@mastra/code-sdk 1.2.0-alpha.8 → 1.2.0

package/CHANGELOG.md

@@ -1,5 +1,242 @@
 # @mastra/code-sdk
 
+## 1.2.0
+
+### Minor Changes
+
+- Added Dynamic Workflow creation and management to Mastra Code, including discovery-backed authoring, immediate persistence, execution, and deletion. ([#21210](https://github.com/mastra-ai/mastra/pull/21210))
+
+  ```ts
+  import { listWorkflows, runWorkflow } from '@mastra/code-sdk/workflows/service';
+
+  const { workflows } = await listWorkflows(mastra);
+  const workflow = workflows[0];
+  if (workflow) {
+    await runWorkflow(mastra, workflow.id, { topic: 'dynamic workflows' });
+  }
+  ```
+
+- Added `processors` and `signalProviders` to the Mastra Code plugin contract, so a plugin can contribute more than tools. ([#20848](https://github.com/mastra-ai/mastra/pull/20848))
+
+  **Processors**
+
+  A plugin can now extend the agent pipeline directly, passing a bare array for input processors or an object for both lanes. Plugin processors run after the processors Mastra Code configures and before the channel and memory layers the Agent appends. The slot isn't configurable, and the processors are resolved before every LLM call. Enabling, disabling, or updating a plugin applies on the next request instead of requiring a restart.
+
+  **Signal providers**
+
+  A plugin can also ship a signal provider, which monitors an external source and pushes notifications into a thread. Providers are long-lived, so the SDK owns their lifecycle instead of handing them to the agent: it registers Mastra on them, connects them to the coding agent, starts them polling, and stops them when the plugin is updated, disabled, or uninstalled. That makes a provider installed from a GitHub repository survive a mid-session update of that repository. Only one provider with a given id runs at a time, and a provider that fails to start is isolated from the rest of its plugin.
+
+  Field resolvers also receive `getController()` and `getActiveSession()` on the plugin context, so a plugin can read the running session lazily at the moment it needs it.
+
+  For embedders that inject their own `pluginManager` into `createMastraCode`: the `PluginManager` contract now also requires `onReload`, `getPluginSignalProviders`, and `setRuntime`, so hand-built manager implementations must add these methods.
+
+- Add a reasoning-effort configuration surface across mastracode and Factory (fixes #20766): ([#20884](https://github.com/mastra-ai/mastra/pull/20884))
+
+  - New `max` thinking level (mapped to `reasoning effort: max` for OpenAI Codex and Anthropic `effort`).
+  - Anthropic extended-thinking wiring: the session thinking level now applies to anthropic/claude-opus-4-7 and other Anthropic models via provider thinking/effort options (previously OpenAI-only).
+  - New `models.modeThinkingDefaults` setting: per-mode (build/plan/fast) default thinking levels, resolved at request time with precedence session override → mode default → global `preferences.thinkingLevel`. Configuration changes now apply to the next request of every session, including automated Factory runs.
+  - Factory: new Settings → Defaults controls for editing global and per-mode thinking defaults in local deployments.
+  - TUI: `/think` now sets a session-only override, supports `/think default` to clear it, and `/think status` reports the effective level with provenance (session override / mode default / global default).
+
+  Example `settings.json` configuration:
+
+  ```json
+  {
+    "preferences": { "thinkingLevel": "medium" },
+    "models": {
+      "modeThinkingDefaults": {
+        "build": "high",
+        "plan": "max",
+        "fast": "off"
+      }
+    }
+  }
+  ```
+
+- Added MCP disable-state controls to the MCP manager. Servers can be disabled for the current project or for every project, the state persists across runs in an app-data `mcp-state.json` (user MCP config files are never mutated), and disabled servers stay visible in statuses via the new `disabled`/`disabledScope` fields on `McpServerStatus`. ([#20834](https://github.com/mastra-ai/mastra/pull/20834))
+
+  ```ts
+  await mcpManager.setServerDisabled('filesystem', true); // project scope
+  await mcpManager.setServerDisabled('filesystem', true, { global: true }); // all projects
+  await mcpManager.setAllDisabled(true, { global: true }); // global kill switch
+  mcpManager.isAllDisabledGlobally();
+  mcpManager.getDisabledServers();
+  ```
+
+### Patch Changes
+
+- dependencies updates: ([#19783](https://github.com/mastra-ai/mastra/pull/19783))
+  - Updated dependency [`posthog-node@^5.46.1` ↗︎](https://www.npmjs.com/package/posthog-node/v/5.46.1) (from `^5.37.0`, in `dependencies`)
+
+- dependencies updates: ([#20406](https://github.com/mastra-ai/mastra/pull/20406))
+  - Updated dependency [`@aws-sdk/credential-providers@^3.1095.0` ↗︎](https://www.npmjs.com/package/@aws-sdk/credential-providers/v/3.1095.0) (from `^3.864.0`, in `dependencies`)
+
+- Persist the browser viewport as a preset name, a `{ width, height }` size, or `'window'` in settings, and drop unusable stored values back to the default rather than passing them to the browser. ([#21010](https://github.com/mastra-ai/mastra/pull/21010))
+
+- Fixed tenant credential resolution for session-based authentication providers. Background Factory runs now resolve the authenticated user and active organization from session-wrapped request context values instead of falling back to an empty credential store. ([#21008](https://github.com/mastra-ai/mastra/pull/21008))
+
+- Added a `model` option to Stagehand browser settings, so browser automation can run on a chosen provider instead of a fixed default: ([#20993](https://github.com/mastra-ai/mastra/pull/20993))
+
+  ```ts
+  import { createBrowserFromSettings } from '@mastra/code-sdk/onboarding/settings';
+
+  const browser = await createBrowserFromSettings({
+    enabled: true,
+    provider: 'stagehand',
+    headless: true,
+    stagehand: { env: 'LOCAL', model: 'anthropic/claude-sonnet-4-5' },
+  });
+  ```
+
+  The model must be provider-qualified as `<provider>/<model>`. Values Stagehand cannot resolve, such as a bare `gpt-4.1`, are ignored so the browser still starts.
+
+- Fixed woken notifications running against the controller-level workspace, which is `undefined` for dynamic workspace factories. They now use the workspace of the session that owns the target thread. ([#21144](https://github.com/mastra-ai/mastra/pull/21144))
+
+- Fixed deferred notification deliveries failing with "No model selected" when the woken thread had no request context. The notification dispatch workflow re-sends deferred and summarized notifications long after the originating send, so any stream options attached to the original signal are gone; waking an idle thread then had no way to resolve a model. The notification delivery policy now drives the fix: `NotificationDeliveryDecision` accepts `streamOptions`, and the dispatcher re-runs the agent's delivery policy at delivery time (via the new `agent.resolveNotificationDeliveryDecision()`) to attach freshly resolved stream options to both individual and summary deliveries. Only `streamOptions` is honored at dispatch time; the record's persisted schedule still governs when and how it is delivered. Receipt-time sends also honor the policy's `streamOptions` now: an immediate deliver or summarize-now wake attaches them too, with caller-supplied stream options taking precedence. Mastra Code wires its session-based stream options resolver through the Code Agent's `deliveryPolicy.decide`, layered on the default decision logic. This fix applies to threads whose session is live in the current process; deliveries with no resolvable session fall back to a bare wake, and the "No model selected" error now distinguishes the case where a run started without any controller session context. This re-lands the capability removed by the #18637 revert in the policy-driven shape that revert called for, and the `@mastra/github-signals` bump only widens its `getNotificationStreamOptions` callback return type to allow `undefined`. ([#21113](https://github.com/mastra-ai/mastra/pull/21113))
+
+- Fixed custom provider models saved with a stray `mastracode/` prefix in settings, which broke selecting and using them after choosing them from `/models` (#20799) ([#20804](https://github.com/mastra-ai/mastra/pull/20804))
+
+- Fixed authentication failures for unprefixed direct-provider models by using provider environment credentials when no provider-specific stored credential exists. ([#21197](https://github.com/mastra-ai/mastra/pull/21197))
+
+- Fixed Factory interactive plans so they are stored as browsable artifacts. ([#21173](https://github.com/mastra-ai/mastra/pull/21173))
+
+- Include worktree identity (`worktree_path`, `branch`, `main_repo_path`) in `SessionStart` and `SessionEnd` hook payloads when a session runs in a git worktree, so hooks can provision and tear down per-worktree resources. ([#21034](https://github.com/mastra-ai/mastra/pull/21034))
+
