openmerit 0.3.0 → 0.3.2

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.3.2 — 2026-11-05
+
+- Fix Claude Code project-scope installation for a project inside a larger git repository, such as a monorepo package. Claude Code keeps its private MCP entry under the repository root, so the installer reported "Claude Code did not save the openmerit MCP server", `status` reported it missing and a rerun clashed with it. The installer now reads the entry where Claude Code keeps it and says that the MCP server applies to the whole repository; the skill and rule stay in the project.
+- Make device approvals easier to relay. The installer prints the code's expiry in UTC with the minutes left instead of an unlabelled local time. When it returns with exit code 10, the message includes the approval link and expiry, `--json` adds `approvalUrl`, `userCode` and `expiresAt`, and the installed Scout instructions tell the agent to show the link and expiry to the person approving. The matching Scout serving release extends device codes from 10 to 30 minutes.
+
+## 0.3.1 — 2026-11-05
+
+- `openmerit scout --conversation <id> --results <file.json>` reports aggregate results you measured with your own evals. Scout then ranks them above public benchmarks for your recommendations only. Terminal output shows each task's evaluation plan, its agent and calls per run, and marks quality that rests on measured results.
+- The installed Scout instructions for Claude Code and Pi tell the agent to measure shortlisted candidates as extra runs of the project's existing evals, never a second eval, and to report aggregates only. This needs the matching hosted Scout release, whose `scout_models` tool returns `tasks[].evaluation` and accepts `results`.
+
 ## 0.3.0 — 2026-11-04
 
 - **Breaking:** retire the standalone `openmerit eval` CLI, including `init`, `doctor`, `plan`, `run`, `watch` and `report`. Existing local evaluation files are not deleted. The hosted `/scout/api/recommend` retirement ships separately with the Scout serving Worker.

package/README.md

@@ -24,7 +24,7 @@ Package publication and website deployment are separate releases. Pi setup requi
 
 ## Evaluation CLI retirement — npm 0.3.0
 
-OpenMerit 0.3.0 removes the standalone `openmerit eval` commands, including planning. Existing local evaluation files are not deleted. The old Scout JSON recommendation API is retired in source and goes offline with a separate Scout deployment. Scout's conversational agent, MCP, coding-agent setup and catalogue publication remain available; see [retirement details](docs/agent-evaluations.md).
+OpenMerit 0.3.0 removes the standalone `openmerit eval` commands, including planning. Existing local evaluation files are not deleted. The old Scout JSON recommendation API went offline with the Scout serving deploy of 5 November 2026. Scout's conversational agent, MCP, coding-agent setup and catalogue publication remain available; see [retirement details](docs/agent-evaluations.md).
 
 ## Inference Terminal retirement — npm 0.3.0
 

package/dist/pi/scout-extension.mjs

@@ -117,6 +117,26 @@ Compare quality evidence, reference cost/time and watch-outs. Prefer the cheapes
 usable option the evidence supports, without claiming benchmarks prove
 application accuracy. Use the project's actual SDK and protocol.
 
+## Measuring candidates
+
+Each task also has an \`evaluation\`: the checks (per model call, agent run or
+end-to-end outcome), labelled examples per candidate and when to stop a losing
+candidate. Offer to run it before the user commits to a model. Use the evals the
+project already has, such as datasets and experiments in its eval or tracing
+platform, or its own eval scripts and fixtures: add the shortlisted candidates as
+extra runs of those evals. Do not build a second eval or copy their datasets.
+If the project has no eval for the task, propose a minimal one built from its
+existing traces and the decisions people already make, and create it only where
+and how the user agrees. Ask before spending on model calls or writing to an
+eval platform.
+
+Report what you measured with \`scout_models\`: the same \`conversationId\` and
+\`results\`, one aggregate per model and metric (\`level\`, \`task\` title, \`scope\`,
+\`model\` endpoint or model ID, \`metric\`, \`value\`, \`higherIsBetter\`,
+\`samples\`, \`source\`). Send aggregates only, never inputs, outputs or records.
+Scout ranks them above public benchmarks for this user only; models whose
+\`quality.measured\` is true rest on them.
+
 ## Boundaries
 
