Skip to content

chore: version packages#1012

Open
github-actions[bot] wants to merge 1 commit into
mainfrom
changeset-release/main
Open

chore: version packages#1012
github-actions[bot] wants to merge 1 commit into
mainfrom
changeset-release/main

Conversation

@github-actions

@github-actions github-actions Bot commented Jul 21, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@gemstack/ai-autopilot@0.12.0

Minor Changes

  • decace4: Breaking (@gemstack/ai-autopilot): the framework-detection exports are renamed so they no
    longer collide with the user-facing domain presets.

    Two unrelated subsystems were distinguished only by a directory's singular-vs-plural: src/preset/
    (the Open Loop domain bundles, {loops, prompts}) and src/presets/ (framework detection). Both
    were re-exported side by side from the one entry point, so definePreset and selectPreset read
    like a pair while being from different subsystems. "Preset" now means the user-facing domain bundle
    only, which is what the shipped root presets/ markdown and the dashboard already meant by it.

    src/presets/ moves to src/framework-detection/ (internal), and the exports rename:

    Before After
    definePreset defineFrameworkPreset
    Preset FrameworkPreset
    PresetSpec FrameworkPresetSpec
    PresetSignals FrameworkPresetSignals
    PresetScore FrameworkPresetScore
    PresetRegistry FrameworkPresetRegistry
    PresetError FrameworkPresetError
    builtinPresets builtinFrameworkPresets
    builtinPresetRegistry builtinFrameworkPresetRegistry

    detectFramework, vikePreset, nextPreset, FrameworkSignals and FrameworkDetection are
    unchanged: they were already unambiguous. The domain-preset exports (defineDomainPreset,
    selectPreset, composeDomainPresets, loadDomainPreset, builtinDomainPresets,
    builtinPresetsDir, ...) are unchanged.

    No behavior change. As a side effect @gemstack/ai-autopilot no longer exports a definePreset
    that clashes with the unrelated definePreset in @gemstack/framework.

Patch Changes

  • Updated dependencies [cd121f4]
  • Updated dependencies [780ef3e]
  • Updated dependencies [23c1fb7]
    • @gemstack/ai-sdk@0.6.1

@gemstack/framework@1.2.0

Minor Changes

  • 61451a1: Dashboard: the session page says its branch once. The handoff card that repeated the branch name under the action bar is gone; its verdict ("2 commits · 0 files", "no changes", "branch gone") and its Push branch / Open PR buttons now ride in the branch row itself, and clicking that row expands the commits and changed files. A live session's Changes panel folds into the same row: the file count sits beside the branch, the rows open underneath.

  • ecf2ce4: feat(framework): context fragment lists the recorded conversations (Update Context prompt fragment #683)

    Adds .the-framework/conversations/**.md to CONTEXT_DOCS, so a run is told to read the human conversations (Discord/chat turns) that earlier runs committed there. A read-only pointer, like tickets/ and TODO_AGENTS.md, so it stays out of the merge-update set. The path is pinned by a test to the canonical THE_FRAMEWORK_DIR/CONVERSATIONS_DIR constants so it cannot drift from where runs actually commit.

  • 053db85: feat(framework): context fragment points the knowledge base at a knowledge-base/ folder (Update Context prompt fragment #683)

    Splits the flat KNOWLEDGE-BASE.md into knowledge-base/FACTS.md and knowledge-base/INSIGHTS.md, and moves DECISIONS.md and MARKET_RESEARCH.md under knowledge-base/, with a knowledge-base/**.md catch-all. The on-before-mergeable prompt and the Market research preset name the same paths.

  • 7c70533: Dashboard: the session log reads like a conversation. Your prompt is its own YOU row and the agent's turn is AGENT, so a message is no longer echoed twice — the redundant You: … log line is gone and the first prompt and every follow-up now read the same way.

    Both your prompts and the agent's replies render as Markdown (headings, bold, lists, code, links), at the log's own density. A short message renders inline; a long one collapses to its first line with a chevron and expands in place on click, so the log stays scannable without hiding the text.

    Also: the sessions rail's home row is now "New session" (a launcher) instead of "Live", and a session's id is always offered for copy in the toolbar (the exact string --resume takes), beside the existing copy-branch action.

