works on every screen, not only inside a project: the account menu,
+AI setup, Courses and any other screen outside your projects. Ask how to do
+something there and it reads this release's guide first. Ask “Show me where…”
+and it points at the place, or opens it when you ask it to; it only offers
+places this Vibe64 has and that you may use. It can list your projects, and
+watch what this Vibe64 lets it watch for you. A studio lesson's step bar points
+at the same places.
+
 Recognized words can appear as a Pending user message in chat; acceptance replaces
 it with the saved message once. Unsent or discarded words are not saved history.
 Review and recording controls live beside the same typed composer. Microphone permission

package/docs/colleague-usage/connected-contributions.md

@@ -31,13 +31,13 @@ but the receiving repository's branch permissions still apply.
 
 If the contribution was discarded, further work needs a new contribution. If both
 active and discarded branches exist, resolve the interrupted discard first. A
-missing branch after a publication attempt needs its retained exact revision and
+missing branch after a push attempt needs its retained exact revision and
 receiving history checked; its absence alone proves nothing. An integrated
 revision remains accepted after its branch is discarded or deleted. Explicit
 Update can consume the proven accepted source, but further contribution pushes
 cannot recreate that removed branch. Keep the checkout and ask the coding agent
 to inspect uncertain saved work. After a lost acknowledgement, existing Save recovery can reconcile a
-publication when the remote confirms its exact prepared commit.
+push when the remote confirms its exact prepared commit.
 
 To revise a discarded, unmerged contribution, return to the originating session
 and press Save. Its refusal offers **Review replacement contribution**. Review
@@ -77,7 +77,7 @@ It runs the same preparation.
 The update uses the receiving session's normal **Update this session (rebase)**,
 which carries that checkout's own changes, including unsaved edits, onto the
 branch's latest commit. A checkout is left as it is while its agent is working,
-while a Save or Update runs in it, while its published contribution still awaits
+while a Save or Update runs in it, while its pushed contribution still awaits
 acceptance, or when the update would conflict with its changes. The preparation
 output shows one line per repository: `[Connected] jskit updated to abc1234.`,
 `[Connected] genesis is up to date.`, or `[Connected] vibe64 was not updated:`
@@ -95,7 +95,7 @@ then Save.` This keeps saved and deployed work tested against the current
 connected sources.
 
 When a session has recorded submitted contributions, its normal Save checks those
-exact revisions before publishing, including when the main repository has no new
+exact revisions before pushing, including when the main repository has no new
 file changes. If one is still pending, the error names the receiving repository,
 contribution, revision and integration branch. Once all are accepted and the
 selected sources are reconciled, a clean main repository reports no work to save. Open that
@@ -145,7 +145,7 @@ an instruction to merge or discard.
 For prepared sources, the chat Save button changes to a branch icon labelled **Save connected
 first** when connected work needs submission. It remains available when the main
 checkout has no changes. Its confirmation lists receiving repositories and
-contribution branches. Choose **Push connected contributions** to publish those
+contribution branches. Choose **Push connected contributions** to push those
 changes; the main project's changes stay local. The same controls work on phones.
 The result lists confirmed pushes, including those completed before a later
 failure. Review them in each receiving project's **Incoming changes**.
@@ -164,21 +164,21 @@ pass that list unchanged with the main `destinationReview` to Save.
 This saves one connected repository at a time: commit, push, record its commit,
 then continue. It never pushes the main project in that action. A failure reports
 the completed repositories and the repository that failed. Inspect and retry:
-completed repositories without new changes are skipped. If publication succeeded
+completed repositories without new changes are skipped. If the push succeeded
 but recording failed, retry records the confirmed receiving commit without
-republishing it. These commit records identify required integration, not a
+pushing it again. These commit records identify required integration, not a
 separate retry ledger. Files and existing receiving Save recovery remain available.
 Each receiver is checked again when its turn arrives, so new edits made while
 an earlier repository was saving are included in the next receiver's Save.
 If a push loses its acknowledgement, the exact remote commit can confirm it.
-If the remote is unavailable too, publication stays unknown: restore repository
+If the remote is unavailable too, the push outcome stays unknown: restore repository
 access and inspect work again. Recovery checks the original prepared commit
 without another push, including after restarting the editor. Main Save remains
 blocked while that outcome is unknown.
 The same recovery checks a failed local update after a push. When Git proves
-the attempt did not publish, its original rejection remains visible. A receiving
+the attempt did not push, its original rejection remains visible. A receiving
 project with its own connected work must satisfy its own Save checks; this action
-does not recursively publish further repositories.
+does not recursively save further repositories.
 If preparation fails before any push is attempted, fix the reported cause and
 retry; the contribution's files and recovery checkpoint remain. A missing branch
 after an uncertain push is still unknown, so Vibe64 does not recreate it merely
@@ -186,11 +186,11 @@ because it is absent.
 
