eve 0.64.1 → 0.66.0

package/dist/src/tools/schema.js

@@ -1 +1 @@
-import{asSchema}from"ai";import{z}from"#compiled/zod/index.js";import{toErrorMessage}from"#shared/errors.js";import{parseJsonObject}from"#shared/json.js";const rehydratedSchemas={input:new WeakMap,output:new WeakMap};function toInputSchema(e){return toSchema(e,`input`)}function toOutputSchema(e){return toSchema(e,`output`)}function serializeInputSchema(e){return serializeSchema(e,`input`)}function serializeOutputSchema(e){return serializeSchema(e,`output`)}function isToolSchema(e){let t=getStandardSchemaProperties(e);if(t===void 0)return!1;let n=t.jsonSchema;return typeof t.validate==`function`&&typeof n==`object`&&!!n&&typeof n.input==`function`&&typeof n.output==`function`}const UNSPECIFIED_INPUT_SCHEMA=z.fromJSONSchema({});function toSchema(e,t){if(e==null||isToolSchema(e))return e;let n=rehydratedSchemas[t],r=n.get(e);return r===void 0&&(r=rehydrateJsonSchema(toJsonObject(e,t)),n.set(e,r)),r}function rehydrateJsonSchema(e){try{return z.fromJSONSchema(e,{defaultTarget:`draft-7`})}catch{}try{return z.fromJSONSchema(e,{defaultTarget:`draft-2020-12`})}catch(t){return console.warn(`[eve] Tool schema uses JSON Schema features outside local validation support; passing input through unvalidated: ${toErrorMessage(t)}`),toPassthroughSchema(e)}}function toPassthroughSchema(e){let emit=()=>structuredClone(e);return{"~standard":{version:1,vendor:`eve`,validate:e=>({value:e}),jsonSchema:{input:emit,output:emit}}}}function serializeSchema(e,t){return e==null?e:toJsonObject(e,t)}function toJsonObject(e,t){let n=getStandardSchemaProperties(e),r=n?.jsonSchema,i=typeof r==`object`&&r?r[t]:void 0,a=typeof n?.vendor==`string`?n.vendor:`unknown`;if(n!==void 0&&typeof i!=`function`&&a===`zod`){if(t===`input`){let t=asSchema(e),{$schema:n,...r}=parseJsonObject(t.jsonSchema);return r}throw Error(`Zod 3 cannot emit an output JSON Schema. Upgrade to Zod 4 or provide a plain JSON Schema object.`)}if(n!==void 0&&typeof i!=`function`)throw Error(`Standard Schema vendor "${a}" does not support JSON Schema conversion. Provide a Standard Schema implementation with JSON Schema conversion or a plain JSON Schema object.`);let{$schema:o,...s}=parseJsonObject(n===void 0?e:i({target:`draft-07`}));return s}function getStandardSchemaProperties(e){if(typeof e!=`object`||!e||!(`~standard`in e))return;let t=e[`~standard`];return typeof t==`object`&&t?t:void 0}export{UNSPECIFIED_INPUT_SCHEMA,isToolSchema,serializeInputSchema,serializeOutputSchema,toInputSchema,toOutputSchema};
+import{TypeValidationError,asSchema,jsonSchema}from"ai";import{isObject}from"#shared/guards.js";import{toErrorMessage}from"#shared/errors.js";import{parseJsonObject}from"#shared/json.js";import{Validator}from"#compiled/@cfworker/json-schema/index.js";function toInputSchema(e){return toSchema(e,`input`)}function toOutputSchema(e){return toSchema(e,`output`)}function serializeInputSchema(e){return serializeSchema(e,`input`)}function serializeOutputSchema(e){return serializeSchema(e,`output`)}function isToolSchema(e){let t=getStandardSchemaProperties(e);if(t===void 0)return!1;let n=t.jsonSchema;return typeof t.validate==`function`&&typeof n==`object`&&!!n&&typeof n.input==`function`&&typeof n.output==`function`}function defineJsonSchema(e,t){let emit=()=>structuredClone(e),n={"~standard":{version:1,vendor:`eve`,validate:createJsonSchemaValidator(e,t),jsonSchema:{input:emit,output:emit}}};return plainJsonSchemas.add(n),n}const plainJsonSchemas=new WeakSet,UNSPECIFIED_INPUT_SCHEMA=defineJsonSchema({});function toModelSchema(e,t){if(e===void 0||typeof e==`function`||!(`~standard`in e))return e;let n=e,r=plainJsonSchemas.has(n);return jsonSchema(()=>{let e=serializeSchema(n,t);return r?e:closeObjectSchemas(e)},{validate:async e=>{let t=await n[`~standard`].validate(e);return t.issues===void 0?{success:!0,value:t.value}:{success:!1,error:new TypeValidationError({value:e,cause:t.issues})}}})}function closeObjectSchemas(e){if(!isObject(e))return e;let t={...e},{type:n}=t;(n===`object`||Array.isArray(n)&&n.includes(`object`))&&(t.additionalProperties=isObject(t.additionalProperties)?closeObjectSchemas(t.additionalProperties):!1,isObject(t.properties)&&(t.properties=mapValues(t.properties,closeObjectSchemas))),Array.isArray(t.items)?t.items=t.items.map(closeObjectSchemas):isObject(t.items)&&(t.items=closeObjectSchemas(t.items));for(let e of[`allOf`,`anyOf`,`oneOf`]){let n=t[e];Array.isArray(n)&&(t[e]=n.map(closeObjectSchemas))}return isObject(t.definitions)&&(t.definitions=mapValues(t.definitions,closeObjectSchemas)),t}function mapValues(e,t){return Object.fromEntries(Object.entries(e).map(([e,n])=>[e,t(n)]))}function toSchema(e,t){return e==null||isToolSchema(e)?e:defineJsonSchema(toJsonObject(e,t))}function serializeSchema(e,t){return e==null?e:toJsonObject(e,t)}function toJsonObject(e,t){let n=getStandardSchemaProperties(e),r=n?.jsonSchema,i=typeof r==`object`&&r?r[t]:void 0,a=typeof n?.vendor==`string`?n.vendor:`unknown`;if(n!==void 0&&typeof i!=`function`&&a===`zod`){if(t===`input`){let t=asSchema(e),{$schema:n,...r}=parseJsonObject(t.jsonSchema);return r}throw Error(`Zod 3 cannot emit an output JSON Schema. Upgrade to Zod 4 or provide a plain JSON Schema object.`)}if(n!==void 0&&typeof i!