  • a1552ac: feat(framework): remote-daemon lane, non-loopback bind guarded by a shared token (Remote-daemon lane: non-loopback bind + shared token (atomic) #1051)

    framework --daemon --host <addr> binds the dashboard daemon to a non-loopback address so a device you own can reach it. Because a daemon that spawns processes is code execution for anyone who finds the port, a non-loopback bind generates and persists a shared token (crypto.randomBytes(32), a top-level Registry.daemonToken, never a preference and never shipped to the browser bundle), and one guard fronts every route (static bundle, /_telefunc, /browser): a request needs a valid fw_daemon cookie or a matching ?token=, else 401, compared with crypto.timingSafeEqual. A valid ?token= sets an HttpOnly; SameSite=Strict cookie and 302s to the clean path, so one cookie rides the RPC, the live-events Channel, and the MJPEG screencast alike. A loopback bind generates nothing and the guard is a no-op, so the local zero-config path is byte-identical. The CLI prints a loud warning and the token URL on any non-loopback bind. Composes with (does not replace) the existing CSRF origin check.

  • 969b9bb: Dashboard launcher and rail UX pass:

    • The prompt editor and its run controls are grouped into one rounded "composer box"; the editor loses its own border and focus ring, and the placeholder and typed text get roomier padding.
    • The submit button is a single arrow icon that stays hidden until the prompt has text, then fades in and slides into place, pushing the model select over smoothly (a spinner shows while starting or sending).
    • The agent/model select sits just left of the submit button, borderless, with the model shown as selected (the default reads "Default", no separator dot).
    • Presets is a borderless slash icon, the options gear is borderless with a smaller count badge, and all controls share one height.
    • "Open in editor" moved out of the options gear onto the workspace editor button, which now opens the checkout and picks the preferred editor.
    • The right rail holds one fixed width for every tab (no expand on Views/Browser).
    • The sessions rail: "New session" is now "New", the "Sessions" heading is replaced by a "Recents" label over the list, the agent logo is smaller, and rows clear the scrollbar.
    • The session toolbar drops the copy-branch and copy-session-id buttons, and the branch dirty/clean indicator sits closer to the branch name.

  • f20add1: Dashboard: the session view holds still. A run ending used to swap the whole page for a different one — the action bar blanked, the output was replaced by "Loading session…", the run overview disappeared and the composer was rebuilt. Live and finished are now the same view, so only what the bar, feed and composer say changes. The action bar is one row at any width (the branch truncates, the least important facts drop out, the buttons never wrap under it), and the composer no longer vanishes on a session that ended without a resumable id: it stays and starts a new session instead.

  • d3cd883: Custom presets can now be saved to a project, not just to you. When you create a preset with a project open, a "Save to" choice lets you keep it private (as before) or commit it into the repo's .the-framework/custom-presets.json, so everyone who clones the project gets it. Shared presets show up in their own "Project presets" group in the Presets menu and under / in the editor, and delete from there. Personal presets still live in your home config and follow you across every project.

  • df6ac36: Run target: wire GitHub Actions into the "Run on" gear (Run target: wire GitHub Actions into the "Run on" gear selector (driver axis) #1050). The options gear gains a single-select "Run on" submenu (Current device, the default; GitHub Actions; Claude web as a disabled placeholder), and picking GitHub Actions runs the turn on a fresh Actions runner via the already-merged ActionsDriver instead of on this device. The choice sticks per project like the agent and model. A new --run-on <local|actions> flag drives it from the CLI; actions reads the repo owner/repo from the origin remote and a user token from GH_TOKEN (repo + workflow scopes). local is unchanged.

  • 9202800: Run-view polish for a GitHub Actions target (Run-view polish for non-local run targets (Actions burst-mode + remote) #1053). A run started with --run-on actions now records its target on the run's meta, and the run view reads it: instead of an apparently-stalled live feed (the ActionsDriver replays its transcript in a burst at the end, on a fresh runner per turn), it shows a "running on GitHub Actions, updates arrive when the run finishes" affordance with a clickable link through to the live Actions run (from the action/notice events the driver emits, which carry the run's html_url). The right rail's Browser pane is gated off for an Actions run, since there is no browser on the runner to screencast. A local or remote-device run is unchanged.