 After connected work is saved, omit `connectedReview` for ordinary Main Save.
 Pending integration is reported by the existing exact-revision gate. New connected
-work or unreconciled publication blocks Main Save; inspect and save/recover the
+work or an unreconciled push blocks Main Save; inspect and save/recover the
 receiving session first. Colleague must review these destinations and have an
-explicit Save instruction before publishing.
+explicit Save instruction before saving.
 Main Save checks those sources again after preparing its commit, immediately
 before pushing. If a receiving worktree, HEAD, or recorded saved commit changed,
 inspect connected work and save it before retrying. This is a final snapshot
 check, not a lock across projects; avoid editing connected sources while Main
-Save is publishing, because changes after the final read belong to later work.
+Save is pushing, because changes after the final read belong to later work.

package/docs/colleague-usage/connected-projects.md

@@ -8,10 +8,10 @@ as a merged contribution.
 Checking connected Save status can take longer on a busy development workspace
 because it checks each receiving repository. Allow up to two minutes for the
 check. If status remains unavailable, use Save to check again and read its
-reported blocker; an incomplete check never publishes the main project.
+reported blocker; an incomplete check never saves the main project.
 
 The session badge shows **Unsaved changes** when a connected repository needs
-publication, even if the main repository is clean. After connected Save succeeds,
+saving, even if the main repository is clean. After connected Save succeeds,
 the result lists the pushed commits. An empty progress card closes when the action
 finishes; opened details with real output remain available until dismissed.
 

package/docs/colleague-usage/conversation-context.md

@@ -37,7 +37,7 @@ the operator's access.
 Managed Claude conversations use Vibe64's Git workflow without Claude's automatic
 startup Git snapshot or built-in commit/PR instructions. Agents can still inspect
 current changes through the ordinary managed Git tools when the task needs them;
-Save and publication keep their existing controls and permissions.
+Save and push keep their existing controls and permissions.
 
 Project guidance can change while Codex is working. The current turn keeps its
 installed instructions; the next idle admission reads the latest guidance without

package/docs/colleague-usage/dashboard-navigation.md

@@ -2,8 +2,9 @@
 
 Open a project and select **Dashboard** beside **Preview**. The Dashboard provides
 the available session and project tools, including Files, Changes, Database,
-Architecture, AI Terminal, Project settings, Git log, Env, App access, Integrations, Issues
-and Pull requests. Availability depends on the selected session, host and permissions.
+Architecture, AI Terminal, Project settings, Git log, Env, App access, Integrations, Publish, Issues
+and Pull requests. **Publish** turns the project into an app or package people install; see
+[Publish your project](publishing.md). Availability depends on the selected session, host and permissions.
 On desktop the section list is grouped: tools under the **Project** label come first,
 then a **Session** label with the selected session's name and its tools. The
 highlighted pill marks the open section, and each page's header repeats that
@@ -11,7 +12,8 @@ section's icon and title with a one-line description and its actions.
 