=`function`)throw Error(`Standard Schema vendor "${a}" does not support JSON Schema conversion. Provide a Standard Schema implementation with JSON Schema conversion or a plain JSON Schema object.`);let{$schema:o,...s}=parseJsonObject(n===void 0?e:i({target:`draft-07`}));return s}function getStandardSchemaProperties(e){if(typeof e!=`object`||!e||!(`~standard`in e))return;let t=e[`~standard`];return typeof t==`object`&&t?t:void 0}function createJsonSchemaValidator(e,t){let n,r=!1,passThrough=(e,t)=>(r||(r=!0,console.warn(`[eve] Tool schema cannot validate values locally; passing them through unvalidated: ${t}`)),{value:e});return r=>{if(n===void 0){let t=buildValidator(e);if(typeof t==`string`)return n=null,passThrough(r,t);n=t}if(n===null)return{value:r};let i;try{i=n.validate(r)}catch(e){return r===void 0?{issues:[{message:`Expected a JSON value.`}]}:passThrough(r,toErrorMessage(e))}if(!i.valid)return{issues:toIssues(i.errors)};let a=applyDefaults(e,e,r),o=t?.(a);return o===void 0?{value:a}:{issues:[{message:o}]}}}function applyDefaults(e,t,n){let r=n;for(let n of followRefs(e,t)){let{items:t,properties:i}=n;if(isObject(r)&&isObject(i))r=fillProperties(e,i,r);else if(Array.isArray(r)&&isObject(t)){let n=r,i=n.map(n=>applyDefaults(e,t,n));i.some((e,t)=>e!==n[t])&&(r=i)}}return r}function fillProperties(e,t,n){let r=n;for(let[i,a]of Object.entries(t)){let t=Object.hasOwn(n,i)?n[i]:void 0,o=t===void 0?findDefault(e,a):applyDefaults(e,a,t);o!==t&&(r===n&&(r={...n}),r[i]=o)}return r}function findDefault(e,t){for(let n of followRefs(e,t))if(`default`in n)return structuredClone(n.default)}function*followRefs(e,t){let n=new Set,r=t;for(;isObject(r)&&!n.has(r);)n.add(r),yield r,r=resolveLocalRef(e,r.$ref)}function resolveLocalRef(e,t){if(typeof t!=`string`||t!==`#`&&!t.startsWith(`#/`))return;let n=e;for(let e of decodePointer(t)){if(typeof n!=`object`||!n||!Object.hasOwn(n,e))return;n=n[e]}return n}function buildValidator(e){try{let t=structuredClone(e);return prepareForValidation(t)??new Validator(t,`2020-12`,!1)}catch(e){return toErrorMessage(e)}}const JSON_TYPES=new Set([`array`,`boolean`,`integer`,`null`,`number`,`object`,`string`]),EXCLUSIVE_BOUNDS=[[`exclusiveMinimum`,`minimum`],[`exclusiveMaximum`,`maximum`]],SUBSCHEMA_KEYWORDS=[`additionalItems`,`additionalProperties`,`contains`,`else`,`if`,`not`,`propertyNames`,`then`,`unevaluatedItems`,`unevaluatedProperties`],SUBSCHEMA_LIST_KEYWORDS=[`allOf`,`anyOf`,`oneOf`,`prefixItems`],SUBSCHEMA_MAP_KEYWORDS=[`$defs`,`definitions`,`dependencies`,`dependentSchemas`,`patternProperties`,`properties`];function prepareForValidation(e){if(typeof e==`boolean`)return;if(typeof e!=`object`||!e||Array.isArray(e))return`a subschema is neither an object nor a boolean`;let t=e;delete t.format;for(let[e,n]of EXCLUSIVE_BOUNDS)typeof t[e]==`boolean`&&(t[e]&&typeof t[n]==`number`?(t[e]=t[n],delete t[n]):delete t[e]);let{enum:n,pattern:r,required:i,type:a}=t;if(a!==void 0&&!(Array.isArray(a)?a.every(isJsonType):isJsonType(a)))return`unsupported type ${JSON.stringify(a)}`;if(n!==void 0&&!Array.isArray(n))return"`enum` is not an array";if(i!==void 0&&!(Array.isArray(i)&&i.every(e=>typeof e==`string`)))return"`required` is not an array of strings";if(r!==void 0&&!isRegExpSource(r))return`invalid pattern ${JSON.stringify(r)}`;for(let e of SUBSCHEMA_KEYWORDS)if(e in t){let n=prepareForValidation(t[e]);if(n!==void 0)return n}let o=t.items;for(let e of Array.isArray(o)?o:o===void 0?[]:[o]){let t=prepareForValidation(e);if(t!==void 0)return t}for(let e of SUBSCHEMA_LIST_KEYWORDS){let n=t[e];if(n!==void 0){if(!Array.isArray(n))return`\`${e}\` is not an array`;for(let e of n){let t=prepareForValidation(e);if(t!==void 0)return t}}}for(let e of SUBSCHEMA_MAP_KEYWORDS){let n=t[e];if(n!==void 0){if(typeof n!=`object`||!n||Array.isArray(n))return`\`${e}\` is not an object`;for(let[t,r]of Object.entries(n)){if(e===`patternProperties`&&!isRegExpSource(t))return`invalid pattern ${JSON.stringify(t)}`;if(e===`dependencies`&&Array.isArray(r))continue;let n=prepareForValidation(r);if(n!==void 0)return n}}}}function isJsonType(e){return typeof e==`string`&&JSON_TYPES.has(e)}function isRegExpSource(e){if(typeof e!=`string`)return!1;try{return new RegExp(e,`u`),!0}catch{return!1}}const CLOSED_OBJECT_KEYWORD=/\/(?:additional|unevaluated)Properties$/;function toIssues(e){let t=[];for(let n=0;n<e.length;n++){let r=e[n],i=e[n+1];if(CLOSED_OBJECT_KEYWORD.test(r.keywordLocation)&&i?.keyword===`false`){let e=decodePointer(i.instanceLocation).at(-1);t.push({keyword:`false`,location:i.instanceLocation,message:`Unrecognized key: ${JSON.stringify(e)}`,object:r.instanceLocation}),n++}else t.push({keyword:r.keyword,location:r.instanceLocation,message:r.error})}let n=t.filter(e=>e.object===void 0||!t.some(t=>t.object===void 0&&(t.location===e.location||t.location.startsWith(`${e.location}/`))));return n.filter((e,t)=>{let r=n[t+1];return r===void 0?!0:r.location.startsWith(`${e.location}/`)?!1:e.keyword!==`$ref`||r.location!==e.location}).map(({location:e,message:t,object:n})=>{let r=decodePointer(n??e);return r.length===0?{message:t}:{message:t,path:r}})}function decodePointer(e){return e.split(`/`).slice(1).map(e=>{let t=e;try{t=decodeURIComponent(e)}catch{}return t.replaceAll(`~1`,`/`).replaceAll(`~0`,`~`)})}export{UNSPECIFIED_INPUT_SCHEMA,defineJsonSchema,isToolSchema,serializeInputSchema,serializeOutputSchema,toInputSchema,toModelSchema,toOutputSchema};