  • c55dcca: Dashboard: a stopped session can be deleted. The sessions rail only ever grew — remove-worktree reclaimed a checkout on disk but kept the row, and nothing removed the record. Delete (a trash button beside Remove, which is now a folder-x icon so the two read apart) takes the session out of the dashboard: its run record and event log, and its worktree if one remains. It confirms first, because unlike remove-worktree the replayable history can't be recovered. It deliberately leaves the git branch and its commits, the committed LOGS.md line and the conversation record — deleting a branch that may carry merged work or an open PR is not something a trash icon should do silently. It refuses while a run is still going.

  • 3dd90eb: Dashboard: the session's branch row fills in about three times faster. It was waiting on a gh pr view (≈574ms, against ~10ms for every git read beside it) that ran twice per session, on every navigation and every poll. That lookup is now read through a single-flight, stale-while-revalidate cache, and it no longer blocks the branch, dirty flag, size, commits or files — they render at git speed while the PR arrives behind them. Opening a PR invalidates the cached answer, and a lookup still in flight is reported as pending rather than as "no PR", so the Open PR button holds off instead of offering to open a second one.

  • c2c5798: Dashboard: semantic status colours, a real Checkbox, a stable Sessions rail, and no emoji glyphs.

    The status colours are now four tokens (--success, --warning, --danger, --info) tuned per
    theme, replacing every raw palette value. Before this, "good" was six different greens, amber-500
    meant both "stopped" and "building, fine", and the flat -500 tones sat near 2:1 contrast on the
    light canvas.

    Checkboxes are a shadcn-style primitive on Base UI instead of bare <input type="checkbox">, so
    they follow the theme and carry the same focus ring as everything else.

    The Sessions rail no longer collapses to a strip when the right rail opens the Browser or Views
    tab; its width is now constant.

    The ✅/❌/⚠️ glyphs in the Enhanced System Prompt disclosure are replaced by a status dot and plain
    text, matching how every other state in the app is drawn.

    Sessions rail rows now show which agent ran them, and lead with something that identifies the
    session: the agent's own session name or its branch when there is no typed prompt, and the start
    time when there is neither. They used to print a bold "(no prompt)" while the timestamp that
    actually told them apart sat beside it in small muted text.

    A clean git tree's dot is neutral rather than green. Green means "added / new / done" everywhere
    else, so a green dot for "nothing changed" sat one pane from the file tree's green dot for "this
    folder has changes".

    Claude's logo is its own starburst rather than the Anthropic wordmark.

  • 6e4eb69: Dashboard: the session toolbar leads with the session's name — the same label the rail shows (your prompt) — instead of the branch. The agent renames its branch near the end of a run, and with the branch as the headline that mutation read as the whole view changing; as muted git context beside a stable name, it no longer does. The shared the-framework/ prefix is dropped from the branch for legibility (the full name stays in the tooltip and copy). The branch summary also stops blanking for a beat when a run stops: it holds the last file counts until the handoff has loaded, so it swaps once instead of flashing empty. One line still, with the label truncating last and the branch, size and summary dropping out as the pane narrows.

Patch Changes

  • 6740853: Never silently downgrade a run into your own checkout when its worktree could not be created (fix(framework): one 10s timeout covers every git op, and a timed-out 'worktree add' silently drops the run into the user's main checkout #997).

    Every run gets its own git worktree (Per-worktree concurrency: allow multiple live runs per project #736) so the agent never edits the working tree you are sitting
    in. When git worktree add failed, the daemon logged a line and ran the agent in the project's main
    checkout instead, mixing its edits into your uncommitted work. That is reachable in normal use: on a
    large repo worktree add writes a whole checkout and can outrun its budget and be killed.

    A project that is not a git repo still falls back to the main checkout, which is the supported
    pre-Per-worktree concurrency: allow multiple live runs per project #736 behavior. A project that is a repo and whose worktree add failed now fails the Start
    instead, and the dashboard shows why. A failed Start is recoverable by starting it again; a checkout
    with agent edits mixed into it is not.

    A worktree add killed mid-write leaves the partial checkout it had already written behind (git
    drops its own administrative entry on the way out, so git worktree prune finds nothing to do), so
    that directory is now removed on the timeout path.

    The fallback's log line also says which case it is: "is not a git repository, so it gets no
    worktree" rather than a generic "no worktree" that read the same either way.