+- Updated dependencies [[`e7109ee`](https://github.com/mastra-ai/mastra/commit/e7109ee6f731bacc79c885906f3c7dca8d8f013a), [`b8ce7ec`](https://github.com/mastra-ai/mastra/commit/b8ce7ec96e39343c6c2f36d12d68a9ad816c09f7), [`2e4624e`](https://github.com/mastra-ai/mastra/commit/2e4624edb6917e61249cb60ee377735e7af7e4a9), [`45a9147`](https://github.com/mastra-ai/mastra/commit/45a914741f578754d79d8b7de7b4e4f304d8e14a), [`a3a3624`](https://github.com/mastra-ai/mastra/commit/a3a3624f646b98e409424d8defccbd334da9e8b8), [`772c0c8`](https://github.com/mastra-ai/mastra/commit/772c0c897cec383258de2e6178147f8014767c7b), [`6246914`](https://github.com/mastra-ai/mastra/commit/62469146636911f3cbbe0880bd011c6a897a59a7), [`6445eba`](https://github.com/mastra-ai/mastra/commit/6445eba6020abac681aba1cc9289f446cb400cbe), [`1315d8f`](https://github.com/mastra-ai/mastra/commit/1315d8f17e8e7acb61cca46b72a1d42f6d00d289), [`86b7b77`](https://github.com/mastra-ai/mastra/commit/86b7b777980d30f66e1fd134a37d2af4c22e54cc), [`1c75e32`](https://github.com/mastra-ai/mastra/commit/1c75e32f7fc0b9fb6f548b4407feaec8a1440212), [`296dc9a`](https://github.com/mastra-ai/mastra/commit/296dc9af29f3616e786c7825ec32e0df92d754c5), [`f59032a`](https://github.com/mastra-ai/mastra/commit/f59032a73699443555a08a479e7ac578975784f2), [`f59032a`](https://github.com/mastra-ai/mastra/commit/f59032a73699443555a08a479e7ac578975784f2), [`1bdfde1`](https://github.com/mastra-ai/mastra/commit/1bdfde1f4061b7c0550253a1512a2118debe7996), [`cdd5c33`](https://github.com/mastra-ai/mastra/commit/cdd5c33ac6c7118a9f139e6dc0e14e6a8ae31658), [`3f73c07`](https://github.com/mastra-ai/mastra/commit/3f73c076727e8c36b4fff7a1b40290fb68957fa8), [`cdd5c33`](https://github.com/mastra-ai/mastra/commit/cdd5c33ac6c7118a9f139e6dc0e14e6a8ae31658), [`6bff877`](https://github.com/mastra-ai/mastra/commit/6bff877e214695ff8d9c84b06c13a6e6bcf9f1ed), [`772c0c8`](https://github.com/mastra-ai/mastra/commit/772c0c897cec383258de2e6178147f8014767c7b), [`d7cf7fa`](https://github.com/mastra-ai/mastra/commit/d7cf7fafc1ae1b50bd8462dd0e6c671a8606db93), [`7c1ebb1`](https://github.com/mastra-ai/mastra/commit/7c1ebb15690c4b3f0eabb19077cf8af573311e57), [`0f9a448`](https://github.com/mastra-ai/mastra/commit/0f9a448502157e59f7b76f24360ad497168f5ef8), [`f5a17d9`](https://github.com/mastra-ai/mastra/commit/f5a17d95c19e7d4149996932bd8d1905089f031d), [`578bf2e`](https://github.com/mastra-ai/mastra/commit/578bf2e6a88e9d5b8bf502204e15a95dfbb679ae), [`c47165c`](https://github.com/mastra-ai/mastra/commit/c47165c983c87594c6952f1fd2fa51a90205034c), [`289f4ce`](https://github.com/mastra-ai/mastra/commit/289f4ce16e3293370440172132c52ee787cbc09f), [`deaf0c4`](https://github.com/mastra-ai/mastra/commit/deaf0c4c38fe2e988a094c63bbd8899436f4e579), [`df31eb0`](https://github.com/mastra-ai/mastra/commit/df31eb0c7087d782a0d9346e467f9a4af4b0eef6), [`9571e3a`](https://github.com/mastra-ai/mastra/commit/9571e3a06ed2c5220196460bf82a2129255c3a8b), [`4f16ff8`](https://github.com/mastra-ai/mastra/commit/4f16ff824bf2f9b0ddc93f210477c10c8a4fb1ab), [`b4c89b4`](https://github.com/mastra-ai/mastra/commit/b4c89b4371b0c86da57403ad1a3b3ef0681f3128), [`e6534fa`](https://github.com/mastra-ai/mastra/commit/e6534fab031216f6cb48c4c9907cbfdce9d60bc6), [`210cb7a`](https://github.com/mastra-ai/mastra/commit/210cb7a167998c7bbf72cb3b93e6eb0563330239), [`06b2d87`](https://github.com/mastra-ai/mastra/commit/06b2d87e63bcdd0ed59215c6789692b9b12de376), [`1c67d85`](https://github.com/mastra-ai/mastra/commit/1c67d85e9da8285662f4dbbf47e0378c3fee0747), [`ac01d63`](https://github.com/mastra-ai/mastra/commit/ac01d6355974aec73fdb8781449ed12bac582094), [`8627c23`](https://github.com/mastra-ai/mastra/commit/8627c23d794485fb97d533fc2a7a8323bfec2225), [`03ebe06`](https://github.com/mastra-ai/mastra/commit/03ebe06aeb671d7127b700e53853ed0759e4c19f), [`80a3324`](https://github.com/mastra-ai/mastra/commit/80a33245d3110204de6f56d61211523ffe338692), [`e44e8f3`](https://github.com/mastra-ai/mastra/commit/e44e8f370b66c339ddcaba946d33da6d3c3f06cd), [`d9d2881`](https://github.com/mastra-ai/mastra/commit/d9d2881ede6dd6c023d144215fc812062aed0890), [`a810a05`](https://github.com/mastra-ai/mastra/commit/a810a058f62ad407cfc1701e0be36ae91145d7cf), [`ba24be6`](https://github.com/mastra-ai/mastra/commit/ba24be662439c331ab23a600041f93803c89eca8), [`842b5fe`](https://github.com/mastra-ai/mastra/commit/842b5fe22b6a7fa811bd14e48eb9af523ac989f2), [`26ff3b9`](https://github.com/mastra-ai/mastra/commit/26ff3b9144132bd8570e4796d436fb57a9a2bf24), [`3ff2be3`](https://github.com/mastra-ai/mastra/commit/3ff2be3a5a579d2e2d7084ba0624ec24faadce4d), [`990611b`](https://github.com/mastra-ai/mastra/commit/990611ba76eb876d86c9c594371ae5f02f94b432), [`80bdf3a`](https://github.com/mastra-ai/mastra/commit/80bdf3ae16ade6ff63bde0cb16fa2df8ab7dd4dd), [`c967a5e`](https://github.com/mastra-ai/mastra/commit/c967a5eec150c5dc5418c4a4388982d1fb7ad27c), [`195e83c`](https://github.com/mastra-ai/mastra/commit/195e83c077687f6016fd8090324975c8d8cea50b), [`dc4a25d`](https://github.com/mastra-ai/mastra/commit/dc4a25d41af4e2fe97a816070eaec6aa963ab53b), [`9ba1247`](https://github.com/mastra-ai/mastra/commit/9ba12470c77f1c03642d720ce67e517e878f666e), [`fd96298`](https://github.com/mastra-ai/mastra/commit/fd96298a8367622f4ebfcaa97b5b6c1fbbd14564), [`66bbfb5`](https://github.com/mastra-ai/mastra/commit/66bbfb5f05b473d39f88c0e4a481ccac41634f3a), [`dc4a25d`](https://github.com/mastra-ai/mastra/commit/dc4a25d41af4e2fe97a816070eaec6aa963ab53b), [`f8da216`](https://github.com/mastra-ai/mastra/commit/f8da21633e7eb0e31c9ce0fc30567870d19416d3), [`9168b80`](https://github.com/mastra-ai/mastra/commit/9168b80120a82816b1b9f2385e788bf86fa5d9cd), [`e7a5da4`](https://github.com/mastra-ai/mastra/commit/e7a5da4ef8e4dd452d2f232961b4e682a85ffe43), [`4a09a9c`](https://github.com/mastra-ai/mastra/commit/4a09a9c0474ef643558fcb5f0edc542b82f1cab0), [`5f798b3`](https://github.com/mastra-ai/mastra/commit/5f798b3362e9bdf4d690f85245606e146eef60b9), [`6a84954`](https://github.com/mastra-ai/mastra/commit/6a84954a2667f85b6d59da652dab1bbff007ccb0), [`1e83a47`](https://github.com/mastra-ai/mastra/commit/1e83a4734ab61ba5926af6793e3569a78b72ed37), [`52d8ef0`](https://github.com/mastra-ai/mastra/commit/52d8ef03801f1deb7ee48532fc4190dd4a33916c), [`cdd5c33`](https://github.com/mastra-ai/mastra/commit/cdd5c33ac6c7118a9f139e6dc0e14e6a8ae31658), [`7fdcaa6`](https://github.com/mastra-ai/mastra/commit/7fdcaa66105d64290f9b14432a12ec99f39c4d3a), [`d6c56f9`](https://github.com/mastra-ai/mastra/commit/d6c56f951db3213330b98b0abafa9778c8770e58), [`e08e789`](https://github.com/mastra-ai/mastra/commit/e08e789c1bf4cd2fe46363f7a4728536ceccc9bd), [`bf936e2`](https://github.com/mastra-ai/mastra/commit/bf936e2c89b2ff0dad5695b873ddc009ba96d41e), [`7fb580a`](https://github.com/mastra-ai/mastra/commit/7fb580ac73fbcacf2ff00872a3395f73ae1b9fa5), [`59866f2`](https://github.com/mastra-ai/mastra/commit/59866f2e0a7986bea3b418aaa2c2f79a77d33719), [`ed5d606`](https://github.com/mastra-ai/mastra/commit/ed5d606739c5e3fbdfa9f272df7809aa5ab43b1d), [`c71e307`](https://github.com/mastra-ai/mastra/commit/c71e3077e69eae3f25aa628e3778f153a9d6ab36), [`e7a5da4`](https://github.com/mastra-ai/mastra/commit/e7a5da4ef8e4dd452d2f232961b4e682a85ffe43), [`289f4ce`](https://github.com/mastra-ai/mastra/commit/289f4ce16e3293370440172132c52ee787cbc09f), [`f53d5bd`](https://github.com/mastra-ai/mastra/commit/f53d5bd4885b29e4ac29a428a6044088ea8d6aa3), [`32980a3`](https://github.com/mastra-ai/mastra/commit/32980a3e2413d0274ac244d32c37d910edc13f00), [`a4f6806`](https://github.com/mastra-ai/mastra/commit/a4f68065d082af57770e7ee3d34996fdc914dbd1), [`01a2943`](https://github.com/mastra-ai/mastra/commit/01a2943a7d886edefdff072bfa51f055bab54437), [`82e3365`](https://github.com/mastra-ai/mastra/commit/82e3365ef7c9bf7bee2e7a7029035ea262d68895), [`6104347`](https://github.com/mastra-ai/mastra/commit/61043473ba6bfd0a25156824e853e13165562e6c), [`261edb9`](https://github.com/mastra-ai/mastra/commit/261edb9bc0cfbed2b77090b87562307a360f1a04), [`35cc901`](https://github.com/mastra-ai/mastra/commit/35cc90102cf834a84827acaf9eee0b6d6d1e2a3b), [`a8b4cf0`](https://github.com/mastra-ai/mastra/commit/a8b4cf02823cffebc4751a53337dfacf097c1ae1), [`a53f19b`](https://github.com/mastra-ai/mastra/commit/a53f19b9badb8bd349fe2f885814e9db6b7d3c4b), [`edfe906`](https://github.com/mastra-ai/mastra/commit/edfe906f7926faf8c63c21b3c87e797985557feb), [`ffdfa25`](https://github.com/mastra-ai/mastra/commit/ffdfa25b5cdf3ecfd0c5a2e96363c3ba50995ce9), [`9571e3a`](https://github.com/mastra-ai/mastra/commit/9571e3a06ed2c5220196460bf82a2129255c3a8b), [`65b1183`](https://github.com/mastra-ai/mastra/commit/65b11832834f87a9bc8719391deb27559de5138a), [`333785c`](https://github.com/mastra-ai/mastra/commit/333785c93cbb01e42c60167e995457c28897ddbf), [`bda2235`](https://github.com/mastra-ai/mastra/commit/bda22353ee28f2df0eaea555f7cae1549f979c0b), [`efd5c81`](https://github.com/mastra-ai/mastra/commit/efd5c81cc25fde3c2ddd86fc1178deb4ec176e19), [`1b482c2`](https://github.com/mastra-ai/mastra/commit/1b482c2d89244dd758c41e5f927a2b44041388d2), [`5dba2a4`](https://github.com/mastra-ai/mastra/commit/5dba2a41600385751f5aace79878904e1972609d), [`45bfb88`](https://github.com/mastra-ai/mastra/commit/45bfb88fd52f1dd3be20e2a38905777c96499c90), [`ff28284`](https://github.com/mastra-ai/mastra/commit/ff2828416f14daff9d956e6a352fdaa23c950979), [`4bcdfaf`](https://github.com/mastra-ai/mastra/commit/4bcdfaf0eac3199d7cb171b0a19a92c9c341eea4), [`e3b9307`](https://github.com/mastra-ai/mastra/commit/e3b9307098daefbfae2a52ae2ef51bc9fc701190), [`d6834c5`](https://github.com/mastra-ai/mastra/commit/d6834c5a7866b16734d23900163c2414ed70d791), [`43ea2a1`](https://github.com/mastra-ai/mastra/commit/43ea2a1f5c5867d0f41254047a786e96cd00798b), [`f33264f`](https://github.com/mastra-ai/mastra/commit/f33264f517ae603279afd5c4251e2b40f6dd3618), [`689f2c4`](https://github.com/mastra-ai/mastra/commit/689f2c4b6c0835fe455702b01d21daa8abcd9331), [`fcd0667`](https://github.com/mastra-ai/mastra/commit/fcd0667a4e378be35c9a1b1eb19cce78fbfd7282), [`cfd0d9e`](https://github.com/mastra-ai/mastra/commit/cfd0d9ec77ec3c69dd96f79cdb579e03d79f22ce), [`acc3513`](https://github.com/mastra-ai/mastra/commit/acc3513b19f79bf0a7ec2998694580edca54086c), [`1670533`](https://github.com/mastra-ai/mastra/commit/1670533986f6bacf567746245348125e3a106448), [`a7eb4a1`](https://github.com/mastra-ai/mastra/commit/a7eb4a11450f6170274ed5141bffe821d4fdd5a6), [`1b1dd7b`](https://github.com/mastra-ai/mastra/commit/1b1dd7bc0e59b7a8bfabd09a3eec1ccd95b4c2f3), [`0976933`](https://github.com/mastra-ai/mastra/commit/0976933142333ec78451feef265b68bcb45aa5e7), [`242b945`](https://github.com/mastra-ai/mastra/commit/242b94558777bfbdeb42cbfea84afff0b6ad0633), [`c52d346`](https://github.com/mastra-ai/mastra/commit/c52d3462ec831a5d95926ecd3d3373f5928ad2e5), [`af4636a`](https://github.com/mastra-ai/mastra/commit/af4636a74463275d71c1d13a38f7d2b738f128bf), [`01a2943`](https://github.com/mastra-ai/mastra/commit/01a2943a7d886edefdff072bfa51f055bab54437), [`2eabc09`](https://github.com/mastra-ai/mastra/commit/2eabc097d86d52fbd0123da36a7c874154cc384f), [`0023e79`](https://github.com/mastra-ai/mastra/commit/0023e7919431078280abd11c89d1edeae35fcc69), [`c2ad51e`](https://github.com/mastra-ai/mastra/commit/c2ad51e2467f901eecba8c9f4a45e22a50bd7c18), [`25ca73d`](https://github.com/mastra-ai/mastra/commit/25ca73d25dee7ce9f0ca72939e3a505c4db7257e), [`2f9ef3f`](https://github.com/mastra-ai/mastra/commit/2f9ef3f4ca06fc2dcdd5088c26b7f4da6a016791), [`e7eefcb`](https://github.com/mastra-ai/mastra/commit/e7eefcb162cda7c493e8c3bf43050ead0efbcb2c), [`fea5cae`](https://github.com/mastra-ai/mastra/commit/fea5caedc7e2cfea51784a15e015952692027abf), [`4d7aca2`](https://github.com/mastra-ai/mastra/commit/4d7aca2fe75f225c83d1502d63079568e6ec163f