package/dist/src/vue/index.js

@@ -1,3 +1,3 @@
-import { n as defaultMessageReducer, t as useEveAgent } from "../chunks/use-eve-agent-BdKlQF32.js";
+import { n as defaultMessageReducer, t as useEveAgent } from "../chunks/use-eve-agent-BUeSjwIk.js";
 
 export { defaultMessageReducer, useEveAgent };

package/dist/src/vue/use-eve-agent.js

@@ -1,3 +1,3 @@
-import { t as useEveAgent } from "../chunks/use-eve-agent-BdKlQF32.js";
+import { t as useEveAgent } from "../chunks/use-eve-agent-BUeSjwIk.js";
 
 export { useEveAgent };

package/docs/README.md

@@ -97,7 +97,7 @@ eve then gives you:
 - a reconnectable session stream
 - durable session state across turns
 - a per-agent sandbox with a shared runtime workspace
-- typed runtime helpers accessed through `ctx` (`ctx.session`, `ctx.getSandbox()`, `ctx.getSkill()`)
+- typed runtime helpers accessed through `ctx` (`ctx.session`, `ctx.getSandbox()`)
 
 ## The runtime shape
 

package/docs/agent-config.md

@@ -13,13 +13,13 @@ A typical config selects a model:
 import { defineAgent } from "eve";
 
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
 });
 ```
 
 For a static AI Gateway model ID, you can make the same source change from the
-project root with `eve set --model anthropic/claude-opus-4.8` or from the local
-dev TUI with `/model anthropic/claude-opus-4.8`.
+project root with `eve set --model anthropic/claude-opus-5.5` or from the local
+dev TUI with `/model anthropic/claude-opus-5.5`.
 
 The root `agent.ts` can be omitted when no runtime config is needed. eve then selects its default `agent.ts` source at the same slot, configured with `spacexai/grok-4.7`; authoring the file replaces that source.
 When `agent.ts` is present, `model` is required.
@@ -39,9 +39,9 @@ export default defineAgent({
 });
 ```
 
-`openai()` from `eve/models/openai` defaults to `gpt-5.6-luna-fast`. Both helpers accept an optional native provider model ID and use `OPENAI_API_KEY` or `ANTHROPIC_API_KEY`. During local development they can also use credentials saved through `/login`. Deployments require their API key in the server environment.
+`openai()` from `eve/models/openai` defaults to `gpt-6-luna-fast`. Both helpers accept an optional native provider model ID and use `OPENAI_API_KEY` or `ANTHROPIC_API_KEY`. During local development they can also use credentials saved through `/login`. Deployments require their API key in the server environment.
 
-For a local ChatGPT subscription, use `chatgpt()` from `eve/models/openai` and sign in with `/login`. It defaults to `gpt-5.6-luna-fast` and cannot run in a deployment.
+For a local ChatGPT subscription, use `chatgpt()` from `eve/models/openai` and sign in with `/login`. It defaults to `gpt-6-luna-fast` and cannot run in a deployment.
 
 `/login` can switch a static Gateway string to an eve helper and manage its import. `/model` changes the selected model and settings immediately. Custom provider SDK calls and dynamic expressions retain their authored behavior and require manual source edits. You can still install an AI SDK provider package and pass its `LanguageModel` when you need provider-specific configuration.
 