  • 38bb24a: gitTimeoutMs() no longer mistakes the value of a global git option for the subcommand, and a
    conversation commit that keeps failing now says why.

    gitTimeoutMs() picked the subcommand by dropping every word that starts with -, which drops
    the flags but keeps the values of the global options that take one. git -C /repo push therefore
    read as the subcommand /repo, and the push silently got the 30s local-mutation budget instead of
    its intended 120s network budget. The same held for -c key=val, --git-dir, --work-tree,
    --namespace and --exec-path. The leading global options are now skipped properly, value and
    all, so the real subcommand is what picks the budget. No call site in the package passes -C
    today, so no timeout that was already correct changes; this closes the trap before a future call
    site falls into it. gitTimeoutMs()'s signature is unchanged.

    The conversation committer logged only its successes. commitConversations() returns a reason
    when it declines or fails, and the poller dropped it, so a project whose commit failed every tick
    re-queued itself forever while printing nothing at all. The reason is now logged, but on change
    only: the first failure prints one line, and repeats of the same reason stay quiet until the
    reason changes or the commit lands, so a stuck project cannot flood the daemon log one line per
    poll window. The ordinary "no conversation changes" outcome is never reported as a failure.

  • fd8e13f: Stop a throwing onEvent listener from escaping a prompt run (runPrompt).

    runFramework wraps each opts.onEvent(event) call in a try/catch and logs-and-ignores a listener
    that throws, but runPrompt's emit did not. Because emit is called both inside and outside the run's
    try block (the session-start and system-prompt events fire before it), an onEvent listener that threw
    could escape runPrompt uncaught, or skip the run's end event. runPrompt now guards the listener the
    same way, so a bad listener can no longer take the run down.

  • 96f7b6d: Dashboard launcher polish. The run Context picker is now a dropdown that sits in the composer control row next to the presets button, instead of an inline section that pushed the form down. The options gear moved next to the model selector, before Send, and shows a small presence dot rather than a count. "Enhanced System Prompt" is now a dropdown too, sharing the "In play" row. Every menu trigger stays highlighted while its menu is open, and menus open toward the side with room. The prompt editor and all dropdowns now scroll through the shared thin overlay scrollbar, which also removes the doubled border/track the native bar left inside popups.

  • c9864a6: Dashboard: fix the Usage card's dividers rendering at full text brightness, and stop the
    Overview cards stretching into empty space.

    Tailwind v4 defaults an uncoloured border-t to currentColor, so the three dividers in the
    Usage card painted at oklch(0.95 0 0), the body text colour, against hairlines that are
    oklch(0.3 0.01 264) everywhere else in the app. They now use the border token.

    The two Overview card rows also stretched each card to its neighbour's height, so on a quiet
    board half of "Session outcomes" and most of "Working now" were empty card. They are now two
    column stacks that size to their content, which pairs the tall chart against the tall list and
    brings the Projects table a full screen higher.

  • deb130e: Dashboard: the "New preset" form is now a centered modal dialog instead of a panel that pushed the composer controls down. Same fields (name, prompt, and the "Save to" scope choice); a backdrop click, Esc, or the close button dismisses it. Adds a reusable shadcn-style Dialog on Base UI for future forms.

  • 8a8cd75: Dashboard: make the dark: utilities follow the theme toggle instead of the OS.

    Tailwind v4 compiles dark: to @media (prefers-color-scheme: dark) unless a custom variant
    says otherwise, but the dashboard's tokens live on the .dark class that LayoutDefault toggles.
    The two disagreed: picking Dark on a light OS applied the dark tokens while leaving every
    dark:text-* rule unapplied, so diff counts, the stream-lost banner and the daemon-health banner
    kept their light-mode colours on a dark canvas (and the reverse on a dark OS set to Light).
    Declaring @custom-variant dark binds both to the same signal.

  • decace4: Breaking (@gemstack/ai-autopilot): the framework-detection exports are renamed so they no
    longer collide with the user-facing domain presets.