 Ask before adding providers or credentials, spending, deploying or external
@@ -127,7 +147,9 @@ own model.
 If Scout fails or has no suitable candidate/evidence for the required deployment,
 say so and pause the model decision; do not substitute a remembered recommendation
 or SDK default. Continue only model-independent work. Local/browser models are not
-exempt when the catalogue lacks coverage. If it reports that the device key is missing,
+exempt when the catalogue lacks coverage. When an install or login waits for approval (exit code
+10), show the user the approval link and its UTC expiry in your reply, not only in
+command output, then rerun the same command once they approve. If it reports that the device key is missing,
 expired or revoked, run \`npx -y openmerit@latest login\` and then
 \`npx -y openmerit@latest install\` again. Never ask the user to paste a key.
 `;

package/dist/terminal/scout-update.json

@@ -1 +1 @@
-{"format":1,"contract":"scout-recommend-v1","version":"0.3.0","skill":"---\nname: openmerit-scout\ndescription: Consults OpenMerit Scout before choosing, adding, downloading, replacing or configuring any application inference model, including implicit choices while building features, pretrained browser/local models, SDK defaults, embeddings, OCR, speech, classification, detection, segmentation, LLMs and agents. Not for unrelated edits, ordinary inference calls or this coding agent's own model.\n---\n\n# OpenMerit · Scout\n\nBefore introducing, recommending, replacing or configuring an application\ninference model, call the `scout_models` tool on the `openmerit` MCP server.\nThis applies even when the user never mentions models or Scout. Downloading local\nweights, using a pretrained browser model or accepting an SDK default is a model\ndecision: consult Scout before model-specific code, installs or downloads.\nPreserve the user's explicit model choice; assess it rather than silently replacing\nit. Leave unchanged, approved application selections in place.\n\n## Project policy\n\nRead `.openmerit/scout.json` before searching. If it is absent, ask once whether this\nproject uses **existing-only** (only endpoints it can already reach) or\n**existing-first** (prefer them, ask before searching the whole catalogue), and\nwhich public catalogue endpoint IDs it can access. Save the answer:\n\n```sh\nnpx -y openmerit@latest project --policy existing-only --endpoint-id <id> ...\n```\n\nNever infer access from a model name, SDK or environment-variable name. Never\nrecord credentials, private URLs, prompts or customer data. An empty inventory\nis not permission to search all models.\n\n## Searching\n\nCall `scout_models` with a non-confidential task description as `prompt` and\n`endpointIds` set to the saved inventory, including `[]` when empty. Send\nonly task requirements, never source files, secrets or real customer records.\nIf the MCP tool is unavailable, run\n`npx -y openmerit@latest scout \"<task>\" --json`; it applies the saved inventory.\n\nScout is a conversation with the agent behind openmerit.ai/scout. When\n`status` is `needs_answers`, fill in what the user or project already states,\nshow the user Scout's brief and remaining questions, and wait for their answers.\nThen call again with the returned `conversationId` and `brief` set to the\nrewritten components, or `skip: true` if the user wants Scout to assume. Never\nanswer Scout's questions with guesses. Send later refinements to the same\n`conversationId`. A call with only a `conversationId` reads an interrupted turn.\n\nWhen `status` is `recommended`, Scout has split the work into `tasks`, one\nper model, each with up to five ranked `models`; cover every task. Each\nmodel's `cost` names the `endpointId` and `providerModelId` to call.\nExisting-only never broadens. For existing-first, if no existing candidate fits a\ntask, explain the gap and ask before a new conversation without `endpointIds`.\nCompare quality evidence, reference cost/time and watch-outs. Prefer the cheapest\nusable option the evidence supports, without claiming benchmarks prove\napplication accuracy. Use the project's actual SDK and protocol.\n\n## Boundaries\n\nAsk before adding providers or credentials, spending, deploying or external\nwrites. Report recommendation, endpoint connectivity and application checks as\nseparate results, each only after it succeeds. Do not switch this coding agent's\nown model.