), [`e1cead1`](https://github.com/mastra-ai/mastra/commit/e1cead17b5f3653cf00d2f90cc19b113119c02ba), [`01a2943`](https://github.com/mastra-ai/mastra/commit/01a2943a7d886edefdff072bfa51f055bab54437), [`25956fc`](https://github.com/mastra-ai/mastra/commit/25956fc8841780d506acb22b618fdb4dcf6c4e21), [`d9d93b2`](https://github.com/mastra-ai/mastra/commit/d9d93b25e4a65ad5fa153fa35be7ed149c8d587f), [`c4ec889`](https://github.com/mastra-ai/mastra/commit/c4ec889561c0264c43f66d04d587bee4ce35e792), [`4b59f78`](https://github.com/mastra-ai/mastra/commit/4b59f786cbc9a7d1ef07a07517dbd4b96865e99d), [`eeae63e`](https://github.com/mastra-ai/mastra/commit/eeae63e7fbe8e1f237adc69bca6e2ac13c5ca907), [`3dc97ea`](https://github.com/mastra-ai/mastra/commit/3dc97ea415fad353b48a13095fad1835933cc12a), [`9ba1247`](https://github.com/mastra-ai/mastra/commit/9ba12470c77f1c03642d720ce67e517e878f666e), [`94e7ae9`](https://github.com/mastra-ai/mastra/commit/94e7ae970b37c888cd1244ef013292639a2fe6d1), [`d97107a`](https://github.com/mastra-ai/mastra/commit/d97107a7edc517ae8feddf914fc43cd80a66c0a8), [`e6a2860`](https://github.com/mastra-ai/mastra/commit/e6a2860649cc51f87d32d78b766ae2126446ba07), [`7010c5d`](https://github.com/mastra-ai/mastra/commit/7010c5d15728bf9c5dfe4fb6b1bf80ce23bf143a), [`bab06b1`](https://github.com/mastra-ai/mastra/commit/bab06b18923873a584bdfc71a6b4ec7fb4727fb7), [`3d01cd3`](https://github.com/mastra-ai/mastra/commit/3d01cd387321b6f9c5cac31d487c84bf51b19c78), [`7bf3086`](https://github.com/mastra-ai/mastra/commit/7bf308663f0115ca74ad20554ade740f06640859), [`4c186a0`](https://github.com/mastra-ai/mastra/commit/4c186a017275f45e6ed4c09de0f89550e2d09e8c), [`b0fa077`](https://github.com/mastra-ai/mastra/commit/b0fa077bcbc9b08551846fe372a0d3d15b71ed72), [`0282e16`](https://github.com/mastra-ai/mastra/commit/0282e16115538c8e9b248b90f0748eb01cb5dc98), [`a8dd139`](https://github.com/mastra-ai/mastra/commit/a8dd1391a9fe9a6632c25809ef236980afa9a020), [`6a667b4`](https://github.com/mastra-ai/mastra/commit/6a667b4b7cd6a93fe41fcdd357b08c5a8c09b9ab), [`9be8878`](https://github.com/mastra-ai/mastra/commit/9be8878dcf0388e84fc4873e0eec27bd49b881a4), [`e5786be`](https://github.com/mastra-ai/mastra/commit/e5786be02bb903073082bd9d6da880ebaacc343f), [`e5786be`](https://github.com/mastra-ai/mastra/commit/e5786be02bb903073082bd9d6da880ebaacc343f), [`2440e09`](https://github.com/mastra-ai/mastra/commit/2440e096ea6c2def1ccc1eb2d0f3f5b88c4af940), [`d6c63e4`](https://github.com/mastra-ai/mastra/commit/d6c63e4b757babb95a735d0e421d4dac67c1dcf1), [`2093fbd`](https://github.com/mastra-ai/mastra/commit/2093fbd53bb744bae19ec89f6d73db9a66fbe8a7), [`a59049b`](https://github.com/mastra-ai/mastra/commit/a59049b1652a13efff66ac826326b5ed9a550342), [`7bd85ea`](https://github.com/mastra-ai/mastra/commit/7bd85ea7588b71c25ce9f4019c88f8539be5dcbc), [`83fa004`](https://github.com/mastra-ai/mastra/commit/83fa0044bfda8b703a83883dbd8bef204844d13f), [`a463cdf`](https://github.com/mastra-ai/mastra/commit/a463cdf1c95c3059e70f0bff27959e8558bb899d), [`e7a5da4`](https://github.com/mastra-ai/mastra/commit/e7a5da4ef8e4dd452d2f232961b4e682a85ffe43), [`e4d3761`](https://github.com/mastra-ai/mastra/commit/e4d376143cf3322885a7e6e4048e536ce441f785), [`7b4393d`](https://github.com/mastra-ai/mastra/commit/7b4393d557411fdcf07b0e30e5acaf7cc85154ae), [`0ea6b80`](https://github.com/mastra-ai/mastra/commit/0ea6b8001408ce02b56e8be0536b0fd8cbaf8ad2)]:
+  - @mastra/core@1.58.0
+  - @mastra/libsql@1.20.0
+  - @mastra/pg@1.20.0
+  - @mastra/github-signals@0.2.5
+  - @mastra/agent-browser@0.5.1
+  - @mastra/stagehand@0.3.2
+  - @mastra/observability@1.16.6
+  - @mastra/schema-compat@1.3.6
+  - @mastra/memory@1.26.1
+  - @mastra/mcp@1.16.0
+  - @mastra/duckdb@1.6.1
+
+## 1.2.0-alpha.18
+
+### Minor Changes
+
+- Added Dynamic Workflow creation and management to Mastra Code, including discovery-backed authoring, immediate persistence, execution, and deletion. ([#21210](https://github.com/mastra-ai/mastra/pull/21210))
+
+  ```ts
+  import { listWorkflows, runWorkflow } from '@mastra/code-sdk/workflows/service';
+
+  const { workflows } = await listWorkflows(mastra);
+  const workflow = workflows[0];
+  if (workflow) {
+    await runWorkflow(mastra, workflow.id, { topic: 'dynamic workflows' });
+  }
+  ```
+
+### Patch Changes
+
+- Updated dependencies [[`296dc9a`](https://github.com/mastra-ai/mastra/commit/296dc9af29f3616e786c7825ec32e0df92d754c5), [`4a09a9c`](https://github.com/mastra-ai/mastra/commit/4a09a9c0474ef643558fcb5f0edc542b82f1cab0), [`1e83a47`](https://github.com/mastra-ai/mastra/commit/1e83a4734ab61ba5926af6793e3569a78b72ed37), [`ff28284`](https://github.com/mastra-ai/mastra/commit/ff2828416f14daff9d956e6a352fdaa23c950979), [`1670533`](https://github.com/mastra-ai/mastra/commit/1670533986f6bacf567746245348125e3a106448)]:
+  - @mastra/core@1.58.0-alpha.16
+
+## 1.2.0-alpha.17
+
+### Patch Changes
+
+- Updated dependencies [[`dc4a25d`](https://github.com/mastra-ai/mastra/commit/dc4a25d41af4e2fe97a816070eaec6aa963ab53b), [`dc4a25d`](https://github.com/mastra-ai/mastra/commit/dc4a25d41af4e2fe97a816070eaec6aa963ab53b)]:
+  - @mastra/memory@1.26.1-alpha.7
+  - @mastra/core@1.58.0-alpha.15
+  - @mastra/libsql@1.20.0-alpha.3
+  - @mastra/pg@1.20.0-alpha.4
+
+## 1.2.0-alpha.16
+
+### Patch Changes
+
+- Updated dependencies [[`210cb7a`](https://github.com/mastra-ai/mastra/commit/210cb7a167998c7bbf72cb3b93e6eb0563330239), [`5f798b3`](https://github.com/mastra-ai/mastra/commit/5f798b3362e9bdf4d690f85245606e146eef60b9), [`01a2943`](https://github.com/mastra-ai/mastra/commit/01a2943a7d886edefdff072bfa51f055bab54437), [`01a2943`](https://github.com/mastra-ai/mastra/commit/01a2943a7d886edefdff072bfa51f055bab54437), [`25ca73d`](https://github.com/mastra-ai/mastra/commit/25ca73d25dee7ce9f0ca72939e3a505c4db7257e), [`e1cead1`](https://github.com/mastra-ai/mastra/commit/e1cead17b5f3653cf00d2f90cc19b113119c02ba), [`01a2943`](https://github.com/mastra-ai/mastra/commit/01a2943a7d886edefdff072bfa51f055bab54437), [`d97107a`](https://github.com/mastra-ai/mastra/commit/d97107a7edc517ae8feddf914fc43cd80a66c0a8)]:
+  - @mastra/core@1.58.0-alpha.14
+  - @mastra/observability@1.16.6-alpha.4
+  - @mastra/mcp@1.16.0-alpha.2
+
+## 1.2.0-alpha.15
+
+### Patch Changes
+
+- Fixed authentication failures for unprefixed direct-provider models by using provider environment credentials when no provider-specific stored credential exists. ([#21197](https://github.com/mastra-ai/mastra/pull/21197))
+
+- Updated dependencies [[`9571e3a`](https://github.com/mastra-ai/mastra/commit/9571e3a06ed2c5220196460bf82a2129255c3a8b), [`d6c56f9`](https://github.com/mastra-ai/mastra/commit/d6c56f951db3213330b98b0abafa9778c8770e58), [`9571e3a`](https://github.com/mastra-ai/mastra/commit/9571e3a06ed2c5220196460bf82a2129255c3a8b), [`acc3513`](https://github.com/mastra-ai/mastra/commit/acc3513b19f79bf0a7ec2998694580edca54086c), [`94e7ae9`](https://github.com/mastra-ai/mastra/commit/94e7ae970b37c888cd1244ef013292639a2fe6d1), [`6a667b4`](https://github.com/mastra-ai/mastra/commit/6a667b4b7cd6a93fe41fcdd357b08c5a8c09b9ab), [`2440e09`](https://github.com/mastra-ai/mastra/commit/2440e096ea6c2def1ccc1eb2d0f3f5b88c4af940), [`a59049b`](https://github.com/mastra-ai/mastra/commit/a59049b1652a13efff66ac826326b5ed9a550342)]:
+  - @mastra/core@1.58.0-alpha.13
+
+## 1.2.0-alpha.14
+
+### Patch Changes
+
+- Fixed Factory interactive plans so they are stored as browsable artifacts. ([#21173](https://github.com/mastra-ai/mastra/pull/21173))
+
+## 1.2.0-alpha.13
+
+### Patch Changes
+
+- Fixed woken notifications running against the controller-level workspace, which is `undefined` for dynamic workspace factories. They now use the workspace of the session that owns the target thread. ([#21144](https://github.com/mastra-ai/mastra/pull/21144))
+
+- Updated dependencies [[`2e4624e`](https://github.com/mastra-ai/mastra/commit/2e4624edb6917e61249cb60ee377735e7af7e4a9), [`e6534fa`](https://github.com/mastra-ai/mastra/commit/e6534fab031216f6cb48c4c9907cbfdce9d60bc6), [`7fdcaa6`](https://github.com/mastra-ai/mastra/commit/7fdcaa66105d64290f9b14432a12ec99f39c4d3a), [`65b1183`](https://github.com/mastra-ai/mastra/commit/65b11832834f87a9bc8719391deb27559de5138a), [`5dba2a4`](https://github.com/mastra-ai/mastra/commit/5dba2a41600385751f5aace79878904e1972609d), [`cfd0d9e`](https://github.com/mastra-ai/mastra/commit/cfd0d9ec77ec3c69dd96f79cdb579e03d79f22ce), [`d9d93b2`](https://github.com/mastra-ai/mastra/commit/d9d93b25e4a65ad5fa153fa35be7ed149c8d587f)]:
+  - @mastra/core@1.58.0-alpha.12
+  - @mastra/observability@1.16.6-alpha.3
+  - @mastra/schema-compat@1.3.6-alpha.3
+  - @mastra/mcp@1.16.0-alpha.2
+  - @mastra/memory@1.26.1-alpha.6
+
+## 1.2.0-alpha.12
+
+### Minor Changes
+
+- Added `processors` and `signalProviders` to the Mastra Code plugin contract, so a plugin can contribute more than tools. ([#20848](https://github.com/mastra-ai/mastra/pull/20848))
+
+  **Processors**
+
+  A plugin can now extend the agent pipeline directly, passing a bare array for input processors or an object for both lanes. Plugin processors run after the processors Mastra Code configures and before the channel and memory layers the Agent appends. The slot isn't configurable, and the processors are resolved before every LLM call. Enabling, disabling, or updating a plugin applies on the next request instead of requiring a restart.
+
+  **Signal providers**
+
+  A plugin can also ship a signal provider, which monitors an external source and pushes notifications into a thread. Providers are long-lived, so the SDK owns their lifecycle instead of handing them to the agent: it registers Mastra on them, connects them to the coding agent, starts them polling, and stops them when the plugin is updated, disabled, or uninstalled. That makes a provider installed from a GitHub repository survive a mid-session update of that repository. Only one provider with a given id runs at a time, and a provider that fails to start is isolated from the rest of its plugin.
+
+  Field resolvers also receive `getController()` and `getActiveSession()` on the plugin context, so a plugin can read the running session lazily at the moment it needs it.
+
+  For embedders that inject their own `pluginManager` into `createMastraCode`: the `PluginManager` contract now also requires `onReload`, `getPluginSignalProviders`, and `setRuntime`, so hand-built manager implementations must add these methods.