    Two unrelated subsystems were distinguished only by a directory's singular-vs-plural: src/preset/
    (the Open Loop domain bundles, {loops, prompts}) and src/presets/ (framework detection). Both
    were re-exported side by side from the one entry point, so definePreset and selectPreset read
    like a pair while being from different subsystems. "Preset" now means the user-facing domain bundle
    only, which is what the shipped root presets/ markdown and the dashboard already meant by it.

    src/presets/ moves to src/framework-detection/ (internal), and the exports rename:

    Before After
    definePreset defineFrameworkPreset
    Preset FrameworkPreset
    PresetSpec FrameworkPresetSpec
    PresetSignals FrameworkPresetSignals
    PresetScore FrameworkPresetScore
    PresetRegistry FrameworkPresetRegistry
    PresetError FrameworkPresetError
    builtinPresets builtinFrameworkPresets
    builtinPresetRegistry builtinFrameworkPresetRegistry

    detectFramework, vikePreset, nextPreset, FrameworkSignals and FrameworkDetection are
    unchanged: they were already unambiguous. The domain-preset exports (defineDomainPreset,
    selectPreset, composeDomainPresets, loadDomainPreset, builtinDomainPresets,
    builtinPresetsDir, ...) are unchanged.

    No behavior change. As a side effect @gemstack/ai-autopilot no longer exports a definePreset
    that clashes with the unrelated definePreset in @gemstack/framework.

  • ef4f007: Dashboard: the session log labels each row with a plain word instead of its internal event name (Make the session-log labels self-explanatory (e.g. DRIVER) #1035). The badge used to show the raw event kind, so a turn of the AI read DRIVER, the paused state read SETTLED, and the spend row read USAGE. These now read AGENT, WAITING, and COST, and the resume-link row reads RESUME. Kinds that were already clear are unchanged.

  • 8154e20: Dashboard: widen the session-log badge column so "SYSTEM PROMPT" fits on one line. The badge column was narrow enough that the one two-word label wrapped to two lines; the column now fits it (measured at 106px, column is 112px).

  • Updated dependencies [decace4]

    • @gemstack/ai-autopilot@0.12.0

@gemstack/mcp@0.5.0

Minor Changes

  • 9258b84: Remove the dead Mcp server registry and correct the docs to the real mounting API.

    @gemstack/mcp exported Mcp.web() / Mcp.local() / Mcp.getWebServers() / Mcp.getLocalServers(), backed by a store on the __gemstack_mcp_servers__ global. Nothing in the repo ever read that store: no package, no example, no runtime path. Six README and docs pages nevertheless presented Mcp.web(path, ServerClass) as the way to mount a server, so anyone following them registered into a map nobody read, got no error, and got no working endpoint.

    Removed: Mcp, McpWebEntry, McpWebBuilder, and the global store. The real, already-working mounts are unchanged: createMcpHttpHandler(server) for raw node:http / Express / Connect (main entry), createWebRequestHandler(server) for Fetch-style hosts and startStdio(server) for stdio (both from @gemstack/mcp/runtime). Note the shape difference the old docs hid: these take a server instance, not a class.

    @gemstack/mcp takes a minor rather than a major: it is pre-1.0, where a breaking removal is conventionally a minor, and no working code can be relying on the removed surface since nothing ever consumed it. The connector packages take a patch: their published READMEs (and, for @gemstack/mcp-connectors, the mountConnectors JSDoc that ships in its .d.ts) carried the false mount snippet, so the corrected text is worth a release.

@gemstack/ai-sdk@0.6.1

Patch Changes

  • cd121f4: fix(ai-sdk): a store whose unscoped list() hides other users' threads no longer fails open on resume-by-id

    The ai-sdk: a conversation id is loaded and appended to without any owner check #984 owner check settles ownership from ConversationStoreListEntry.userId, falling back to an unscoped store.list() when the caller's scoped listing does not hold the thread. A thread missing from that unscoped listing was read as "no owner recorded" and allowed, which is right for a pre-ai-sdk: a conversation id is loaded and appended to without any owner check #984 ownerless row but wrong for a store whose list() with no user id returns nothing. Such a store implements list(userId) correctly and is fully owner-aware, and a cross-user resume was still allowed.

    The unscoped listing reporting nothing at all, while the store demonstrably holds rows (the caller has threads of their own, or the target thread has messages), now proves the listing is not enumerating the backend, and the resume is refused with ConversationOwnershipError.

    Deliberately unchanged: a listing that reports rows but not this one stays permissive, since an absent thread and a partial listing are indistinguishable there. Ownerless legacy threads on a store that enumerates normally stay resumable, with or without a user on the run, and a store that omits userId from its entries keeps its old permissive behavior.

    ConversationStore gains no members. The listing contract the check depends on is now documented on the interface itself, where an implementer sees it, rather than only on the entry type's userId field.