\n\nIf Scout fails or has no suitable candidate/evidence for the required deployment,\nsay so and pause the model decision; do not substitute a remembered recommendation\nor SDK default. Continue only model-independent work. Local/browser models are not\nexempt when the catalogue lacks coverage. If it reports that the device key is missing,\nexpired or revoked, run `npx -y openmerit@latest login` and then\n`npx -y openmerit@latest install` again. Never ask the user to paste a key.\n","rule":"# Application model decisions · OpenMerit Scout\n\nBefore choosing, recommending, adding, downloading, replacing or configuring\nANY application inference model, load the `openmerit-scout` skill and consult\n`scout_models` on the `openmerit` MCP server before committing to a model.\nThis applies even when the user only asks you to build a feature, never mentions\nmodels or Scout, or names a model themselves (assess it; do not silently replace it).\n\nA pretrained model, local weights, browser inference, an SDK's default model,\nembeddings, OCR, speech, image classification, detection and segmentation all\ncount as model choices, not implementation details. Identify this decision before\nwriting model-specific code, installing a model package or downloading weights.\n\nFollow the skill's project access policy. If it is missing, ask and save it first.\nScout may answer with clarifying questions: relay them to the user and continue\nthe same Scout conversation with their answers before choosing.\nIf Scout is unavailable, fails or has no suitable evidence/candidate, disclose the\ngap and pause the model decision; do not fall back to a remembered or SDK-default\nmodel. You may continue work that does not commit to a model.\n\nDo not consult Scout for unrelated edits, normal inference calls using an unchanged\napproved model, or this coding agent's own model. Do not reselect approved models\nunless the task introduces or changes a model decision.\n\nWhen the user says \"update openmerit\", run\n`npx -y openmerit@latest update --scope user` (use `--scope project` if this\nrule is in the project's .claude/rules directory). Report the result, including\nany conflict or manual-upgrade requirement. Do not delete edited files, bypass\npermissions or rerun login automatically. A new session loads updated instructions.\n"}
+{"format":1,"contract":"scout-recommend-v1","version":"0.3.2","skill":"---\nname: openmerit-scout\ndescription: Consults OpenMerit Scout before choosing, adding, downloading, replacing or configuring any application inference model, including implicit choices while building features, pretrained browser/local models, SDK defaults, embeddings, OCR, speech, classification, detection, segmentation, LLMs and agents. Not for unrelated edits, ordinary inference calls or this coding agent's own model.\n---\n\n# OpenMerit · Scout\n\nBefore introducing, recommending, replacing or configuring an application\ninference model, call the `scout_models` tool on the `openmerit` MCP server.\nThis applies even when the user never mentions models or Scout. Downloading local\nweights, using a pretrained browser model or accepting an SDK default is a model\ndecision: consult Scout before model-specific code, installs or downloads.\nPreserve the user's explicit model choice; assess it rather than silently replacing\nit. Leave unchanged, approved application selections in place.\n\n## Project policy\n\nRead `.openmerit/scout.json` before searching. If it is absent, ask once whether this\nproject uses **existing-only** (only endpoints it can already reach) or\n**existing-first** (prefer them, ask before searching the whole catalogue), and\nwhich public catalogue endpoint IDs it can access. Save the answer:\n\n```sh\nnpx -y openmerit@latest project --policy existing-only --endpoint-id <id> ...\n```\n\nNever infer access from a model name, SDK or environment-variable name. Never\nrecord credentials, private URLs, prompts or customer data. An empty inventory\nis not permission to search all models.\n\n## Searching\n\nCall `scout_models` with a non-confidential task description as `prompt` and\n`endpointIds` set to the saved inventory, including `[]` when empty. Send\nonly task requirements, never source files, secrets or real customer records.\nIf the MCP tool is unavailable, run\n`npx -y openmerit@latest scout \"<task>\" --json`; it applies the saved inventory.\n\nScout is a conversation with the agent behind openmerit.ai/scout. When\n`status` is `needs_answers`, fill in what the user or project already states,\nshow the user Scout's brief and remaining questions, and wait for their answers.\nThen call again with the returned `conversationId` and `brief` set to the\nrewritten components, or `skip: true` if the user wants Scout to assume. Never\nanswer Scout's questions with guesses. Send later refinements to the same\n`conversationId`. A call with only a `conversationId` reads an interrupted turn.