+
+### Patch Changes
+
+- Fixed deferred notification deliveries failing with "No model selected" when the woken thread had no request context. The notification dispatch workflow re-sends deferred and summarized notifications long after the originating send, so any stream options attached to the original signal are gone; waking an idle thread then had no way to resolve a model. The notification delivery policy now drives the fix: `NotificationDeliveryDecision` accepts `streamOptions`, and the dispatcher re-runs the agent's delivery policy at delivery time (via the new `agent.resolveNotificationDeliveryDecision()`) to attach freshly resolved stream options to both individual and summary deliveries. Only `streamOptions` is honored at dispatch time; the record's persisted schedule still governs when and how it is delivered. Receipt-time sends also honor the policy's `streamOptions` now: an immediate deliver or summarize-now wake attaches them too, with caller-supplied stream options taking precedence. Mastra Code wires its session-based stream options resolver through the Code Agent's `deliveryPolicy.decide`, layered on the default decision logic. This fix applies to threads whose session is live in the current process; deliveries with no resolvable session fall back to a bare wake, and the "No model selected" error now distinguishes the case where a run started without any controller session context. This re-lands the capability removed by the #18637 revert in the policy-driven shape that revert called for, and the `@mastra/github-signals` bump only widens its `getNotificationStreamOptions` callback return type to allow `undefined`. ([#21113](https://github.com/mastra-ai/mastra/pull/21113))
+
+- Updated dependencies [[`b8ce7ec`](https://github.com/mastra-ai/mastra/commit/b8ce7ec96e39343c6c2f36d12d68a9ad816c09f7), [`a3a3624`](https://github.com/mastra-ai/mastra/commit/a3a3624f646b98e409424d8defccbd334da9e8b8), [`6246914`](https://github.com/mastra-ai/mastra/commit/62469146636911f3cbbe0880bd011c6a897a59a7), [`1315d8f`](https://github.com/mastra-ai/mastra/commit/1315d8f17e8e7acb61cca46b72a1d42f6d00d289), [`3f73c07`](https://github.com/mastra-ai/mastra/commit/3f73c076727e8c36b4fff7a1b40290fb68957fa8), [`7c1ebb1`](https://github.com/mastra-ai/mastra/commit/7c1ebb15690c4b3f0eabb19077cf8af573311e57), [`32980a3`](https://github.com/mastra-ai/mastra/commit/32980a3e2413d0274ac244d32c37d910edc13f00), [`261edb9`](https://github.com/mastra-ai/mastra/commit/261edb9bc0cfbed2b77090b87562307a360f1a04), [`4bcdfaf`](https://github.com/mastra-ai/mastra/commit/4bcdfaf0eac3199d7cb171b0a19a92c9c341eea4), [`1b1dd7b`](https://github.com/mastra-ai/mastra/commit/1b1dd7bc0e59b7a8bfabd09a3eec1ccd95b4c2f3), [`af4636a`](https://github.com/mastra-ai/mastra/commit/af4636a74463275d71c1d13a38f7d2b738f128bf), [`a463cdf`](https://github.com/mastra-ai/mastra/commit/a463cdf1c95c3059e70f0bff27959e8558bb899d), [`0ea6b80`](https://github.com/mastra-ai/mastra/commit/0ea6b8001408ce02b56e8be0536b0fd8cbaf8ad2)]:
+  - @mastra/core@1.58.0-alpha.11
+  - @mastra/github-signals@0.2.5-alpha.1
+  - @mastra/memory@1.26.1-alpha.5
+  - @mastra/mcp@1.16.0-alpha.2
+
+## 1.2.0-alpha.11
+
+### Patch Changes
+
+- Updated dependencies [[`66bbfb5`](https://github.com/mastra-ai/mastra/commit/66bbfb5f05b473d39f88c0e4a481ccac41634f3a)]:
+  - @mastra/core@1.58.0-alpha.10
+
+## 1.2.0-alpha.10
+
+### Patch Changes
+
+- Include worktree identity (`worktree_path`, `branch`, `main_repo_path`) in `SessionStart` and `SessionEnd` hook payloads when a session runs in a git worktree, so hooks can provision and tear down per-worktree resources. ([#21034](https://github.com/mastra-ai/mastra/pull/21034))
+
+- Updated dependencies [[`86b7b77`](https://github.com/mastra-ai/mastra/commit/86b7b777980d30f66e1fd134a37d2af4c22e54cc), [`80a3324`](https://github.com/mastra-ai/mastra/commit/80a33245d3110204de6f56d61211523ffe338692), [`d9d2881`](https://github.com/mastra-ai/mastra/commit/d9d2881ede6dd6c023d144215fc812062aed0890), [`82e3365`](https://github.com/mastra-ai/mastra/commit/82e3365ef7c9bf7bee2e7a7029035ea262d68895), [`1b482c2`](https://github.com/mastra-ai/mastra/commit/1b482c2d89244dd758c41e5f927a2b44041388d2), [`e6a2860`](https://github.com/mastra-ai/mastra/commit/e6a2860649cc51f87d32d78b766ae2126446ba07), [`7bd85ea`](https://github.com/mastra-ai/mastra/commit/7bd85ea7588b71c25ce9f4019c88f8539be5dcbc)]:
+  - @mastra/core@1.58.0-alpha.9
+
+## 1.2.0-alpha.9
+
+### Patch Changes
+
+- Updated dependencies [[`1c75e32`](https://github.com/mastra-ai/mastra/commit/1c75e32f7fc0b9fb6f548b4407feaec8a1440212), [`c47165c`](https://github.com/mastra-ai/mastra/commit/c47165c983c87594c6952f1fd2fa51a90205034c), [`e08e789`](https://github.com/mastra-ai/mastra/commit/e08e789c1bf4cd2fe46363f7a4728536ceccc9bd), [`35cc901`](https://github.com/mastra-ai/mastra/commit/35cc90102cf834a84827acaf9eee0b6d6d1e2a3b), [`a8b4cf0`](https://github.com/mastra-ai/mastra/commit/a8b4cf02823cffebc4751a53337dfacf097c1ae1), [`a53f19b`](https://github.com/mastra-ai/mastra/commit/a53f19b9badb8bd349fe2f885814e9db6b7d3c4b), [`f33264f`](https://github.com/mastra-ai/mastra/commit/f33264f517ae603279afd5c4251e2b40f6dd3618), [`689f2c4`](https://github.com/mastra-ai/mastra/commit/689f2c4b6c0835fe455702b01d21daa8abcd9331), [`eeae63e`](https://github.com/mastra-ai/mastra/commit/eeae63e7fbe8e1f237adc69bca6e2ac13c5ca907), [`4c186a0`](https://github.com/mastra-ai/mastra/commit/4c186a017275f45e6ed4c09de0f89550e2d09e8c), [`b0fa077`](https://github.com/mastra-ai/mastra/commit/b0fa077bcbc9b08551846fe372a0d3d15b71ed72)]:
+  - @mastra/core@1.58.0-alpha.8
+  - @mastra/memory@1.26.1-alpha.4
+  - @mastra/libsql@1.20.0-alpha.2
+  - @mastra/pg@1.20.0-alpha.3
+
 ## 1.2.0-alpha.8
 
 ### Patch Changes

package/README.md

@@ -57,6 +57,28 @@ const prepared = await prepareAgentControllerMount({
 
 Configured processors run before Mastra Code's built-in input processors. Keep processor instances stateless because the mounted agent shares them across sessions and runs.
 
+## Dynamic workflows
+
+The local controller registers the Workflow Builder before workers start. In build mode, users can ask the code agent to create a workflow in natural language. The builder discovers registered agents, tools, and workflows, validates a complete definition, then persists and registers it immediately.
+
+Use the workflow service to manage saved workflows from a custom SDK surface:
+
+```ts
+import { deleteWorkflow, getWorkflow, listWorkflows, runWorkflow } from '@mastra/code-sdk/workflows/service';
+
+const { workflows } = await listWorkflows(mastra);
+const firstWorkflow = workflows[0];
+if (!firstWorkflow) throw new Error('No Dynamic Workflows are available.');
+
+const definition = await getWorkflow(mastra, firstWorkflow.id);
+if (!definition) throw new Error(`Workflow "${firstWorkflow.id}" was not found.`);
+
+const result = await runWorkflow(mastra, definition.id, { topic: 'dynamic workflows' });
+await deleteWorkflow(mastra, definition.id);
+```
+
+Pass the session request context to `runWorkflow` when workflow agent steps need the session-selected model. You can also pass an event callback as the fifth argument to render workflow step progress.
+
 Deep modules are available as subpath imports, e.g.:
 
 ```ts

package/dist/agents/model.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../src/agents/model.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,+BAA+B,CAAC;AACnF,OAAO,KAAK,EAAE,oBAAoB,EAAE,2BAA2B,EAAE,MAAM,kBAAkB,CAAC;AAC1F,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,8BAA8B,CAAC;AAOnE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,8BAA8B,CAAC;AAGlE,OAAO,EAGL,iBAAiB,EAGlB,MAAM,yBAAyB,CAAC;AACjC,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AAExE,OAAO,EACL,kBAAkB,EAClB,eAAe,EACf,qBAAqB,EACrB,iBAAiB,EACjB,6BAA6B,EAC7B,WAAW,GACZ,MAAM,yBAAyB,CAAC;AACjC,YAAY,EAAE,wBAAwB,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AAClG,OAAO,EACL,0BAA0B,EAC1B,0BAA0B,EAC1B,+BAA+B,GAChC,MAAM,0BAA0B,CAAC;AAClC,YAAY,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,0BAA0B,CAAC;AAC1F,OAAO,EACL,wBAAwB,EACxB,wBAAwB,EACxB,sBAAsB,GACvB,MAAM,6BAA6B,CAAC;AACrC,YAAY,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAEzE,KAAK,aAAa,GAAG,oBAAoB,CAAC;AAa1C,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,wBAAwB,GAAG,iBAAiB,CAE5F;AAED,wBAAgB,oCAAoC,CAAC,OAAO,EAAE,2BAA2B,6DAIxF;AAED;;;;GAIG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAEtD;AAED;;;;;;;;GAQG;AACH,wBAAgB,YAAY,CAC1B,OAAO,EAAE,MAAM,EACf,OAAO,CAAC,EAAE;IAAE,aAAa,CAAC,EAAE,aAAa,CAAC;IAAC,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAAC,cAAc,CAAC,EAAE,cAAc,CAAA;CAAE,GACzG,oBAAoB,CA+EtB;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,2BAA2B,CACzC,sBAAsB,EAAE,6BAA6B,CAAC,GAAG,CAAC,GAAG,SAAS,EACtE,YAAY,CAAC,EAAE,MAAM,GACpB,aAAa,CAKf;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,EAAE,cAAc,EAAE,EAAE;IAAE,cAAc,EAAE,cAAc,CAAA;CAAE,EACtD,YAAY,CAAC,EAAE,MAAM,GACpB,aAAa,CAWf;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,iBAAiB,CAC/B,EAAE,cAAc,EAAE,EAAE;IAAE,cAAc,EAAE,cAAc,CAAA;CAAE,EACtD,YAAY,CAAC,EAAE,MAAM,GACpB,aAAa,GAAG,SAAS,CAI3B"}
+{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../src/agents/model.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,+BAA+B,CAAC;AACnF,OAAO,KAAK,EAAE,oBAAoB,EAAE,2BAA2B,EAAE,MAAM,kBAAkB,CAAC;AAC1F,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,8BAA8B,CAAC;AAOnE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,8BAA8B,CAAC;AAGlE,OAAO,EAGL,iBAAiB,EAGlB,MAAM,yBAAyB,CAAC;AACjC,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AAExE,OAAO,EACL,kBAAkB,EAClB,eAAe,EACf,qBAAqB,EACrB,iBAAiB,EACjB,6BAA6B,EAC7B,WAAW,GACZ,MAAM,yBAAyB,CAAC;AACjC,YAAY,EAAE,wBAAwB,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AAClG,OAAO,EACL,0BAA0B,EAC1B,0BAA0B,EAC1B,+BAA+B,GAChC,MAAM,0BAA0B,CAAC;AAClC,YAAY,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,0BAA0B,CAAC;AAC1F,OAAO,EACL,wBAAwB,EACxB,wBAAwB,EACxB,sBAAsB,GACvB,MAAM,6BAA6B,CAAC;AACrC,YAAY,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAEzE,KAAK,aAAa,GAAG,oBAAoB,CAAC;AAa1C,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,wBAAwB,GAAG,iBAAiB,CAE5F;AAED,wBAAgB,oCAAoC,CAAC,OAAO,EAAE,2BAA2B,6DAIxF;AAED;;;;GAIG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAEtD;AAED;;;;;;;;GAQG;AACH,wBAAgB,YAAY,CAC1B,OAAO,EAAE,MAAM,EACf,OAAO,CAAC,EAAE;IAAE,aAAa,CAAC,EAAE,aAAa,CAAC;IAAC,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAAC,cAAc,CAAC,EAAE,cAAc,CAAA;CAAE,GACzG,oBAAoB,CA+EtB;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,2BAA2B,CACzC,sBAAsB,EAAE,6BAA6B,CAAC,GAAG,CAAC,GAAG,SAAS,EACtE,YAAY,CAAC,EAAE,MAAM,GACpB,aAAa,CAKf;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,EAAE,cAAc,EAAE,EAAE;IAAE,cAAc,EAAE,cAAc,CAAA;CAAE,EACtD,YAAY,CAAC,EAAE,MAAM,GACpB,aAAa,CAoBf;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,iBAAiB,CAC/B,EAAE,cAAc,EAAE,EAAE;IAAE,cAAc,EAAE,cAAc,CAAA;CAAE,EACtD,YAAY,CAAC,EAAE,MAAM,GACpB,aAAa,GAAG,SAAS,CAI3B"}

package/dist/agents/model.js

@@ -85,7 +85,7 @@ function resolveModel(modelId, options) {
 	return gateway.resolveLanguageModel({
 		providerId,
 		modelId: bareModelId,
-		apiKey: auth?.apiKey ?? mgApiKey ?? "",
+		apiKey: auth?.apiKey ?? "",
 		headers
 	});
 }