@@ -77,7 +77,7 @@ export default defineAgent({
     events: {
       "session.started": (_event, ctx) => {
         if (ctx.session.auth.initiator?.attributes.plan === "enterprise") {
-          return "anthropic/claude-opus-4.8";
+          return "anthropic/claude-opus-5.5";
         }
 
         return "anthropic/claude-sonnet-5";
@@ -141,7 +141,7 @@ Compaction summarizes older turns as you approach the context window. It's on by
 
 ```ts title="agent/agent.ts"
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   compaction: {
     thresholdPercent: 0.75, // default 0.9
   },
@@ -158,7 +158,7 @@ provider-reported tokens or model token cost reaches a configured limit:
 
 ```ts title="agent/agent.ts"
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   limits: {
     maxInputTokensPerSession: 200_000,
     maxOutputTokensPerSession: 20_000,
@@ -228,7 +228,7 @@ root `agent.ts`:
 import { defineAgent } from "eve";
 
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   experimental: {
     workflow: {
       world: "@workflow/world-postgres",
@@ -264,7 +264,7 @@ sequential model calls:
 import { defineAgent } from "eve";
 
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   experimental: {
     workflow: {
       modelCallsPerStep: 4,
@@ -302,7 +302,7 @@ finishes instead:
 import { defineAgent } from "eve";
 
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   experimental: {
     workflow: {
       retention: 0,

package/docs/channels/slack.mdx

@@ -135,7 +135,7 @@ Check the delivery path in order so you can identify where a Slack event stops:
 | Mentions work but DMs or thread replies do not                        | Check the Slack app's Event Subscriptions and OAuth scopes. DMs require `message.im` and `im:history`; channel replies require the matching message event and history scope.           | Add the missing events or scopes, then reinstall the Slack app if Slack requires it.                                                                                 |
 | A long response cannot be uploaded as a snippet                       | Check whether the installed Slack app has the `files:write` bot scope.                                                                                                                 | Add `files:write`, then reinstall the Slack app so the new scope takes effect.                                                                                       |
 | A preview request receives an authentication page or protection error | Check whether the branch domain is exempt or the direct Slack request URL includes `x-vercel-protection-bypass`. Slack cannot complete an interactive Vercel Authentication challenge. | Add a Deployment Protection Exception for the branch domain, or add the automation bypass query parameter to a Slack app that uses environment-variable credentials. |
-| The webhook reaches the deployment but the agent does not reply       | Open the project's Vercel runtime logs and locate the request for the inbound Slack event.                                                                                             | Use the first missing or failed stage to narrow the issue to channel dispatch, model or tool execution, or outbound Slack delivery.                                  |
+| The webhook reaches the deployment but the agent does not reply       | Open the project's Vercel runtime logs, then inspect the session in [**Agent Runs**](/docs/observability/agent-runs) when your team has access.                                        | Use the first missing or failed stage to narrow the issue to channel dispatch, the agent run, or outbound Slack delivery.                                            |
 
 ## Audience
 

package/docs/concepts/built-in-tools.md

@@ -213,72 +213,6 @@ import { disableTool } from "eve/tools";
 export default disableTool();
 ```
 
-### `todo`
-
-`todo` maintains a durable todo list for the session.
-
-```sh
-eve add tool/todo
-```
-
-```ts title="agent/tools/todo.ts"
-export { default } from "eve/tools/todo";
-```
-
-Override it. Spreading the definition preserves its durable state key:
-
-```ts title="agent/tools/todo.ts"
-import { defineTool } from "eve/tools";
-import { todo } from "eve/tools/todo";
-
-export default defineTool({
-  ...todo,
-  description: "Track the current implementation plan.",
-});
-```
-
-Disable it:
-
-```ts title="agent/tools/todo.ts"
-import { disableTool } from "eve/tools";
-
-export default disableTool();
-```
-
-### `ask_question`
-
-`ask_question` asks the user for clarification or a choice, then parks the turn until they answer. It appears only when the session can request user input. See [Human-in-the-loop](/docs/human-in-the-loop).
-
-```sh
-eve add tool/ask_question
-```
-
-```ts title="agent/tools/ask_question.ts"
-export { default } from "eve/tools/ask_question";
-```
-
-Replace its request-input behavior with an ordinary authored tool:
-
-```ts title="agent/tools/ask_question.ts"
-import { defineTool } from "eve/tools";
-
-export default defineTool({
-  description: "Record a clarification request.",
-  inputSchema: { type: "object" },
-  async execute(input) {
-    return { recorded: input };
-  },
-});
-```
-
-Disable it:
-
-```ts title="agent/tools/ask_question.ts"
-import { disableTool } from "eve/tools";
-
-export default disableTool();
-```
-
 ### `agent`
 
 `agent` delegates a subtask to a fresh copy of the root agent. It is root-only, always runs in the background, and returns a task receipt immediately. The child receives the root's instructions, tools, connections, and sandbox, but starts with fresh conversation history and [state](./state). See [Subagents](../subagents).
@@ -365,6 +299,22 @@ You can also add the opt-in framework tools described below.
 
 These framework-provided tools are not added by default. Add only the ones the agent needs.
 
+### `ask_question`
+
+`ask_question` lets the model ask the user one question, then waits for the answer. The model can offer two or three options, each with a label and a short description, and the user can always type their own answer instead. Channels render the options as native UI, such as Slack select menus. Without the tool, the model can still ask in its reply text and the user answers with their next message. See [Human-in-the-loop](/docs/human-in-the-loop). Add it:
+
+```sh
+eve add tool/ask_question
+```
+
+```ts title="agent/tools/ask_question.ts"
+import { askQuestion } from "eve/tools/ask_question";
+
+export default askQuestion();
+```
+
+`ask_question` is a [workflow tool](/docs/tools/workflows) that calls `ctx.ask()`. The model receives `{ status: "answered", answer }`, where `answer` is the chosen option's label or the user's own words. A plain follow-up message answers the question too when it is the only pending question. When other questions are also pending, a message does not answer any of them: `ask_question` resolves as `{ status: "dismissed" }` and the message reaches the model normally. In a session that cannot request input, such as a scheduled run, the result is `{ status: "unavailable" }` and the model continues on its own judgment. Remove the file to remove the tool.
+
 ### `glob`
 
 `glob` finds sandbox files by glob pattern. Add it:

package/docs/concepts/default-harness.md

@@ -11,7 +11,7 @@ The harness keeps a long session from overflowing the model's context window. Be
 
 ```ts title="agent/agent.ts"
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   compaction: {
     thresholdPercent: 0.75,
   },
@@ -24,8 +24,6 @@ input token count plus an estimate of new messages. A smaller character estimate
 alone cannot satisfy compaction triggered by a higher provider count. If trimming
 cannot free enough space, eve summarizes the older history.
 