\n\nWhen `status` is `recommended`, Scout has split the work into `tasks`, one\nper model, each with up to five ranked `models`; cover every task. Each\nmodel's `cost` names the `endpointId` and `providerModelId` to call.\nExisting-only never broadens. For existing-first, if no existing candidate fits a\ntask, explain the gap and ask before a new conversation without `endpointIds`.\nCompare quality evidence, reference cost/time and watch-outs. Prefer the cheapest\nusable option the evidence supports, without claiming benchmarks prove\napplication accuracy. Use the project's actual SDK and protocol.\n\n## Measuring candidates\n\nEach task also has an `evaluation`: the checks (per model call, agent run or\nend-to-end outcome), labelled examples per candidate and when to stop a losing\ncandidate. Offer to run it before the user commits to a model. Use the evals the\nproject already has, such as datasets and experiments in its eval or tracing\nplatform, or its own eval scripts and fixtures: add the shortlisted candidates as\nextra runs of those evals. Do not build a second eval or copy their datasets.\nIf the project has no eval for the task, propose a minimal one built from its\nexisting traces and the decisions people already make, and create it only where\nand how the user agrees. Ask before spending on model calls or writing to an\neval platform.\n\nReport what you measured with `scout_models`: the same `conversationId` and\n`results`, one aggregate per model and metric (`level`, `task` title, `scope`,\n`model` endpoint or model ID, `metric`, `value`, `higherIsBetter`,\n`samples`, `source`). Send aggregates only, never inputs, outputs or records.\nScout ranks them above public benchmarks for this user only; models whose\n`quality.measured` is true rest on them.\n\n## Boundaries\n\nAsk before adding providers or credentials, spending, deploying or external\nwrites. Report recommendation, endpoint connectivity and application checks as\nseparate results, each only after it succeeds. Do not switch this coding agent's\nown model.\n\nIf Scout fails or has no suitable candidate/evidence for the required deployment,\nsay so and pause the model decision; do not substitute a remembered recommendation\nor SDK default. Continue only model-independent work. Local/browser models are not\nexempt when the catalogue lacks coverage. When an install or login waits for approval (exit code\n10), show the user the approval link and its UTC expiry in your reply, not only in\ncommand output, then rerun the same command once they approve. If it reports that the device key is missing,\nexpired or revoked, run `npx -y openmerit@latest login` and then\n`npx -y openmerit@latest install` again. Never ask the user to paste a key.\n","rule":"# Application model decisions · OpenMerit Scout\n\nBefore choosing, recommending, adding, downloading, replacing or configuring\nANY application inference model, load the `openmerit-scout` skill and consult\n`scout_models` on the `openmerit` MCP server before committing to a model.\nThis applies even when the user only asks you to build a feature, never mentions\nmodels or Scout, or names a model themselves (assess it; do not silently replace it).\n\nA pretrained model, local weights, browser inference, an SDK's default model,\nembeddings, OCR, speech, image classification, detection and segmentation all\ncount as model choices, not implementation details. Identify this decision before\nwriting model-specific code, installing a model package or downloading weights.\n\nFollow the skill's project access policy. If it is missing, ask and save it first.\nScout may answer with clarifying questions: relay them to the user and continue\nthe same Scout conversation with their answers before choosing.\nIf Scout is unavailable, fails or has no suitable evidence/candidate, disclose the\ngap and pause the model decision; do not fall back to a remembered or SDK-default\nmodel. You may continue work that does not commit to a model.\n\nDo not consult Scout for unrelated edits, normal inference calls using an unchanged\napproved model, or this coding agent's own model. Do not reselect approved models\nunless the task introduces or changes a model decision.\n\nWhen the user says \"update openmerit\", run\n`npx -y openmerit@latest update --scope user` (use `--scope project` if this\nrule is in the project's .claude/rules directory). Report the result, including\nany conflict or manual-upgrade requirement. Do not delete edited files, bypass\npermissions or rerun login automatically. A new session loads updated instructions.\n"}