@@ -115,7 +115,10 @@ function resolveRequestThinkingLevel(agentControllerContext, settingsPath) {
 function getDynamicModel({ requestContext }, settingsPath) {
 	const agentControllerContext = requestContext.get("controller");
 	const modelId = agentControllerContext?.session?.modelId;
-	if (!modelId) throw new Error("No model selected. Use /models to select a model first.");
+	if (!modelId) {
+		if (!agentControllerContext) throw new Error("No model available: this run started without a controller session context, so no model selection could be resolved.");
+		throw new Error("No model selected. Use /models to select a model first.");
+	}
 	return resolveModel(modelId, {
 		thinkingLevel: resolveRequestThinkingLevel(agentControllerContext, settingsPath),
 		remapForCodexOAuth: true,

package/dist/agents/model.js.map

@@ -1 +1 @@
-{"version":3,"file":"model.js","names":[],"sources":["../../src/agents/model.ts"],"sourcesContent":["import type { AgentControllerRequestContext } from '@mastra/core/agent-controller';\nimport type { GatewayLanguageModel, MastraModelGatewayInterface } from '@mastra/core/llm';\nimport type { RequestContext } from '@mastra/core/request-context';\nimport {\n  loadSettings,\n  resolveDefaultThinkingLevel,\n  stripMastraCodeCustomProviderPrefix,\n} from '../onboarding/settings.js';\nimport { AMAZON_BEDROCK_GATEWAY_ID, createAmazonBedrockGateway } from '../providers/amazon-bedrock-gateway.js';\nimport type { ThinkingLevel } from '../providers/openai-codex.js';\nimport { resolveCredentialStore } from './credential-resolver.js';\nimport { resolveCustomProviders } from './custom-provider-source.js';\nimport {\n  MASTRA_GATEWAY_PREFIX,\n  MASTRACODE_GATEWAY_ID,\n  MastraCodeGateway,\n  reloadAuthStorage,\n  stripMastraGatewayPrefix,\n} from './mastracode-gateway.js';\nimport type { MastraCodeGatewayOptions } from './mastracode-gateway.js';\n\nexport {\n  getAnthropicApiKey,\n  getOpenAIApiKey,\n  MASTRACODE_GATEWAY_ID,\n  MastraCodeGateway,\n  remapOpenAIModelForCodexOAuth,\n  resolveAuth,\n} from './mastracode-gateway.js';\nexport type { MastraCodeCustomProvider, MastraCodeGatewayOptions } from './mastracode-gateway.js';\nexport {\n  setCredentialStoreProvider,\n  hasCredentialStoreProvider,\n  resolveTenantFromRequestContext,\n} from './credential-resolver.js';\nexport type { CredentialTenant, CredentialStoreProvider } from './credential-resolver.js';\nexport {\n  setCustomProvidersSource,\n  hasCustomProvidersSource,\n  resolveCustomProviders,\n} from './custom-provider-source.js';\nexport type { CustomProvidersSource } from './custom-provider-source.js';\n\ntype ResolvedModel = GatewayLanguageModel;\ntype ModelRequestHeaders = Record<string, string>;\n\nfunction getAgentControllerHeaders(requestContext?: RequestContext): ModelRequestHeaders | undefined {\n  const agentControllerContext = requestContext?.get('controller') as AgentControllerRequestContext<any> | undefined;\n  const headers = {\n    ...(agentControllerContext?.threadId ? { 'x-thread-id': agentControllerContext.threadId } : {}),\n    ...(agentControllerContext?.resourceId ? { 'x-resource-id': agentControllerContext.resourceId } : {}),\n  };\n\n  return Object.keys(headers).length > 0 ? headers : undefined;\n}\n\nexport function createMastraCodeGateway(options: MastraCodeGatewayOptions): MastraCodeGateway {\n  return new MastraCodeGateway(options);\n}\n\nexport function createMastraCodeModelCatalogProvider(gateway: MastraModelGatewayInterface) {\n  return gateway instanceof MastraCodeGateway\n    ? gateway.createModelCatalogProvider()\n    : MastraCodeGateway.createModelCatalogProvider(gateway);\n}\n\n/**\n * Placeholder for future model ID normalization.\n * Currently returns the input unchanged, but exists as a seam\n * for aliasing, casing fixes, or validation in the future.\n */\nexport function resolveModelId(modelId: string): string {\n  return modelId;\n}\n\n/**\n * Resolve a model ID to the correct provider instance.\n * Shared by the main agent, observer, and reflector.\n *\n * - For anthropic/* models: Uses stored OAuth credentials when present, otherwise direct API key\n * - For openai/* models: Uses OAuth when configured, otherwise direct API key from AuthStorage\n * - For moonshotai/* models: Uses Moonshot AI Anthropic-compatible endpoint\n * - For all other providers: Uses Mastra's model router (models.dev gateway)\n */\nexport function resolveModel(\n  modelId: string,\n  options?: { thinkingLevel?: ThinkingLevel; remapForCodexOAuth?: boolean; requestContext?: RequestContext },\n): GatewayLanguageModel {\n  reloadAuthStorage();\n  const headers = getAgentControllerHeaders(options?.requestContext);\n  const settings = loadSettings();\n  // Bedrock was previously cataloged under the MastraCode gateway namespace\n  // (`mastracode/amazon-bedrock/<model>`). Normalize any legacy saved ids to the\n  // standalone `amazon-bedrock/<model>` form so they resolve through the\n  // dedicated Bedrock gateway.\n  const bedrockLegacyPrefix = `${MASTRACODE_GATEWAY_ID}/amazon-bedrock/`;\n  const bedrockNormalizedInput = modelId.startsWith(bedrockLegacyPrefix)\n    ? modelId.slice(MASTRACODE_GATEWAY_ID.length + 1)\n    : modelId;\n  // Deployed web registers a custom providers source (DB-backed, tenant\n  // scoped); when registered it is authoritative and settings.json custom\n  // providers are ignored. Undefined = local settings-based behavior.\n  const customProviders = resolveCustomProviders(options?.requestContext) ?? settings.customProviders;\n  // Ids selected from the shared /models catalog were previously persisted in\n  // the gateway-qualified `mastracode/<customProviderId>/<model>` form, which\n  // parses the provider as `mastracode` and breaks provider config lookup.\n  // Normalize at resolution time (in addition to stripping at selection time)\n  // so already-saved ids and any surface that persists the raw catalog id\n  // still resolve to the custom provider.\n  const normalizedInput = stripMastraCodeCustomProviderPrefix(bedrockNormalizedInput, customProviders);\n  const isMastraGatewayModel = normalizedInput.startsWith(MASTRA_GATEWAY_PREFIX);\n  const normalizedModelId = stripMastraGatewayPrefix(normalizedInput);\n  const [providerId, ...modelParts] = normalizedModelId.split('/');\n  const bareModelId = modelParts.join('/');\n  if (!providerId || !bareModelId) {\n    throw new Error(`Invalid model id: ${modelId}`);\n  }\n\n  if (providerId === AMAZON_BEDROCK_GATEWAY_ID) {\n    const bedrockGateway = createAmazonBedrockGateway();\n    const routerId = `${AMAZON_BEDROCK_GATEWAY_ID}/${bareModelId}`;\n    const auth = bedrockGateway.resolveAuth({\n      gatewayId: AMAZON_BEDROCK_GATEWAY_ID,\n      providerId: AMAZON_BEDROCK_GATEWAY_ID,\n      modelId: bareModelId,\n      routerId,\n    });\n    return bedrockGateway.resolveLanguageModel({\n      providerId: AMAZON_BEDROCK_GATEWAY_ID,\n      modelId: bareModelId,\n      apiKey: auth?.apiKey ?? '',\n      headers,\n    });\n  }\n\n  const routerId = `${MASTRACODE_GATEWAY_ID}/${normalizedModelId}`;\n\n  const mgApiKey = MastraCodeGateway.getMastraGatewayApiKey();\n  const rawGatewayBase =\n    settings.memoryGateway?.baseUrl ?? process.env['MASTRA_GATEWAY_URL'] ?? 'https://gateway-api.mastra.ai';\n  // Deployed web registers a per-tenant credential store provider; when the\n  // request carries an authenticated tenant, resolve credentials through the\n  // caller's own store (user > org > env). Undefined = global AuthStorage.\n  const credentialStore = resolveCredentialStore(options?.requestContext);\n  const gateway = createMastraCodeGateway({\n    mastraGatewayBaseUrl: rawGatewayBase.replace(/\\/+$/, '').replace(/\\/v1$/, ''),\n    mastraGatewayApiKey: mgApiKey,\n    routeThroughMastraGateway: Boolean(mgApiKey && isMastraGatewayModel),\n    thinkingLevel: options?.thinkingLevel,\n    customProviders,\n    credentialStore,\n  });\n\n  const auth = gateway.resolveAuth({\n    gatewayId: MASTRACODE_GATEWAY_ID,\n    providerId,\n    modelId: bareModelId,\n    routerId,\n  });\n\n  return gateway.resolveLanguageModel({\n    providerId,\n    modelId: bareModelId,\n    apiKey: auth?.apiKey ?? mgApiKey ?? '',\n    headers,\n  });\n}\n\n/**\n * Resolve the effective thinking level for the current request.\n *\n * Precedence:\n *   1. Session override (`state.thinkingLevel`, set via /think or the session\n *      settings panel).\n *   2. Per-mode default from settings (`models.modeThinkingDefaults[mode]`).\n *   3. Global default (`preferences.thinkingLevel`).\n *\n * Resolved per-request (not seeded at session start) so configuration changes\n * apply to the next request of every session — including automated\n * (rule-driven) Factory runs that nobody ever opens interactively.\n */\nexport function resolveRequestThinkingLevel(\n  agentControllerContext: AgentControllerRequestContext<any> | undefined,\n  settingsPath?: string,\n): ThinkingLevel {\n  const override = agentControllerContext?.state?.thinkingLevel as ThinkingLevel | undefined;\n  if (override !== undefined) return override;\n  const modeId = agentControllerContext?.session?.modeId;\n  return resolveDefaultThinkingLevel(loadSettings(settingsPath), modeId).level;\n}\n\n/**\n * Dynamic model function that reads the current model from controller state.\n * This allows runtime model switching via the /models picker.\n */\nexport function getDynamicModel(\n  { requestContext }: { requestContext: RequestContext },\n  settingsPath?: string,\n): ResolvedModel {\n  const agentControllerContext = requestContext.get('controller') as AgentControllerRequestContext<any> | undefined;\n\n  const modelId = agentControllerContext?.session?.modelId;\n  if (!modelId) {\n    throw new Error('No model selected. Use /models to select a model first.');\n  }\n\n  const thinkingLevel = resolveRequestThinkingLevel(agentControllerContext, settingsPath);\n\n  return resolveModel(modelId, { thinkingLevel, remapForCodexOAuth: true, requestContext });\n}\n\n/**\n * Goal judge model resolver for the agent's `goal.judge` config. Resolves the\n * configured goal judge model through mastracode's gateway so provider\n * credentials (stored in auth storage, not just env) are injected — a bare model\n * id handed to core's default model router would fail to find the API key.\n *\n * Returns `undefined` when no judge model is configured, which keeps the goal\n * step a complete no-op (the goal mechanism requires a judge to do anything).\n *\n * `settingsPath` must be the same source `createMastraCode()` reads from so the\n * judge model and the goal budget (`goalMaxTurns`) come from one config — with a\n * custom `settingsPath` a bare `loadSettings()` here could read a different file\n * and silently turn the goal step into a no-op.