-Compaction also preserves the framework's own tool state automatically. It resets read-before-write tracking (so a write afterward re-reads the file whose read evidence was summarized away) and re-injects the active todo list, so the model keeps its task list across the summary. There is no per-tool hook to configure.
-
 First-class [memory](../memory) participates in a separate lifecycle. eve asks
 providers to capture before compaction, excludes attributed recalled records
 from the summarizer, keeps their canonical latest values, and recalls again

package/docs/concepts/execution-model-and-durability.mdx

@@ -64,7 +64,7 @@ For advanced self-hosted deployments, the root `agent.ts` can select the install
 import { defineAgent } from "eve";
 
 export default defineAgent({
-  model: "anthropic/claude-opus-4.8",
+  model: "anthropic/claude-opus-5.5",
   experimental: {
     workflow: {
       world: "@workflow/world-postgres",

package/docs/concepts/sessions-runs-and-streaming.md

@@ -77,7 +77,7 @@ The stream is newline-delimited JSON (NDJSON), one event per line:
 | `actions.requested`       | The model requested one or more actions, including tool calls; calls stream before execution.                                                      |
 | `action.partial`          | A locally executed tool generator yielded a preliminary output snapshot.                                                                           |
 | `action.result`           | A tool call returned.                                                                                                                              |
-| `input.requested`         | The run paused for human input ([HITL](/docs/human-in-the-loop) approval or `ask_question`); carries `requests`.                                   |
+| `input.requested`         | The run paused for human input ([HITL](/docs/human-in-the-loop) approval or a `ctx.ask()` question); carries `requests`.                           |
 | `input.resolved`          | The server accepted terminal human-input outcomes; carries `resolutions` with responses when provided.                                             |
 | `subagent.called`         | A subagent was delegated; carries `childSessionId` to attach to.                                                                                   |
 | `subagent.completed`      | The parent recorded a successful subagent invocation result; carries the actual output.                                                            |
@@ -210,7 +210,7 @@ Message sends default to `"steer"`. Before assistant output begins, eve interrup
 
 If the session is waiting on a human-in-the-loop approval, respond with the channel’s Approve or Cancel controls. Text messages do not decide an approval; unrelated text starts an ordinary turn while the approval stays pending and answerable. A later structured `inputResponses` answer keyed by its `requestId` still resumes the original tool call, even after intervening turns.
 
-With one question-only batch, an exact option match or permitted freeform response answers `ask_question`. Any other follow-up marks the question unanswered and starts the new turn. With several approval or question batches pending, eve does not guess which batch plain text addresses: the message starts an ordinary turn and the batches stay open. Use structured responses to target requests unambiguously.
+A pending `ctx.ask()` question from a blocking tool, such as `ask_question`, can be answered with plain text. When it is the only pending question, a message that matches an option answers it, and so does any message when the question allows free text. Otherwise eve does not guess which question the text addresses: the message follows the normal `turnPolicy`, questions created with `dismissible: true` resolve as `dismissed`, and other questions stay open. Questions from background tools and subagents always need a structured response. Use structured responses to target requests unambiguously.
 
 A structured response matches any currently pending request by ID, not only the newest batch. It becomes stale only after that request was answered, cleared, or cancelled. eve delivers a stale response to the model as a new user message, and the model decides whether the old selection still matters. A stale approval never authorizes the earlier tool call; the model must request the action and approval again if they are still needed.
 

package/docs/connections/overview.mdx

@@ -10,7 +10,7 @@ Connections live under `agent/connections/`. A static connection's runtime name
 
 ## MCP connections
 
-Use an MCP connection when the external service already exposes an MCP server. The server publishes its tools and schemas, and eve makes the matched tools callable by the model.
+Use an MCP connection when the external service already exposes an MCP server. The server publishes its tools and schemas, and eve makes the matched tools callable by the model. Each tool's JSON Schema reaches the model as published, minus any arguments you supply with `toolCall.providedArguments`, and eve checks the model's arguments against it before it calls the server.
 
 Read [MCP connections](/docs/connections/mcp) for `defineMcpClientConnection`, transport requirements, and MCP tool filters.
 

package/docs/evals/assertions.mdx

@@ -111,8 +111,8 @@ t.calledSubagent("weather", {
 
 ```ts
 const request = session.requireInputRequest({
-  toolName: "ask_question",
-  optionIds: ["red", "blue"],
+  toolName: "confirm_deploy",
+  optionIds: ["approve", "cancel"],
 });
 ```
 

package/docs/evals/judge.mdx

@@ -3,7 +3,7 @@ title: "Judge"
 description: "Grade evals with evaluation models using criteria, typed questions, or batches, and set thresholds on each assertion."
 ---
 
-Use `t.judge(...)` when a deterministic [assertion](./assertions) cannot capture what a good response looks like. It calls [`evaluate` from `eve/ai`](../guides/evaluate#evaluate-inside-a-tool) and defaults to `typesafe-ai/jev`. The judge is separate from the agent under test and is used only for scoring.
+Use `t.judge(...)` when a deterministic [assertion](./assertions) cannot capture what a good response looks like. It calls [`evaluate` from `eve/ai`](../guides/evaluate#evaluate-inside-a-tool) and defaults to `typesafe-ai/jev`, TypeSafe AI's [Jev evaluation model](https://vercel.com/i/what-is-jev). The judge is separate from the agent under test and is used only for scoring.
 
 ```ts
 import { defineEval } from "eve/evals";
@@ -138,14 +138,14 @@ String IDs resolve through the AI SDK default provider, or Vercel AI Gateway whe
 import { openai } from "@ai-sdk/openai";
 
 t.judge("The response addresses the request.", {
-  model: openai.evaluationModel("gpt-5.6-luna"),
+  model: openai.evaluationModel("gpt-6-luna"),
   modelOptions: { providerOptions: { openai: { reasoningEffort: "high" } } },
 });
 ```
 
 Use the provider's `evaluationModel` factory, not a language model instance. AI SDK supplies evaluation adapters for supported language models. Provider support and the evaluation contract are experimental; see [AI SDK evaluation](https://ai-sdk.dev/docs/ai-sdk-core/evaluation).
 