package/dist/terminal/src/install.mjs

@@ -50,7 +50,9 @@ Exit codes: 0 done, 10 approval pending, 11 denied or expired, 20 conflict,
 30 OpenMerit unreachable, 1 other error.
 `;
 
-class Stop extends Error { constructor(message, code, next) { super(message); this.exit = code; this.next = next; } }
+class Stop extends Error { constructor(message, code, next, details) { super(message); this.exit = code; this.next = next; this.details = details; } }
+// An approval deadline people in any timezone, and agents relaying it, read the same way.
+const deadlineText = at => `${new Date(at).toISOString().slice(11, 16)} UTC, in ${Math.max(0, Math.round((at - Date.now()) / 60_000))} min`;
 
 const server = () => (process.env.OPENMERIT_URL ?? 'https://openmerit.ai').replace(/\/+$/, '');
 const credentialsPath = telemetry => deviceCredentialsPath(telemetry.properties.agent, telemetry.properties.scope, telemetry.root);
@@ -105,7 +107,13 @@ function claudePaths(root, scope) {
     mcpScope: scope === 'user' ? 'user' : 'local',
   };
 }
-const configuredServer = (config, root, scope) => scope === 'user' ? config.mcpServers?.openmerit : config.projects?.[root]?.mcpServers?.openmerit;
+// Claude Code keys its private local scope by the project's git repository root, so a
+// project in a subdirectory of a repository shares the entry of the whole repository.
+async function claudeProject(root) {
+  try { return (await exec('git', ['rev-parse', '--show-toplevel'], { cwd: root, timeout: 10_000, windowsHide: true })).stdout.trim() || root; }
+  catch { return root; }
+}
+const configuredServer = (config, project, scope) => scope === 'user' ? config.mcpServers?.openmerit : config.projects?.[project]?.mcpServers?.openmerit;
 const serverConfig = (key, telemetry) => ({ type: 'http', url: `${server()}/scout/mcp`, headers: { Authorization: `Bearer ${key}`, ...setupHeaders(telemetry.properties, 'agent') } });
 const canonical = value => Array.isArray(value) ? value.map(canonical) : value && typeof value === 'object' ? Object.fromEntries(Object.keys(value).sort().map(k => [k, canonical(value[k])])) : value;
 const serverHash = value => value ? hash(JSON.stringify(canonical(value))) : null;
@@ -123,12 +131,12 @@ async function claude(args, root, key) {
 // Inspect the current Claude Code installation without changing anything.
 async function inspect(root, scope, client = 'claude-code') {
   if (client === 'pi') return { ...await inspectPi(root, scope), client };
-  const paths = claudePaths(root, scope);
+  const paths = { ...claudePaths(root, scope), project: await claudeProject(root) };
   const [record, config, installedSkill, installedRule, legacyRule] = await Promise.all([object(paths.record), object(paths.mcpFile), optional(paths.skill), optional(paths.rule), optional(paths.legacyRule)]);
   const updates = await inspectUpdates(dirname(paths.record), scope, record);
   const owned = [2, 3].includes(record.version) && record.scope === scope;
-  const current = configuredServer(config, root, scope);
-  const otherScope = configuredServer(config, root, scope === 'user' ? 'project' : 'user');
+  const current = configuredServer(config, paths.project, scope);
+  const otherScope = configuredServer(config, paths.project, scope === 'user' ? 'project' : 'user');
   const projectFile = await object(join(root, '.mcp.json'));
   return {
     paths, record, owned, current, installedSkill, installedRule, updates, client,
@@ -189,7 +197,7 @@ async function login({ client, scope, open, log, wait, telemetry }) {
   if (process.env.OPENMERIT_ANALYTICS_ACCOUNT === 'off' || !telemetryEnabled(true)) approvalURL.searchParams.set('analytics_account', 'off');