  • 780ef3e: Internal: the provider layer no longer carries four copies of the same code. Every adapter builds its SDK client through one shared lazyClient() helper, the five pure OpenAI-compatible providers (xai, groq, deepseek, ollama, azure) come from one factory, and the Anthropic stream-event mapping is shared with Bedrock instead of being inlined twice.

    Two side effects worth naming. A first client build is now memoised as a promise, so two concurrent calls share one client rather than racing to construct two, and a failed dynamic import no longer caches. And XaiProvider.name and friends are typed string rather than the literal 'xai', matching the ProviderFactory contract they are consumed through.

    All public exports keep their names, constructor signatures, and default base URLs.

  • 23c1fb7: CachedSubAgentRunStore.load() now returns null (not undefined) when a CacheAdapter resolves undefined on a miss, matching CachedAgentRunStore.load(). The CacheAdapter contract already declares Promise<T | null>, so adapters that honour it are unaffected; adapters that resolve undefined now get the documented null.

    Internal: both run-store families now share one storage implementation. All public exports (InMemoryAgentRunStore, CachedAgentRunStore, newAgentRunId, InMemorySubAgentRunStore, CachedSubAgentRunStore and their types) keep the same shape.

@gemstack/mcp-connector-github@0.2.2

Patch Changes

  • 9258b84: Remove the dead Mcp server registry and correct the docs to the real mounting API.

    @gemstack/mcp exported Mcp.web() / Mcp.local() / Mcp.getWebServers() / Mcp.getLocalServers(), backed by a store on the __gemstack_mcp_servers__ global. Nothing in the repo ever read that store: no package, no example, no runtime path. Six README and docs pages nevertheless presented Mcp.web(path, ServerClass) as the way to mount a server, so anyone following them registered into a map nobody read, got no error, and got no working endpoint.

    Removed: Mcp, McpWebEntry, McpWebBuilder, and the global store. The real, already-working mounts are unchanged: createMcpHttpHandler(server) for raw node:http / Express / Connect (main entry), createWebRequestHandler(server) for Fetch-style hosts and startStdio(server) for stdio (both from @gemstack/mcp/runtime). Note the shape difference the old docs hid: these take a server instance, not a class.

    @gemstack/mcp takes a minor rather than a major: it is pre-1.0, where a breaking removal is conventionally a minor, and no working code can be relying on the removed surface since nothing ever consumed it. The connector packages take a patch: their published READMEs (and, for @gemstack/mcp-connectors, the mountConnectors JSDoc that ships in its .d.ts) carried the false mount snippet, so the corrected text is worth a release.

  • Updated dependencies [9258b84]

    • @gemstack/mcp@0.5.0
    • @gemstack/mcp-connectors@0.2.2

@gemstack/mcp-connector-google-drive@0.2.2

Patch Changes

  • 9258b84: Remove the dead Mcp server registry and correct the docs to the real mounting API.

    @gemstack/mcp exported Mcp.web() / Mcp.local() / Mcp.getWebServers() / Mcp.getLocalServers(), backed by a store on the __gemstack_mcp_servers__ global. Nothing in the repo ever read that store: no package, no example, no runtime path. Six README and docs pages nevertheless presented Mcp.web(path, ServerClass) as the way to mount a server, so anyone following them registered into a map nobody read, got no error, and got no working endpoint.

    Removed: Mcp, McpWebEntry, McpWebBuilder, and the global store. The real, already-working mounts are unchanged: createMcpHttpHandler(server) for raw node:http / Express / Connect (main entry), createWebRequestHandler(server) for Fetch-style hosts and startStdio(server) for stdio (both from @gemstack/mcp/runtime). Note the shape difference the old docs hid: these take a server instance, not a class.

    @gemstack/mcp takes a minor rather than a major: it is pre-1.0, where a breaking removal is conventionally a minor, and no working code can be relying on the removed surface since nothing ever consumed it. The connector packages take a patch: their published READMEs (and, for @gemstack/mcp-connectors, the mountConnectors JSDoc that ships in its .d.ts) carried the false mount snippet, so the corrected text is worth a release.

  • Updated dependencies [9258b84]

    • @gemstack/mcp@0.5.0
    • @gemstack/mcp-connectors@0.2.2

@gemstack/mcp-connectors@0.2.2

Patch Changes

  • 9258b84: Remove the dead Mcp server registry and correct the docs to the real mounting API.