 On mobile, **Show project** reveals the project from chat. Use **Dashboard section**
 to choose a tool and **Back to dashboard** to return from its detail. **Show chat**
-returns to the coding conversation. Preview and Dashboard retain their own selected
+returns to the coding conversation. On a desktop the header's **Preview** and
+**Dash** (Dashboard) switch between the two panes. Preview and Dashboard retain their own selected
 state when hidden; leaving the preview does not stop its application.
 
 The left-arrow **Back to dashboard** control has the same appearance in

package/docs/colleague-usage/development-env.md

@@ -14,13 +14,19 @@ Stack defaults are editable fallbacks: saving one creates a project override,
 and removing that override reveals the current declared default again. Defaults
 are user-owned in Env; platform-injected runtime values remain system-owned and
 non-editable, even when Stack declares the same name. Reserved Vibe64-prefixed
-names require a Stack declaration before a user can supply them.
+names require a Stack declaration before a user can supply them. When a Stack
+resource, such as a starter's database, is not yet configured, its bindings that
+have a default keep their default row; only the values without one, for example
+`DB_PASSWORD`, appear as empty rows to fill, and the setup message lists them
+as missing.
 
 Saved development values are available to the coding agent, its managed browser,
 and managed Playwright commands. For an authorized external site, tell the coding
 agent which variable names contain its URL and credentials. Values need not be
 pasted into chat or copied into source. Presence in Env confirms configuration,
-not a successful login.
+not a successful login. Because the coding agent receives every Development
+value, use test credentials there (for example Stripe test or restricted keys);
+production credentials do not belong in Development Env.
 
 When the application needs a generated secret, such as an application key, the
 coding agent generates it with the framework's own command and stores it as a
@@ -28,6 +34,15 @@ coding agent generates it with the framework's own command and stores it as a
 value into a generated `.env` file: Vibe64 rewrites that file from Env, so an
 edit there would be lost. The key then appears in **Env** with its value hidden.
 
+When the host offers saved values, `vibe64-helper env saved` lists their names,
+descriptions and fields, never their values. To use one, link it to the
+variable the application already reads (`vibe64-helper env link development
+STRIPE_SECRET_KEY <id> apiKey`) instead of asking the person to paste a key.
+A linked row is not editable; `vibe64-helper env unlink` returns it to an
+ordinary empty row. Env shows a linked row as one line with a link icon and
+the host's provenance label; on a phone each variable becomes a small card. A
+managed resource of the same name still wins.
+
 Each managed shell command reads current Development Env, so the coding agent's
 next command receives added or changed values without restarting its conversation.
 The agent can use ordinary application commands; it does not need to copy managed
@@ -49,8 +64,10 @@ resource readiness, environment names, declared Workspace setup commands and
 managed tool entry points. This read exposes no environment values and starts
 no preparation. Repeat it after changing resource declarations. A
 `not-prepared` resource requires Workspace setup; it is not a request to invent
-credentials. In the session chat, wait for the assistant to finish, then choose
-**Prepare workspace**, or **Retry** on a failed preparation. Adding a resource
+credentials. Starting Preview (`vibe64-helper preview ensure --wait`) prepares
+the workspace, so the agent rereads the status before reporting preparation. In
+the session chat, **Prepare workspace** or **Retry** on a failed preparation
+does the same once the assistant has finished. Adding a resource
 requires preparation even if the dependency-install command has not changed.
 An earlier successful preparation does not cover a changed setup recipe or newly
 unprepared resources. Preparation retry checks the current requirements before

package/docs/colleague-usage/file-interfaces.md

@@ -90,7 +90,7 @@ source, person, project/session, verified Helper profile and prompt revision.
 Source changes mark the text out of date. It remains informational and is not
 supplied to coding agents. Generating or updating starts only on an explicit
 button press. A missing Helper disables generation, but saved text and Code stay
-readable. Configure the Helper in **Model routing** and connect its account in
+readable. Choose the **Explain & check** model in **Models** and connect its account in
 **AI Accounts** if necessary.
 
 While generating, the field shows **Generating explanation…**, **Stop**, and any