-String model IDs resolve through the default provider's evaluation API. Gateway accepts native evaluation models such as `typesafe-ai/jev`; passing a language model ID such as `openai/gpt-5.6-luna` does not automatically select the OpenAI adapter. Use an evaluation-model instance as shown above for language models.
+String model IDs resolve through the default provider's evaluation API. Gateway accepts native evaluation models such as `typesafe-ai/jev`; passing a language model ID such as `openai/gpt-6-luna` does not automatically select the OpenAI adapter. Use an evaluation-model instance as shown above for language models.
 
 Missing credentials, invalid answers, unsupported question types, and provider errors become failed gates even for tracked-only judgments. They do not silently skip the eval or select another model. Validation, authentication, and retries use the shared `evaluate` implementation.
 

package/docs/extensions.md

@@ -334,6 +334,10 @@ export default defineHook({
 
 `toolResultFrom` recognizes the mounted `crm__search` result from the original definition, not the namespaced string. Publishers should keep tool descriptions distinct so eve can assign each definition an unambiguous identity.
 
+### Bundled development extensions
+
+Local `eve dev` also mounts bundled development extensions without creating a project mount. The self-modification extension is included by default when `eve dev` starts a local server. Bundled development extensions are not included in production builds. See [Self-Modification](./guides/self-modification) for the local workflow.
+
 ### Compatibility
 
 At build time, eve checks the extension's generated capability metadata. If the extension needs an unsupported capability contract, upgrade eve or install a compatible extension release.

package/docs/getting-started.mdx

@@ -45,7 +45,7 @@ eve adds the missing `eve`, `ai`, and `zod` dependencies without changing files
 To initialize the agent with a different AI Gateway model or reasoning effort, pass `--model` or `--reasoning`:
 
 ```bash
-npx eve@latest init my-agent --model openai/gpt-5.6-terra --reasoning high
+npx eve@latest init my-agent --model openai/gpt-6-luna --reasoning high
 ```
 
 ## Run the agent
@@ -57,7 +57,7 @@ cd my-agent
 npm run dev
 ```
 
-This starts an interactive session where you can send messages to your agent. Edit `agent/instructions.md` to change its behavior and `agent/agent.ts` to configure its model. eve reloads your changes as you work.
+This starts an interactive session where you can send messages to your agent. Edit `agent/instructions.md` to change its behavior and `agent/agent.ts` to configure its model. eve reloads your changes as you work. You can also ask the agent to update its own instructions, tools, or skills; see [Self-Modification](./guides/self-modification).
 
 <span id="project-layout" />
 <span id="agent-directory-layout" />

package/docs/guides/deployment/vercel.mdx

@@ -155,6 +155,12 @@ eve dev https://your_agent.vercel.app/eve/support
 
 Set `VERCEL_AUTOMATION_BYPASS_SECRET` locally before connecting if the deployment uses Deployment Protection.
 
+## Inspect agent runs
+
+Use the **Agent Runs** tab under **Observability** to browse sessions and
+inspect conversation traces. See [Agent Runs](../../observability/agent-runs)
+for beta access, pricing, retention, and tracing setup.
+
 ## Continue configuring production
 
 Use these guides to secure and observe the deployed agent:

package/docs/guides/dev-tui.md

@@ -3,7 +3,7 @@ title: "Terminal UI"
 description: "Use eve locally or connect to a deployed agent from an interactive terminal UI."
 ---
 
-`eve dev` starts a local development server and opens an interactive terminal UI. Use it to talk to your agent, approve tool calls, answer its questions, and configure local development.
+`eve dev` starts a local development server and opens an interactive terminal UI. Use it to talk to your agent, approve tool calls, answer its questions, and configure local development. When `eve dev` starts a local server, self-modification is available by default; see [Self-Modification](./self-modification).
 
 ```bash
 eve dev
@@ -71,7 +71,7 @@ Changes apply together after the final choice, then the picker returns to chat.
 
 A successful login or model change takes effect on the next prompt.
 
-Gateway connections default to `spacexai/grok-4.7`; OpenAI and ChatGPT default to `gpt-5.6-luna-fast`; Anthropic defaults to `claude-sonnet-5`. An explicitly authored compatible model stays selected. If a new default is unavailable, eve offers the connection's available models. Dynamic or custom model expressions must be edited in `agent.ts`.
+Gateway connections default to `spacexai/grok-4.7`; OpenAI and ChatGPT default to `gpt-6-luna-fast`; Anthropic defaults to `claude-sonnet-5`. An explicitly authored compatible model stays selected. If a new default is unavailable, eve offers the connection's available models. Dynamic or custom model expressions must be edited in `agent.ts`.
 
 ## Add an integration
 

package/docs/guides/dynamic-capabilities.md

@@ -422,6 +422,8 @@ The caller's team gets its own playbook advertised as a loadable skill; everyone
 
 Skills follow the same naming rule as tools: a single `defineSkill(...)` is named after the file slug, while a map names each entry by its bare key (namespace the key yourself if it might collide). A dynamic skill overrides a same-named authored one; two dynamic resolvers emitting the same name throws.
 
+A dynamic skill that returns only `markdown` never starts a sandbox: eve keeps its instructions in session state and serves them from `load_skill`. When the skill also returns `files`, eve writes the package to the sandbox skill root when the resolver first returns it, and again only when its contents change or the session gets a new sandbox. A changed package replaces the previous directory, so files omitted from the new result are removed.
+
 ## Dynamic instructions
 
 A dynamic instructions file returns `defineInstructions({ content, role? })` built from the principal, tenant, channel, or external data. Omit `role` for system context:

package/docs/guides/evaluate.md

@@ -14,7 +14,8 @@ patch releases.
 