    @gemstack/mcp exported Mcp.web() / Mcp.local() / Mcp.getWebServers() / Mcp.getLocalServers(), backed by a store on the __gemstack_mcp_servers__ global. Nothing in the repo ever read that store: no package, no example, no runtime path. Six README and docs pages nevertheless presented Mcp.web(path, ServerClass) as the way to mount a server, so anyone following them registered into a map nobody read, got no error, and got no working endpoint.

    Removed: Mcp, McpWebEntry, McpWebBuilder, and the global store. The real, already-working mounts are unchanged: createMcpHttpHandler(server) for raw node:http / Express / Connect (main entry), createWebRequestHandler(server) for Fetch-style hosts and startStdio(server) for stdio (both from @gemstack/mcp/runtime). Note the shape difference the old docs hid: these take a server instance, not a class.

    @gemstack/mcp takes a minor rather than a major: it is pre-1.0, where a breaking removal is conventionally a minor, and no working code can be relying on the removed surface since nothing ever consumed it. The connector packages take a patch: their published READMEs (and, for @gemstack/mcp-connectors, the mountConnectors JSDoc that ships in its .d.ts) carried the false mount snippet, so the corrected text is worth a release.

  • Updated dependencies [9258b84]

    • @gemstack/mcp@0.5.0

@gemstack/framework-dashboard@0.3.0

Minor Changes

  • 3e02886: Client connection profiles + "a device I have" in the gear (Client connection profiles + "a device I have" in the gear #1052). The "Run on" submenu gains an "A device I have" section: Local (this machine's own daemon), your saved daemons, and "Add a device". These rows diverge from the driver rows above them: a driver row writes a preference, a device row NAVIGATES the browser to that daemon's origin carrying its token, where the same-origin bootstrap re-authenticates from the cookie it sets. "Add a device" takes the full http://host:port/?token=… URL a box prints on its network bind (any reachable host: LAN IP, tailnet name, tunnel URL), parsing the origin and token out of one paste. Profiles live in per-browser localStorage, so the token never reaches the daemon's registry file. A "connected to " indicator in the header shows which daemon the dashboard is talking to.
  • 9202800: Run-view polish for a GitHub Actions target (Run-view polish for non-local run targets (Actions burst-mode + remote) #1053). A run started with --run-on actions now records its target on the run's meta, and the run view reads it: instead of an apparently-stalled live feed (the ActionsDriver replays its transcript in a burst at the end, on a fresh runner per turn), it shows a "running on GitHub Actions, updates arrive when the run finishes" affordance with a clickable link through to the live Actions run (from the action/notice events the driver emits, which carry the run's html_url). The right rail's Browser pane is gated off for an Actions run, since there is no browser on the runner to screencast. A local or remote-device run is unchanged.

Patch Changes

@gemstack/example-autopilot-quickstart@0.0.12

Patch Changes

  • Updated dependencies [cd121f4]
  • Updated dependencies [780ef3e]
  • Updated dependencies [23c1fb7]
  • Updated dependencies [decace4]
    • @gemstack/ai-sdk@0.6.1
    • @gemstack/ai-autopilot@0.12.0

@gemstack/example-bootstrap-quickstart@0.0.12

Patch Changes

  • Updated dependencies [cd121f4]
  • Updated dependencies [780ef3e]
  • Updated dependencies [23c1fb7]
  • Updated dependencies [decace4]
    • @gemstack/ai-sdk@0.6.1
    • @gemstack/ai-autopilot@0.12.0

@gemstack/example-connectors-quickstart@0.0.4

Patch Changes

  • Updated dependencies [9258b84]
    • @gemstack/mcp@0.5.0
    • @gemstack/mcp-connectors@0.2.2

@gemstack/example-framework-demo@0.0.9

Patch Changes

@gemstack/example-mcp-quickstart@0.0.3

Patch Changes

  • Updated dependencies [9258b84]
    • @gemstack/mcp@0.5.0

@github-actions
github-actions Bot force-pushed the changeset-release/main branch 30 times, most recently from 3b1f5ab to a6657f4 Compare July 22, 2026 22:09
@github-actions
github-actions Bot force-pushed the changeset-release/main branch 9 times, most recently from e64dc48 to 4d2ce76 Compare July 23, 2026 00:23
@github-actions
github-actions Bot force-pushed the changeset-release/main branch from 4d2ce76 to ec74820 Compare July 23, 2026 00:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

0 participants