\n */\nexport function getGoalJudgeModel(\n  { requestContext }: { requestContext: RequestContext },\n  settingsPath?: string,\n): ResolvedModel | undefined {\n  const judgeModelId = loadSettings(settingsPath).models.goalJudgeModel;\n  if (!judgeModelId) return undefined;\n  return resolveModel(judgeModelId, { remapForCodexOAuth: true, requestContext });\n}\n"],"mappings":";;;;;;AA8CA,SAAS,0BAA0B,gBAAkE;CACnG,MAAM,yBAAyB,gBAAgB,IAAI,YAAY;CAC/D,MAAM,UAAU;EACd,GAAI,wBAAwB,WAAW,EAAE,eAAe,uBAAuB,SAAS,IAAI,CAAC;EAC7F,GAAI,wBAAwB,aAAa,EAAE,iBAAiB,uBAAuB,WAAW,IAAI,CAAC;CACrG;CAEA,OAAO,OAAO,KAAK,OAAO,CAAC,CAAC,SAAS,IAAI,UAAU,KAAA;AACrD;AAEA,SAAgB,wBAAwB,SAAsD;CAC5F,OAAO,IAAI,kBAAkB,OAAO;AACtC;AAEA,SAAgB,qCAAqC,SAAsC;CACzF,OAAO,mBAAmB,oBACtB,QAAQ,2BAA2B,IACnC,kBAAkB,2BAA2B,OAAO;AAC1D;;;;;;AAOA,SAAgB,eAAe,SAAyB;CACtD,OAAO;AACT;;;;;;;;;;AAWA,SAAgB,aACd,SACA,SACsB;CACtB,kBAAkB;CAClB,MAAM,UAAU,0BAA0B,SAAS,cAAc;CACjE,MAAM,WAAW,aAAa;CAK9B,MAAM,sBAAsB,GAAG,sBAAsB;CACrD,MAAM,yBAAyB,QAAQ,WAAW,mBAAmB,IACjE,QAAQ,MAAM,sBAAsB,SAAS,CAAC,IAC9C;CAIJ,MAAM,kBAAkB,uBAAuB,SAAS,cAAc,KAAK,SAAS;CAOpF,MAAM,kBAAkB,oCAAoC,wBAAwB,eAAe;CACnG,MAAM,uBAAuB,gBAAgB,WAAW,qBAAqB;CAC7E,MAAM,oBAAoB,yBAAyB,eAAe;CAClE,MAAM,CAAC,YAAY,GAAG,cAAc,kBAAkB,MAAM,GAAG;CAC/D,MAAM,cAAc,WAAW,KAAK,GAAG;CACvC,IAAI,CAAC,cAAc,CAAC,aAClB,MAAM,IAAI,MAAM,qBAAqB,SAAS;CAGhD,IAAI,eAAA,kBAA0C;EAC5C,MAAM,iBAAiB,2BAA2B;EAClD,MAAM,WAAW,GAAG,0BAA0B,GAAG;EACjD,MAAM,OAAO,eAAe,YAAY;GACtC,WAAW;GACX,YAAY;GACZ,SAAS;GACT;EACF,CAAC;EACD,OAAO,eAAe,qBAAqB;GACzC,YAAY;GACZ,SAAS;GACT,QAAQ,MAAM,UAAU;GACxB;EACF,CAAC;CACH;CAEA,MAAM,WAAW,GAAG,sBAAsB,GAAG;CAE7C,MAAM,WAAW,kBAAkB,uBAAuB;CAC1D,MAAM,iBACJ,SAAS,eAAe,WAAW,QAAQ,IAAI,yBAAyB;CAI1E,MAAM,kBAAkB,uBAAuB,SAAS,cAAc;CACtE,MAAM,UAAU,wBAAwB;EACtC,sBAAsB,eAAe,QAAQ,QAAQ,EAAE,CAAC,CAAC,QAAQ,SAAS,EAAE;EAC5E,qBAAqB;EACrB,2BAA2B,QAAQ,YAAY,oBAAoB;EACnE,eAAe,SAAS;EACxB;EACA;CACF,CAAC;CAED,MAAM,OAAO,QAAQ,YAAY;EAC/B,WAAW;EACX;EACA,SAAS;EACT;CACF,CAAC;CAED,OAAO,QAAQ,qBAAqB;EAClC;EACA,SAAS;EACT,QAAQ,MAAM,UAAU,YAAY;EACpC;CACF,CAAC;AACH;;;;;;;;;;;;;;AAeA,SAAgB,4BACd,wBACA,cACe;CACf,MAAM,WAAW,wBAAwB,OAAO;CAChD,IAAI,aAAa,KAAA,GAAW,OAAO;CACnC,MAAM,SAAS,wBAAwB,SAAS;CAChD,OAAO,4BAA4B,aAAa,YAAY,GAAG,MAAM,CAAC,CAAC;AACzE;;;;;AAMA,SAAgB,gBACd,EAAE,kBACF,cACe;CACf,MAAM,yBAAyB,eAAe,IAAI,YAAY;CAE9D,MAAM,UAAU,wBAAwB,SAAS;CACjD,IAAI,CAAC,SACH,MAAM,IAAI,MAAM,yDAAyD;CAK3E,OAAO,aAAa,SAAS;EAAE,eAFT,4BAA4B,wBAAwB,YAE/B;EAAG,oBAAoB;EAAM;CAAe,CAAC;AAC1F;;;;;;;;;;;;;;;AAgBA,SAAgB,kBACd,EAAE,kBACF,cAC2B;CAC3B,MAAM,eAAe,aAAa,YAAY,CAAC,CAAC,OAAO;CACvD,IAAI,CAAC,cAAc,OAAO,KAAA;CAC1B,OAAO,aAAa,cAAc;EAAE,oBAAoB;EAAM;CAAe,CAAC;AAChF"}
+{"version":3,"file":"model.js","names":[],"sources":["../../src/agents/model.ts"],"sourcesContent":["import type { AgentControllerRequestContext } from '@mastra/core/agent-controller';\nimport type { GatewayLanguageModel, MastraModelGatewayInterface } from '@mastra/core/llm';\nimport type { RequestContext } from '@mastra/core/request-context';\nimport {\n  loadSettings,\n  resolveDefaultThinkingLevel,\n  stripMastraCodeCustomProviderPrefix,\n} from '../onboarding/settings.js';\nimport { AMAZON_BEDROCK_GATEWAY_ID, createAmazonBedrockGateway } from '../providers/amazon-bedrock-gateway.js';\nimport type { ThinkingLevel } from '../providers/openai-codex.js';\nimport { resolveCredentialStore } from './credential-resolver.js';\nimport { resolveCustomProviders } from './custom-provider-source.js';\nimport {\n  MASTRA_GATEWAY_PREFIX,\n  MASTRACODE_GATEWAY_ID,\n  MastraCodeGateway,\n  reloadAuthStorage,\n  stripMastraGatewayPrefix,\n} from './mastracode-gateway.js';\nimport type { MastraCodeGatewayOptions } from './mastracode-gateway.js';\n\nexport {\n  getAnthropicApiKey,\n  getOpenAIApiKey,\n  MASTRACODE_GATEWAY_ID,\n  MastraCodeGateway,\n  remapOpenAIModelForCodexOAuth,\n  resolveAuth,\n} from './mastracode-gateway.js';\nexport type { MastraCodeCustomProvider, MastraCodeGatewayOptions } from './mastracode-gateway.js';\nexport {\n  setCredentialStoreProvider,\n  hasCredentialStoreProvider,\n  resolveTenantFromRequestContext,\n} from './credential-resolver.js';\nexport type { CredentialTenant, CredentialStoreProvider } from './credential-resolver.js';\nexport {\n  setCustomProvidersSource,\n  hasCustomProvidersSource,\n  resolveCustomProviders,\n} from './custom-provider-source.js';\nexport type { CustomProvidersSource } from './custom-provider-source.js';\n\ntype ResolvedModel = GatewayLanguageModel;\ntype ModelRequestHeaders = Record<string, string>;\n\nfunction getAgentControllerHeaders(requestContext?: RequestContext): ModelRequestHeaders | undefined {\n  const agentControllerContext = requestContext?.get('controller') as AgentControllerRequestContext<any> | undefined;\n  const headers = {\n    ...(agentControllerContext?.threadId ? { 'x-thread-id': agentControllerContext.threadId } : {}),\n    ...(agentControllerContext?.resourceId ? { 'x-resource-id': agentControllerContext.resourceId } : {}),\n  };\n\n  return Object.keys(headers).length > 0 ? headers : undefined;\n}\n\nexport function createMastraCodeGateway(options: MastraCodeGatewayOptions): MastraCodeGateway {\n  return new MastraCodeGateway(options);\n}\n\nexport function createMastraCodeModelCatalogProvider(gateway: MastraModelGatewayInterface) {\n  return gateway instanceof MastraCodeGateway\n    ? gateway.createModelCatalogProvider()\n    : MastraCodeGateway.createModelCatalogProvider(gateway);\n}\n\n/**\n * Placeholder for future model ID normalization.\n * Currently returns the input unchanged, but exists as a seam\n * for aliasing, casing fixes, or validation in the future.\n */\nexport function resolveModelId(modelId: string): string {\n  return modelId;\n}\n\n/**\n * Resolve a model ID to the correct provider instance.\n * Shared by the main agent, observer, and reflector.\n *\n * - For anthropic/* models: Uses stored OAuth credentials when present, otherwise direct API key\n * - For openai/* models: Uses OAuth when configured, otherwise direct API key from AuthStorage\n * - For moonshotai/* models: Uses Moonshot AI Anthropic-compatible endpoint\n * - For all other providers: Uses Mastra's model router (models.dev gateway)\n */\nexport function resolveModel(\n  modelId: string,\n  options?: { thinkingLevel?: ThinkingLevel; remapForCodexOAuth?: boolean; requestContext?: RequestContext },\n): GatewayLanguageModel {\n  reloadAuthStorage();\n  const headers = getAgentControllerHeaders(options?.requestContext);\n  const settings = loadSettings();\n  // Bedrock was previously cataloged under the MastraCode gateway namespace\n  // (`mastracode/amazon-bedrock/<model>`). Normalize any legacy saved ids to the\n  // standalone `amazon-bedrock/<model>` form so they resolve through the\n  // dedicated Bedrock gateway.\n  const bedrockLegacyPrefix = `${MASTRACODE_GATEWAY_ID}/amazon-bedrock/`;\n  const bedrockNormalizedInput = modelId.startsWith(bedrockLegacyPrefix)\n    ? modelId.slice(MASTRACODE_GATEWAY_ID.length + 1)\n    : modelId;\n  // Deployed web registers a custom providers source (DB-backed, tenant\n  // scoped); when registered it is authoritative and settings.json custom\n  // providers are ignored. Undefined = local settings-based behavior.\n  const customProviders = resolveCustomProviders(options?.requestContext) ?? settings.customProviders;\n  // Ids selected from the shared /models catalog were previously persisted in\n  // the gateway-qualified `mastracode/<customProviderId>/<model>` form, which\n  // parses the provider as `mastracode` and breaks provider config lookup.\n  // Normalize at resolution time (in addition to stripping at selection time)\n  // so already-saved ids and any surface that persists the raw catalog id\n  // still resolve to the custom provider.\n  const normalizedInput = stripMastraCodeCustomProviderPrefix(bedrockNormalizedInput, customProviders);\n  const isMastraGatewayModel = normalizedInput.startsWith(MASTRA_GATEWAY_PREFIX);\n  const normalizedModelId = stripMastraGatewayPrefix(normalizedInput);\n  const [providerId, ...modelParts] = normalizedModelId.split('/');\n  const bareModelId = modelParts.join('/');\n  if (!providerId || !bareModelId) {\n    throw new Error(`Invalid model id: ${modelId}`);\n  }\n\n  if (providerId === AMAZON_BEDROCK_GATEWAY_ID) {\n    const bedrockGateway = createAmazonBedrockGateway();\n    const routerId = `${AMAZON_BEDROCK_GATEWAY_ID}/${bareModelId}`;\n    const auth = bedrockGateway.resolveAuth({\n      gatewayId: AMAZON_BEDROCK_GATEWAY_ID,\n      providerId: AMAZON_BEDROCK_GATEWAY_ID,\n      modelId: bareModelId,\n      routerId,\n    });\n    return bedrockGateway.resolveLanguageModel({\n      providerId: AMAZON_BEDROCK_GATEWAY_ID,\n      modelId: bareModelId,\n      apiKey: auth?.apiKey ?? '',\n      headers,\n    });\n  }\n\n  const routerId = `${MASTRACODE_GATEWAY_ID}/${normalizedModelId}`;\n\n  const mgApiKey = MastraCodeGateway.getMastraGatewayApiKey();\n  const rawGatewayBase =\n    settings.memoryGateway?.baseUrl ?? process.env['MASTRA_GATEWAY_URL'] ?? 'https://gateway-api.mastra.ai';\n  // Deployed web registers a per-tenant credential store provider; when the\n  // request carries an authenticated tenant, resolve credentials through the\n  // caller's own store (user > org > env). Undefined = global AuthStorage.\n  const credentialStore = resolveCredentialStore(options?.requestContext);\n  const gateway = createMastraCodeGateway({\n    mastraGatewayBaseUrl: rawGatewayBase.replace(/\\/+$/, '').replace(/\\/v1$/, ''),\n    mastraGatewayApiKey: mgApiKey,\n    routeThroughMastraGateway: Boolean(mgApiKey && isMastraGatewayModel),\n    thinkingLevel: options?.thinkingLevel,\n    customProviders,\n    credentialStore,\n  });\n\n  const auth = gateway.resolveAuth({\n    gatewayId: MASTRACODE_GATEWAY_ID,\n    providerId,\n    modelId: bareModelId,\n    routerId,\n  });\n\n  return gateway.resolveLanguageModel({\n    providerId,\n    modelId: bareModelId,\n    apiKey: auth?.apiKey ?? '',\n    headers,\n  });\n}\n\n/**\n * Resolve the effective thinking level for the current request.\n *\n * Precedence:\n *   1. Session override (`state.thinkingLevel`, set via /think or the session\n *      settings panel).\n *   2. Per-mode default from settings (`models.modeThinkingDefaults[mode]`).\n *   3. Global default (`preferences.thinkingLevel`).\n *\n * Resolved per-request (not seeded at session start) so configuration changes\n * apply to the next request of every session — including automated\n * (rule-driven) Factory runs that nobody ever opens interactively.\n */\nexport function resolveRequestThinkingLevel(\n  agentControllerContext: AgentControllerRequestContext<any> | undefined,\n  settingsPath?: string,\n): ThinkingLevel {\n  const override = agentControllerContext?.state?.thinkingLevel as ThinkingLevel | undefined;\n  if (override !== undefined) return override;\n  const modeId = agentControllerContext?.session?.modeId;\n  return resolveDefaultThinkingLevel(loadSettings(settingsPath), modeId).level;\n}\n\n/**\n * Dynamic model function that reads the current model from controller state.\n * This allows runtime model switching via the /models picker.\n */\nexport function getDynamicModel(\n  { requestContext }: { requestContext: RequestContext },\n  settingsPath?: string,\n): ResolvedModel {\n  const agentControllerContext = requestContext.get('controller') as AgentControllerRequestContext<any> | undefined;\n\n  const modelId = agentControllerContext?.session?.modelId;\n  if (!modelId) {\n    // A missing controller context means the run was started without session\n    // request context at all (e.g. a signal delivered to an idle thread) —\n    // \"use /models\" would mislead there, the user's selection was never the\n    // problem.