 ## Choose from Gateway models
 
-By default, `auto` evaluates with `typesafe-ai/jev`. Like other AI SDK
+By default, `auto` evaluates with `typesafe-ai/jev`, TypeSafe AI's
+[Jev evaluation model](https://vercel.com/i/what-is-jev). Like other AI SDK
 model strings, it uses Vercel AI Gateway unless the application has configured a
 different global default provider.
 
@@ -25,8 +26,8 @@ import { auto } from "eve/models";
 export default defineAgent({
   model: auto({
     options: {
-      "openai/gpt-5.6-sol": "Difficult reasoning and engineering tasks",
-      "openai/gpt-5.6-luna": "Routine tasks where fast completion matters",
+      "openai/gpt-6-sol": "Difficult reasoning and engineering tasks",
+      "openai/gpt-6-luna": "Routine tasks where fast completion matters",
     },
   }),
 });
@@ -38,7 +39,7 @@ Gateway evaluator uses the same connection selected through `/login` as Gateway
 language models. A configured AI SDK default provider still owns string model
 resolution during development. The TUI footer displays `dynamic model` when the
 agent uses `auto`, then adds the resolved model for the current turn, such as
-`dynamic model · openai/gpt-5.6-luna`.
+`dynamic model · openai/gpt-6-luna`.
 
 ## Use a provider directly
 
@@ -58,8 +59,8 @@ export default defineAgent({
   model: auto({
     model: typeSafeAi.evaluationModel("jev-latest"),
     options: {
-      "openai/gpt-5.6-sol": "Difficult reasoning and engineering tasks",
-      "openai/gpt-5.6-luna": "Routine tasks where fast completion matters",
+      "openai/gpt-6-sol": "Difficult reasoning and engineering tasks",
+      "openai/gpt-6-luna": "Routine tasks where fast completion matters",
     },
   }),
 });
@@ -83,7 +84,7 @@ export default defineAgent({
   reasoning: "medium",
   model: auto({
     options: {
-      "openai/gpt-5.6-sol": "Hard problems",
+      "openai/gpt-6-sol": "Hard problems",
       my_secret_model: {
         model: anthropic("sonnet-5"),
         description: "Routine work that can use the direct Anthropic provider",
@@ -141,7 +142,7 @@ export default defineTool({
 
 The choice above is typed as `"billing" | "support"`. Each question appears under
 its authored key in `result.answers`. Results also include token usage, warnings,
-provider metadata, and response metadata. To use that choice to delegate while keeping specialist subagents out of the parent model's tools, see [Route to a hidden subagent with JEV](/docs/tools/workflows#route-to-a-hidden-subagent-with-jev).
+provider metadata, and response metadata. To use that choice to delegate while keeping specialist subagents out of the parent model's tools, see [Route to a hidden subagent with Jev](/docs/tools/workflows#route-to-a-hidden-subagent-with-jev).
 
 `evaluate` accepts AI SDK evaluation options, including `maxRetries`, `headers`,
 and `providerOptions`. Pass an `abortSignal` to cancel the request. Input and
@@ -178,7 +179,9 @@ export default defineTool({
 The evaluation model reviews the tool name and input for dangerous effects. A
 caution, failed review, or incomplete input requires human approval. See
 [Human-in-the-loop approvals](/docs/human-in-the-loop#approvals) for classifier
-options and data handling.
+options and data handling, and the
+[Auto-approve tool calls with Jev](https://vercel.com/kb/guide/auto-approve-tool-calls-eve-jev)
+guide for an end-to-end walkthrough.
 
 ## Runtime behavior
 

package/docs/guides/frontend/overview.mdx

@@ -218,7 +218,7 @@ received attachment as a `file` part on the user message. The part includes `med
 
 ## Human-in-the-loop prompts
 
-Tools opt into approval with `approval`, and the model can also ask a question with `ask_question` — see [Human-in-the-loop](/docs/human-in-the-loop) for the server-side model. Either way the stream emits an `input.requested` event, and the pending request rides on a `dynamic-tool` part at `part.toolMetadata?.eve?.inputRequest`. Scan every message because an unrelated turn can add newer messages while an approval stays open, then answer through the same session with `respond()`:
+Tools opt into approval with `approval`, and workflow tools such as the opt-in `ask_question` tool ask the user a question with `ctx.ask()` — see [Human-in-the-loop](/docs/human-in-the-loop) for the server-side model. Either way the stream emits an `input.requested` event, and the pending request rides on a `dynamic-tool` part at `part.toolMetadata?.eve?.inputRequest`. Scan every message because an unrelated turn can add newer messages while an approval stays open, then answer through the same session with `respond()`:
 
 ```tsx
 const pendingRequests = agent.data.messages

package/docs/guides/hooks.md

@@ -88,7 +88,8 @@ failed stop, follows the normal
 
 For `subagent.called` and `subagent.completed`, `ctx.session.id` identifies the
 parent session. Typed handlers and `*` handlers receive this context even when
-the subagent event arrives between parent turns.
+the subagent event arrives between parent turns. These hooks can use
+`ctx.getSandbox()` against the parent session.
 
 ### Narrowing tool results
 
@@ -174,13 +175,15 @@ When a stream event fires, three things happen in order:
 
 1. Emit. The channel adapter handler runs, the event is stamped with its `meta` envelope, then it is written to the durable stream.
 2. Hooks. Stream-event hooks fire (typed handlers first, then the `*` wildcard). Return values are ignored.
-3. Dynamic tool resolvers. Resolvers subscribed to the event type run and update the tool set.
+3. Model preparation, for model lifecycle events. Dynamic resolvers subscribed to those events update the model context. Subagent notifications do not run model preparation.
 
 Hooks always run after the event is durably recorded, so if a hook throws, the stream stays consistent. The persisted event and every hook observe the same `meta.id`.
 
 ## What happens when a hook throws
 
-A thrown handler propagates through the emit composer and surfaces as `turn.failed`. In a conversation session, this includes handlers for `turn.started` and the first `step.started` of a model call: the failed turn ends with `session.waiting`, and the next message can start another turn. Task-mode boundary failures remain terminal. If a hook subscribed to a failure-cascade event also throws, it escalates to `session.failed`. For belt-and-suspenders semantics inside a hook, wrap the body in `try`/`catch`. eve treats a thrown hook as a real failure.
+A thrown handler during a model turn propagates through turn execution and surfaces as `turn.failed`. In a conversation session, this includes handlers for `turn.started` and the first `step.started` of a model call: the failed turn ends with `session.waiting`, and the next message can start another turn. Task-mode boundary failures remain terminal. If a hook subscribed to a failure-cascade event also throws, it escalates to `session.failed`. For belt-and-suspenders semantics inside a hook, wrap the body in `try`/`catch`. eve treats a thrown hook as a real failure.
+
+For `subagent.called` and `subagent.completed`, a thrown handler fails the notification step and follows the workflow runtime's step retry policy. A retry can publish the event again before rerunning its hooks. Parent execution waits for the notification step to finish or exhaust its retries.
 
 ## Subagent isolation
 

package/docs/guides/instrumentation/agent-runs.mdx

@@ -0,0 +1,101 @@
+---
+title: "Agent Runs"
+description: "Inspect eve sessions in Vercel and configure the Agent Runs destination."
+url: "/observability/agent-runs"
+---
+
+<Callout type="info" icon={<Flag aria-hidden="true" className="size-5" />}>
+  Agent Runs are available in [**Beta**](https://vercel.com/docs/release-phases#beta) on
+  [**Enterprise**](https://vercel.com/docs/plans/enterprise),
+  [**Pro**](https://vercel.com/docs/plans/pro), and [**Hobby**](https://vercel.com/docs/plans/hobby)
+  plans.
+</Callout>
+
+Use Vercel Agent Runs to browse eve sessions and inspect conversation traces
+in the project's **Observability** view. The tab requires enablement for your
+Vercel team; contact your Vercel representative if it does not appear.
+
+Agent Runs usage is billed at
+[Always-on Tracing rates](https://vercel.com/docs/tracing/always-on-tracing#usage-and-pricing)
+during beta. All plans include 30-day Agent Runs retention during beta.
+
+## Configure Agent Runs
+
+Preview and production deployments export to Vercel Agent Runs by default.
+Omitting `agent/instrumentation/agent-runs.ts` preserves that default. Export
+`agentRuns(...)` from that file to reconfigure it, or export
+`disableInstrumentation()` to disable it.
+
+## Enable tracing on Vercel
+
+New eve deployments automatically sample 100% of requests. Existing eve
+deployments need a sampling rule configured for the Vercel project to collect
+traces. To enable tracing for an existing deployment, select your team and
+project, then open
+[**Settings → Tracing**](https://vercel.com/d?to=%2F%5Bteam%5D%2F%5Bproject%5D%2Fsettings%2Ftracing&title=Go+to+Tracing+settings):
+
+1. Select **Add Sampling Rule**.
+2. Choose **All Environments**, **Production**, or **Preview**.
+3. Set **Rate** to `100%`.
+4. Leave **Request Path Prefix** blank to sample requests on every path.
+5. Select **Save**.
+
+Vercel applies the first matching rule. If an existing rule matches these
+requests, update it to use a 100% rate; a later rule will not override it.
+
+Tracing starts as soon as you save the rule. You can also manage sampling
+rules with [`vercel traces config`](https://vercel.com/docs/cli/traces#manage-trace-sampling-rules).
+The CLI accepts rates from 1 to 100%; add a 0% rule from the dashboard.
+
+## Filter Agent Runs exports
+
+`agentRuns()` accepts one `exportPolicy` object or an array applied in order.
+Each policy filters spans and attributes before the Agent Runs destination
+receives them:
+
+```ts title="agent/instrumentation/agent-runs.ts"
+import { agentRuns } from "eve/instrumentation/otel";
+
+export default agentRuns({
+  exportPolicy: {
+    span: ({ name }) => ({ emit: name !== "internal.cache.refresh" }),
+    attribute: ({ key }) => (key === "customer.id" ? { emit: false } : { emit: true }),
+  },
+});
+```
+
+Return `{ emit: true }` from `span` to retain a span unchanged, or
+`{ emit: false }` to omit it from Agent Runs. To retain the span while redacting
+content, return `{ redact: true }` with `inputs: true`, `outputs: true`, or
+both. A redaction decision implies emission and requires at least one
+direction. A throwing `span` callback drops the span from Agent Runs. Return
+`{ emit: true }`, `{ emit: false }`, or `{ replace: true, value }` from
+`attribute` to retain, remove, or change one attribute.
+
+## Redact content
+
+The `exportPolicy` can further narrow what Agent Runs receives after the
+process-wide `otel({ tracePolicy })` has admitted a trace. For example, keep
+spans while redacting inputs and outputs unless the session is public:
+
+```ts title="agent/instrumentation/agent-runs.ts"
+import { agentRuns } from "eve/instrumentation/otel";
+
+export default agentRuns({
+  exportPolicy: {
+    span: ({ audience }) =>
+      audience === "public" ? { emit: true } : { redact: true, inputs: true, outputs: true },
+  },
+});
+```
+
+Input redaction removes eve's known prompt, instruction, document, and
+tool-argument attributes. Output redaction removes response, reasoning,
+tool-result, exception, and status attributes. Redaction narrows only Agent
+Runs; it does not mutate spans sent to another destination.
+
+## What to read next
+
+- [OpenTelemetry](./otel): configure process-wide settings and third-party destinations.
+- [Vercel deployment](/docs/guides/deployment/vercel): deploy an eve agent and inspect its runtime.
+- [Migrate instrumentation](/docs/observability/instrumentation-migration): replace `agent/instrumentation.ts`.