\n    if (!agentControllerContext) {\n      throw new Error(\n        'No model available: this run started without a controller session context, so no model selection could be resolved.',\n      );\n    }\n    throw new Error('No model selected. Use /models to select a model first.');\n  }\n\n  const thinkingLevel = resolveRequestThinkingLevel(agentControllerContext, settingsPath);\n\n  return resolveModel(modelId, { thinkingLevel, remapForCodexOAuth: true, requestContext });\n}\n\n/**\n * Goal judge model resolver for the agent's `goal.judge` config. Resolves the\n * configured goal judge model through mastracode's gateway so provider\n * credentials (stored in auth storage, not just env) are injected — a bare model\n * id handed to core's default model router would fail to find the API key.\n *\n * Returns `undefined` when no judge model is configured, which keeps the goal\n * step a complete no-op (the goal mechanism requires a judge to do anything).\n *\n * `settingsPath` must be the same source `createMastraCode()` reads from so the\n * judge model and the goal budget (`goalMaxTurns`) come from one config — with a\n * custom `settingsPath` a bare `loadSettings()` here could read a different file\n * and silently turn the goal step into a no-op.\n */\nexport function getGoalJudgeModel(\n  { requestContext }: { requestContext: RequestContext },\n  settingsPath?: string,\n): ResolvedModel | undefined {\n  const judgeModelId = loadSettings(settingsPath).models.goalJudgeModel;\n  if (!judgeModelId) return undefined;\n  return resolveModel(judgeModelId, { remapForCodexOAuth: true, requestContext });\n}\n"],"mappings":";;;;;;AA8CA,SAAS,0BAA0B,gBAAkE;CACnG,MAAM,yBAAyB,gBAAgB,IAAI,YAAY;CAC/D,MAAM,UAAU;EACd,GAAI,wBAAwB,WAAW,EAAE,eAAe,uBAAuB,SAAS,IAAI,CAAC;EAC7F,GAAI,wBAAwB,aAAa,EAAE,iBAAiB,uBAAuB,WAAW,IAAI,CAAC;CACrG;CAEA,OAAO,OAAO,KAAK,OAAO,CAAC,CAAC,SAAS,IAAI,UAAU,KAAA;AACrD;AAEA,SAAgB,wBAAwB,SAAsD;CAC5F,OAAO,IAAI,kBAAkB,OAAO;AACtC;AAEA,SAAgB,qCAAqC,SAAsC;CACzF,OAAO,mBAAmB,oBACtB,QAAQ,2BAA2B,IACnC,kBAAkB,2BAA2B,OAAO;AAC1D;;;;;;AAOA,SAAgB,eAAe,SAAyB;CACtD,OAAO;AACT;;;;;;;;;;AAWA,SAAgB,aACd,SACA,SACsB;CACtB,kBAAkB;CAClB,MAAM,UAAU,0BAA0B,SAAS,cAAc;CACjE,MAAM,WAAW,aAAa;CAK9B,MAAM,sBAAsB,GAAG,sBAAsB;CACrD,MAAM,yBAAyB,QAAQ,WAAW,mBAAmB,IACjE,QAAQ,MAAM,sBAAsB,SAAS,CAAC,IAC9C;CAIJ,MAAM,kBAAkB,uBAAuB,SAAS,cAAc,KAAK,SAAS;CAOpF,MAAM,kBAAkB,oCAAoC,wBAAwB,eAAe;CACnG,MAAM,uBAAuB,gBAAgB,WAAW,qBAAqB;CAC7E,MAAM,oBAAoB,yBAAyB,eAAe;CAClE,MAAM,CAAC,YAAY,GAAG,cAAc,kBAAkB,MAAM,GAAG;CAC/D,MAAM,cAAc,WAAW,KAAK,GAAG;CACvC,IAAI,CAAC,cAAc,CAAC,aAClB,MAAM,IAAI,MAAM,qBAAqB,SAAS;CAGhD,IAAI,eAAA,kBAA0C;EAC5C,MAAM,iBAAiB,2BAA2B;EAClD,MAAM,WAAW,GAAG,0BAA0B,GAAG;EACjD,MAAM,OAAO,eAAe,YAAY;GACtC,WAAW;GACX,YAAY;GACZ,SAAS;GACT;EACF,CAAC;EACD,OAAO,eAAe,qBAAqB;GACzC,YAAY;GACZ,SAAS;GACT,QAAQ,MAAM,UAAU;GACxB;EACF,CAAC;CACH;CAEA,MAAM,WAAW,GAAG,sBAAsB,GAAG;CAE7C,MAAM,WAAW,kBAAkB,uBAAuB;CAC1D,MAAM,iBACJ,SAAS,eAAe,WAAW,QAAQ,IAAI,yBAAyB;CAI1E,MAAM,kBAAkB,uBAAuB,SAAS,cAAc;CACtE,MAAM,UAAU,wBAAwB;EACtC,sBAAsB,eAAe,QAAQ,QAAQ,EAAE,CAAC,CAAC,QAAQ,SAAS,EAAE;EAC5E,qBAAqB;EACrB,2BAA2B,QAAQ,YAAY,oBAAoB;EACnE,eAAe,SAAS;EACxB;EACA;CACF,CAAC;CAED,MAAM,OAAO,QAAQ,YAAY;EAC/B,WAAW;EACX;EACA,SAAS;EACT;CACF,CAAC;CAED,OAAO,QAAQ,qBAAqB;EAClC;EACA,SAAS;EACT,QAAQ,MAAM,UAAU;EACxB;CACF,CAAC;AACH;;;;;;;;;;;;;;AAeA,SAAgB,4BACd,wBACA,cACe;CACf,MAAM,WAAW,wBAAwB,OAAO;CAChD,IAAI,aAAa,KAAA,GAAW,OAAO;CACnC,MAAM,SAAS,wBAAwB,SAAS;CAChD,OAAO,4BAA4B,aAAa,YAAY,GAAG,MAAM,CAAC,CAAC;AACzE;;;;;AAMA,SAAgB,gBACd,EAAE,kBACF,cACe;CACf,MAAM,yBAAyB,eAAe,IAAI,YAAY;CAE9D,MAAM,UAAU,wBAAwB,SAAS;CACjD,IAAI,CAAC,SAAS;EAKZ,IAAI,CAAC,wBACH,MAAM,IAAI,MACR,qHACF;EAEF,MAAM,IAAI,MAAM,yDAAyD;CAC3E;CAIA,OAAO,aAAa,SAAS;EAAE,eAFT,4BAA4B,wBAAwB,YAE/B;EAAG,oBAAoB;EAAM;CAAe,CAAC;AAC1F;;;;;;;;;;;;;;;AAgBA,SAAgB,kBACd,EAAE,kBACF,cAC2B;CAC3B,MAAM,eAAe,aAAa,YAAY,CAAC,CAAC,OAAO;CACvD,IAAI,CAAC,cAAc,OAAO,KAAA;CAC1B,OAAO,aAAa,cAAc;EAAE,oBAAoB;EAAM;CAAe,CAAC;AAChF"}

package/dist/agents/modes/build.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"build.d.ts","sourceRoot":"","sources":["../../../src/agents/modes/build.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,eAAO,MAAM,SAAS,EAAE,mBAyCvB,CAAC"}
+{"version":3,"file":"build.d.ts","sourceRoot":"","sources":["../../../src/agents/modes/build.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,eAAO,MAAM,SAAS,EAAE,mBA8CvB,CAAC"}

package/dist/agents/modes/build.js

@@ -34,7 +34,12 @@ End with a structured summary:
 . **Completed**: What you implemented (1-2 sentences)
 . **Changes**: Files modified/created
 . **Verification**: How you verified it works
-. **Notes**: Follow-up needed (if any)`,
+. **Notes**: Follow-up needed (if any)
+
+## Workflows
+- To build a workflow: call \`create-workflow\` with the user's request verbatim. A focused sub-agent handles discovery, composition, and saving — don't try to do it inline.
+- To run an existing workflow: call \`run-workflow\` with { workflowId, inputData }. Returns { status, result, error? }.
+- To inspect or manage: \`list-workflows\`, \`get-workflow\`, \`delete-workflow\`.`,
 	defaultModelId: "openai/gpt-5.5",
 	metadata: { default: true }
 };

package/dist/agents/modes/build.js.map

@@ -1 +1 @@
-{"version":3,"file":"build.js","names":[],"sources":["../../../src/agents/modes/build.ts"],"sourcesContent":["/**\n * Execute subagent — focused task execution with write capabilities.\n *\n * This subagent is given a specific implementation task and uses both\n * read and write tools to complete it. It can modify files, run commands,\n * and perform actual development work within a constrained scope.\n */\nimport type { AgentControllerMode } from '@mastra/core/agent-controller';\n\nexport const buildMode: AgentControllerMode = {\n  id: 'build',\n  name: 'Build',\n  description:\n    \"Task execution with write capabilities. Use for 'implement feature X', 'fix bug Y', 'refactor module Z'.\",\n  instructions: `You are a focused execution agent. Your job is to complete a specific, well-defined task by making the necessary changes to the codebase.\n\n## Rules\n- You have FULL ACCESS to read, write, and execute within your task scope.\n- Stay focused on the specific task given. Do not make unrelated changes.\n- Read files before modifying them — use view first, then string_replace_lsp or write_file.\n- Verify your changes work by running relevant tests or checking for errors.\n\n## Tool Strategy\n- **Read first**: Always view a file before editing it\n- **Edit precisely**: Use string_replace_lsp with enough context to match uniquely\n- **Use specialized tools**: Prefer view/search_content/find_files over shell commands for reading\n- **Parallelize**: Make independent tool calls together (e.g., view multiple files at once)\n\n## Workflow\n. Understand the task and explore relevant code\n. For complex tasks (3+ steps): track progress internally and summarize it in your final answer\n. Make changes incrementally — verify each change before moving on\n. Run tests or type-check to verify\n\n## Efficiency\nYour output returns to the parent agent. Be concise:\n- Don't repeat file contents in your response\n- Summarize what changed, don't narrate each step\n- Keep your final summary under 300 words\n\n## Output Format\nEnd with a structured summary:\n. **Completed**: What you implemented (1-2 sentences)\n. **Changes**: Files modified/created\n. **Verification**: How you verified it works\n. **Notes**: Follow-up needed (if any)`,\n  defaultModelId: 'openai/gpt-5.5',\n  metadata: {\n    default: true,\n  },\n};\n"],"mappings":";AASA,MAAa,YAAiC;CAC5C,IAAI;CACJ,MAAM;CACN,aACE;CACF,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAgCd,gBAAgB;CAChB,UAAU,EACR,SAAS,KACX;AACF"}
+{"version":3,"file":"build.js","names":[],"sources":["../../../src/agents/modes/build.ts"],"sourcesContent":["/**\n * Execute subagent — focused task execution with write capabilities.\n *\n * This subagent is given a specific implementation task and uses both\n * read and write tools to complete it. It can modify files, run commands,\n * and perform actual development work within a constrained scope.\n */\nimport type { AgentControllerMode } from '@mastra/core/agent-controller';\n\nexport const buildMode: AgentControllerMode = {\n  id: 'build',\n  name: 'Build',\n  description:\n    \"Task execution with write capabilities. Use for 'implement feature X', 'fix bug Y', 'refactor module Z'.\",\n  instructions: `You are a focused execution agent. Your job is to complete a specific, well-defined task by making the necessary changes to the codebase.\n\n## Rules\n- You have FULL ACCESS to read, write, and execute within your task scope.\n- Stay focused on the specific task given. Do not make unrelated changes.\n- Read files before modifying them — use view first, then string_replace_lsp or write_file.\n- Verify your changes work by running relevant tests or checking for errors.\n\n## Tool Strategy\n- **Read first**: Always view a file before editing it\n- **Edit precisely**: Use string_replace_lsp with enough context to match uniquely\n- **Use specialized tools**: Prefer view/search_content/find_files over shell commands for reading\n- **Parallelize**: Make independent tool calls together (e.g., view multiple files at once)\n\n## Workflow\n. Understand the task and explore relevant code\n. For complex tasks (3+ steps): track progress internally and summarize it in your final answer\n. Make changes incrementally — verify each change before moving on\n. Run tests or type-check to verify\n\n## Efficiency\nYour output returns to the parent agent. Be concise:\n- Don't repeat file contents in your response\n- Summarize what changed, don't narrate each step\n- Keep your final summary under 300 words\n\n## Output Format\nEnd with a structured summary:\n. **Completed**: What you implemented (1-2 sentences)\n. **Changes**: Files modified/created\n. **Verification**: How you verified it works\n. **Notes**: Follow-up needed (if any)\n\n## Workflows\n- To build a workflow: call \\`create-workflow\\` with the user's request verbatim. A focused sub-agent handles discovery, composition, and saving — don't try to do it inline.\n- To run an existing workflow: call \\`run-workflow\\` with { workflowId, inputData }. Returns { status, result, error? }.\n- To inspect or manage: \\`list-workflows\\`, \\`get-workflow\\`, \\`delete-workflow\\`.`,\n  defaultModelId: 'openai/gpt-5.5',\n  metadata: {\n    default: true,\n  },\n};\n"],"mappings":";AASA,MAAa,YAAiC;CAC5C,IAAI;CACJ,MAAM;CACN,aACE;CACF,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqCd,gBAAgB;CAChB,UAAU,EACR,SAAS,KACX;AACF"}

package/dist/agents/modes/explore.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"explore.d.ts","sourceRoot":"","sources":["../../../src/agents/modes/explore.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,eAAO,MAAM,QAAQ,EAAE,mBAkCtB,CAAC"}
+{"version":3,"file":"explore.d.ts","sourceRoot":"","sources":["../../../src/agents/modes/explore.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,eAAO,MAAM,QAAQ,EAAE,mBAqCtB,CAAC"}

package/dist/agents/modes/explore.js

@@ -30,7 +30,10 @@ End with a structured summary:
 . **Key Files**: Most relevant files with line numbers
 . **Details**: Additional context if needed
 
-Keep your summary under 300 words.`,
+Keep your summary under 300 words.
+
+## Workflows
+- You can use \`list-workflows\` and \`get-workflow\` to inspect saved workflows. You cannot create, run, or delete in this mode.`,
 	availableTools: [...EXPLORE_MODE_AVAILABLE_TOOLS]
 };
 //#endregion

package/dist/agents/modes/explore.js.map

@@ -1 +1 @@
-{"version":3,"file":"explore.js","names":[],"sources":["../../../src/agents/modes/explore.ts"],"sourcesContent":["/**\n * Explore subagent — read-only codebase exploration.\n *\n * This subagent is given a focused task (e.g., \"find all usages of X\",\n * \"understand how module Y works\") and uses read-only tools to explore\n * the codebase, then returns a concise summary of its findings.\n */\nimport type { AgentControllerMode } from '@mastra/core/agent-controller';\nimport { EXPLORE_MODE_AVAILABLE_TOOLS } from '../tool-availability.js';\n\nexport const fastMode: AgentControllerMode = {\n  id: 'fast',\n  name: 'Explore',\n  description:\n    \"Read-only codebase exploration. Use for questions like 'find all usages of X', 'how does module Y work'.\",\n  defaultModelId: 'openai/gpt-5.4-mini',\n  instructions: `You are an expert code explorer. Your job is to investigate a codebase and answer a specific question or gather specific information.\n\n## Rules\n- You have READ-ONLY access. You cannot modify files or run commands.\n- Be thorough — search broadly first, then drill into relevant files.\n- After gathering enough information, produce a clear, concise summary of your findings.\n\n## Tool Strategy\n- **Start broad**: Use find_files (glob) to understand project structure\n- **Search smart**: Use search_content (grep) with specific patterns — avoid overly broad searches\n- **Read efficiently**: Use view with view_range for large files — don't read entire files if you only need a section\n- **Parallelize**: Make multiple independent tool calls in one round when exploring different areas\n\n## Efficiency\nYour output returns to the parent agent. Be concise:\n- Don't include raw file contents in your response — summarize what you found\n- Reference files by path and line number, not by copying code\n- If a search returns many results, report the count and key examples, not every match\n\n## Output Format\nEnd with a structured summary:\n. **Answer**: Direct answer to the question (1-2 sentences)\n. **Key Files**: Most relevant files with line numbers\n. **Details**: Additional context if needed\n\nKeep your summary under 300 words.`,\n\n  availableTools: [...EXPLORE_MODE_AVAILABLE_TOOLS],\n};\n"],"mappings":";;AAUA,MAAa,WAAgC;CAC3C,IAAI;CACJ,MAAM;CACN,aACE;CACF,gBAAgB;CAChB,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;CA2Bd,gBAAgB,CAAC,GAAG,4BAA4B;AAClD"}
+{"version":3,"file":"explore.js","names":[],"sources":["../../../src/agents/modes/explore.ts"],"sourcesContent":["/**\n * Explore subagent — read-only codebase exploration.\n *\n * This subagent is given a focused task (e.g., \"find all usages of X\",\n * \"understand how module Y works\") and uses read-only tools to explore\n * the codebase, then returns a concise summary of its findings.\n */\nimport type { AgentControllerMode } from '@mastra/core/agent-controller';\nimport { EXPLORE_MODE_AVAILABLE_TOOLS } from '../tool-availability.js';\n\nexport const fastMode: AgentControllerMode = {\n  id: 'fast',\n  name: 'Explore',\n  description:\n    \"Read-only codebase exploration. Use for questions like 'find all usages of X', 'how does module Y work'.\",\n  defaultModelId: 'openai/gpt-5.4-mini',\n  instructions: `You are an expert code explorer. Your job is to investigate a codebase and answer a specific question or gather specific information.\n\n## Rules\n- You have READ-ONLY access. You cannot modify files or run commands.\n- Be thorough — search broadly first, then drill into relevant files.\n- After gathering enough information, produce a clear, concise summary of your findings.\n\n## Tool Strategy\n- **Start broad**: Use find_files (glob) to understand project structure\n- **Search smart**: Use search_content (grep) with specific patterns — avoid overly broad searches\n- **Read efficiently**: Use view with view_range for large files — don't read entire files if you only need a section\n- **Parallelize**: Make multiple independent tool calls in one round when exploring different areas\n\n## Efficiency\nYour output returns to the parent agent. Be concise:\n- Don't include raw file contents in your response — summarize what you found\n- Reference files by path and line number, not by copying code\n- If a search returns many results, report the count and key examples, not every match\n\n## Output Format\nEnd with a structured summary:\n. **Answer**: Direct answer to the question (1-2 sentences)\n. **Key Files**: Most relevant files with line numbers\n. **Details**: Additional context if needed\n\nKeep your summary under 300 words.\n\n## Workflows\n- You can use \\`list-workflows\\` and \\`get-workflow\\` to inspect saved workflows. You cannot create, run, or delete in this mode.`,\n\n  availableTools: [...EXPLORE_MODE_AVAILABLE_TOOLS],\n};\n"],"mappings":";;AAUA,MAAa,WAAgC;CAC3C,IAAI;CACJ,MAAM;CACN,aACE;CACF,gBAAgB;CAChB,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8Bd,gBAAgB,CAAC,GAAG,4BAA4B;AAClD"}

package/dist/agents/modes/plan.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"plan.d.ts","sourceRoot":"","sources":["../../../src/agents/modes/plan.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,eAAO,MAAM,QAAQ,EAAE,mBAmCtB,CAAC"}
+{"version":3,"file":"plan.d.ts","sourceRoot":"","sources":["../../../src/agents/modes/plan.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AAGzE,eAAO,MAAM,QAAQ,EAAE,mBAwCtB,CAAC"}

package/dist/agents/modes/plan.js

@@ -10,7 +10,7 @@ const planMode = {
 
 ## Rules
 - You have READ-ONLY access to the project. You cannot modify project files or run commands.
-- The one exception is plan files: you can create and edit markdown files inside \`.mastracode/plans/\` using \`write_file\`, \`view\`, and \`string_replace_lsp\`. You may not write anywhere else.
+- The one exception is plan files: you can create and edit markdown files inside the session-specific plan directory identified in your system prompt using \`write_file\`, \`view\`, and \`string_replace_lsp\`. You may not write anywhere else.
 - First, explore the codebase to understand existing patterns, architecture, and conventions.
 - Produce a concrete, actionable plan — not vague suggestions.
 
@@ -21,13 +21,18 @@ const planMode = {
 - **Parallelize**: Make multiple independent tool calls when exploring different areas
 
 ## Plan Delivery
-- Write your plan to a markdown file under \`.mastracode/plans/\` (e.g. \`.mastracode/plans/add-dark-mode.md\`) using \`write_file\`, then call \`submit_plan({ path })\` with the path to that file (never the plan body).
+- Write your plan to a markdown file in the session-specific plan directory identified in your system prompt using \`write_file\`, then call \`submit_plan({ path })\` with the path to that file (never the plan body).
 - Start the file with a \`# Title\` heading describing the plan.
 - Reuse the same file while iterating on the same plan; only create a new file for a genuinely different plan so each plan stays available to review.
 - Do NOT output the plan as text — it MUST live in the plan file.
 - Be concise: reference files by path and line number, don't include raw contents.
 - Focus on actionable details, not general observations.
-- To revise after "Request changes", edit the same file in place with \`string_replace_lsp\`, and call \`submit_plan\` again with the same path.`,
+- To revise after "Request changes", edit the same file in place with \`string_replace_lsp\`, and call \`submit_plan\` again with the same path.
+
+## Workflows
+- You can INSPECT saved workflows via \`list-workflows\` and \`get-workflow\`.
+- You CANNOT build, run, or delete workflows in this mode.
+- If the user asks for a workflow design, include it in the implementation plan stored at the session-specific plan path, then call \`submit_plan({ path })\` as required above. After approval, build mode can save and run it.`,
 	metadata: { default: false },
 	availableTools: [...PLAN_MODE_AVAILABLE_TOOLS]
 };

package/dist/agents/modes/plan.js.map

@@ -1 +1 @@
-{"version":3,"file":"plan.js","names":[],"sources":["../../../src/agents/modes/plan.ts"],"sourcesContent":["/**\n * Plan mode — read-only analysis and planning.\n */\nimport type { AgentControllerMode } from '@mastra/core/agent-controller';\nimport { PLAN_MODE_AVAILABLE_TOOLS } from '../tool-availability.js';\n\nexport const planMode: AgentControllerMode = {\n  id: 'plan',\n  name: 'Plan',\n  transitionsTo: 'build',\n  defaultModelId: 'openai/gpt-5.5',\n  description:\n    \"Read-only analysis and planning. Use for 'create an implementation plan for X', 'analyze the architecture of Y'.\",\n  instructions: `You are an expert software architect and planner. Your job is to analyze a codebase and produce a detailed implementation plan for a given task.\n\n## Rules\n- You have READ-ONLY access to the project. You cannot modify project files or run commands.\n- The one exception is plan files: you can create and edit markdown files inside \\`.mastracode/plans/\\` using \\`write_file\\`, \\`view\\`, and \\`string_replace_lsp\\`. You may not write anywhere else.\n- First, explore the codebase to understand existing patterns, architecture, and conventions.\n- Produce a concrete, actionable plan — not vague suggestions.\n\n## Tool Strategy\n- **Discover structure**: Use find_files (glob) to understand project layout and find relevant files\n- **Find patterns**: Use search_content (grep) to locate existing implementations, imports, and conventions\n- **Understand deeply**: Use view with view_range to read specific sections of key files\n- **Parallelize**: Make multiple independent tool calls when exploring different areas\n\n## Plan Delivery\n- Write your plan to a markdown file under \\`.mastracode/plans/\\` (e.g. \\`.mastracode/plans/add-dark-mode.md\\`) using \\`write_file\\`, then call \\`submit_plan({ path })\\` with the path to that file (never the plan body).\n- Start the file with a \\`# Title\\` heading describing the plan.\n- Reuse the same file while iterating on the same plan; only create a new file for a genuinely different plan so each plan stays available to review.\n- Do NOT output the plan as text — it MUST live in the plan file.\n- Be concise: reference files by path and line number, don't include raw contents.\n- Focus on actionable details, not general observations.\n- To revise after \"Request changes\", edit the same file in place with \\`string_replace_lsp\\`, and call \\`submit_plan\\` again with the same path.`,\n\n  metadata: {\n    default: false,\n  },\n\n  availableTools: [...PLAN_MODE_AVAILABLE_TOOLS],\n};\n"],"mappings":";;AAMA,MAAa,WAAgC;CAC3C,IAAI;CACJ,MAAM;CACN,eAAe;CACf,gBAAgB;CAChB,aACE;CACF,cAAc;;;;;;;;;;;;;;;;;;;;;;CAuBd,UAAU,EACR,SAAS,MACX;CAEA,gBAAgB,CAAC,GAAG,yBAAyB;AAC/C"}
+{"version":3,"file":"plan.js","names":[],"sources":["../../../src/agents/modes/plan.ts"],"sourcesContent":["/**\n * Plan mode — read-only analysis and planning.\n */\nimport type { AgentControllerMode } from '@mastra/core/agent-controller';\nimport { PLAN_MODE_AVAILABLE_TOOLS } from '../tool-availability.js';\n\nexport const planMode: AgentControllerMode = {\n  id: 'plan',\n  name: 'Plan',\n  transitionsTo: 'build',\n  defaultModelId: 'openai/gpt-5.5',\n  description:\n    \"Read-only analysis and planning. Use for 'create an implementation plan for X', 'analyze the architecture of Y'.\",\n  instructions: `You are an expert software architect and planner. Your job is to analyze a codebase and produce a detailed implementation plan for a given task.\n\n## Rules\n- You have READ-ONLY access to the project. You cannot modify project files or run commands.\n- The one exception is plan files: you can create and edit markdown files inside the session-specific plan directory identified in your system prompt using \\`write_file\\`, \\`view\\`, and \\`string_replace_lsp\\`. You may not write anywhere else.\n- First, explore the codebase to understand existing patterns, architecture, and conventions.\n- Produce a concrete, actionable plan — not vague suggestions.\n\n## Tool Strategy\n- **Discover structure**: Use find_files (glob) to understand project layout and find relevant files\n- **Find patterns**: Use search_content (grep) to locate existing implementations, imports, and conventions\n- **Understand deeply**: Use view with view_range to read specific sections of key files\n- **Parallelize**: Make multiple independent tool calls when exploring different areas\n\n## Plan Delivery\n- Write your plan to a markdown file in the session-specific plan directory identified in your system prompt using \\`write_file\\`, then call \\`submit_plan({ path })\\` with the path to that file (never the plan body).\n- Start the file with a \\`# Title\\` heading describing the plan.\n- Reuse the same file while iterating on the same plan; only create a new file for a genuinely different plan so each plan stays available to review.\n- Do NOT output the plan as text — it MUST live in the plan file.\n- Be concise: reference files by path and line number, don't include raw contents.\n- Focus on actionable details, not general observations.\n- To revise after \"Request changes\", edit the same file in place with \\`string_replace_lsp\\`, and call \\`submit_plan\\` again with the same path.\n\n## Workflows\n- You can INSPECT saved workflows via \\`list-workflows\\` and \\`get-workflow\\`.\n- You CANNOT build, run, or delete workflows in this mode.\n- If the user asks for a workflow design, include it in the implementation plan stored at the session-specific plan path, then call \\`submit_plan({ path })\\` as required above. After approval, build mode can save and run it.`,\n\n  metadata: {\n    default: false,\n  },\n\n  availableTools: [...PLAN_MODE_AVAILABLE_TOOLS],\n};\n"],"mappings":";;AAMA,MAAa,WAAgC;CAC3C,IAAI;CACJ,MAAM;CACN,eAAe;CACf,gBAAgB;CAChB,aACE;CACF,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;CA4Bd,UAAU,EACR,SAAS,MACX;CAEA,gBAAgB,CAAC,GAAG,yBAAyB;AAC/C"}

package/dist/agents/prompts/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agents/prompts/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,eAAe,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAChE,OAAO,EAAE,cAAc,EAAE,MAAM,WAAW,CAAC;AAC3C,OAAO,EAAE,cAAc,EAAE,MAAM,WAAW,CAAC;AAG3C,OAAO,KAAK,EAAE,aAAa,IAAI,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAUpF,MAAM,WAAW,aAAc,SAAQ,IAAI,CAAC,iBAAiB,EAAE,cAAc,CAAC;IAC5E,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,GAAG,CAAC;IACZ,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;CACpB;AAQD;;;GAGG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,aAAa,GAAG,MAAM,CAqE1D"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agents/prompts/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,eAAe,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAChE,OAAO,EAAE,cAAc,EAAE,MAAM,WAAW,CAAC;AAC3C,OAAO,EAAE,cAAc,EAAE,MAAM,WAAW,CAAC;AAG3C,OAAO,KAAK,EAAE,aAAa,IAAI,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAWpF,MAAM,WAAW,aAAc,SAAQ,IAAI,CAAC,iBAAiB,EAAE,cAAc,CAAC;IAC5E,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,GAAG,CAAC;IACZ,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;CACpB;AAQD;;;GAGG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,aAAa,GAAG,MAAM,CA0E1D"}

package/dist/agents/prompts/index.js

@@ -1,4 +1,5 @@
 import { buildModePrompt, buildModePromptFn } from "./build.js";
+import { getLocalPlansRelativeDir } from "../../utils/plans.js";
 import { planModePrompt } from "./plan.js";
 import { fastModePrompt } from "./fast.js";
 import { hasTavilyKey } from "../../tools/web-search.js";
@@ -25,9 +26,11 @@ function buildFullPrompt(ctx) {
 	if (permRules?.tools) {
 		for (const [name, policy] of Object.entries(permRules.tools)) if (policy === "deny") deniedTools.add(name);
 	}
+	const factoryProjectId = typeof ctx.state?.factoryProjectId === "string" ? ctx.state.factoryProjectId : void 0;
 	const toolGuidance = buildToolGuidance(ctx.modeId, {
 		hasWebSearch,
-		deniedTools
+		deniedTools,
+		plansDir: getLocalPlansRelativeDir({ factoryProjectId })
 	});
 	const base = buildBasePrompt({
 		projectPath: ctx.workingDir,