   pending = { ...pending, url: approvalURL.href, setup_id: telemetry.properties.setup_id, installation_id: telemetry.properties.installation_id };
   await save(pendingPath(telemetry), pending);
-  log(`Approve this device in your browser (opening it now):\n\n  ${pending.url}\n\n  Code: ${pending.userCode} (expires ${new Date(pending.expiresAt).toLocaleTimeString()})\n`);
+  log(`Approve this device in your browser (opening it now):\n\n  ${pending.url}\n\n  Code: ${pending.userCode} (expires ${deadlineText(pending.expiresAt)})\n`);
   await telemetry.step('browser_open', 'started');
   if (open && !pending.opened && !pending.openFailed) {
     // Only the approval page on the configured OpenMerit server may be opened.
@@ -226,7 +234,9 @@ async function login({ client, scope, open, log, wait, telemetry }) {
       if (data.error === 'access_denied') throw new Stop('The sign-in request was denied. Nothing was installed.', exitCodes.denied, 'Rerun the command to start a new request.');
       throw new Stop('The sign-in code expired or was already used. Nothing was installed.', exitCodes.denied, 'Rerun the command to get a new code.');
     }
-    if (Date.now() + interval > deadline) throw new Stop(`Waiting for approval of code ${pending.userCode}.`, exitCodes.pending, `Open ${pending.url}, click Approve, then rerun the same command. It resumes this request.`);
+    if (Date.now() + interval > deadline) throw new Stop(`Waiting for approval of code ${pending.userCode}: ${pending.url} (expires ${deadlineText(pending.expiresAt)}).`, exitCodes.pending,
+      `Show the person approving this link and its expiry in your reply. Once they click Approve, rerun the same command: it resumes this request.`,
+      { approvalUrl: pending.url, userCode: pending.userCode, expiresAt: new Date(pending.expiresAt).toISOString() });
     await new Promise(resolve => setTimeout(resolve, interval));
   }
 }
@@ -251,7 +261,7 @@ async function install(root, { client, scope, plan, open, log, wait, telemetry,
   const found = conflicts(state);
   if (client === 'pi' && autoUpdate === 'on') throw new Stop('Pi updates are explicit: rerun install --client pi. Automatic hooks are Claude Code only.', exitCodes.error);
   const changes = [
-    `${state.current ? 'Update' : 'Add'} ${client === 'pi' ? 'native Pi MCP extension' : 'MCP server "openmerit" in Claude Code'} (${scope} scope)`,
+    `${state.current ? 'Update' : 'Add'} ${client === 'pi' ? 'native Pi MCP extension' : 'MCP server "openmerit" in Claude Code'} (${scope} scope)${client !== 'pi' && scope === 'project' && state.paths.project !== root ? `; Claude Code applies it to the whole git repository at ${state.paths.project}` : ''}`,
     `${state.installedSkill === null ? 'Create' : 'Update'} ${state.paths.skill}`,
     `${state.installedRule === null ? 'Create' : 'Update'} ${state.paths.rule} (loaded at session startup)`,
     ...(client === 'pi' ? [] : [
@@ -293,7 +303,7 @@ async function install(root, { client, scope, plan, open, log, wait, telemetry,
     await save(state.paths.rule, rule);
     if (state.legacyRule === 'remove') await rm(state.paths.legacyRule, { force: true });
     // Record what Claude actually wrote, so later edits are detected precisely.
-    const written = configuredServer(await object(state.paths.mcpFile), root, scope);
+    const written = configuredServer(await object(state.paths.mcpFile), state.paths.project, scope);
     if (!written) throw new Stop('Claude Code did not save the "openmerit" MCP server.', exitCodes.error, 'Run "openmerit doctor".');
     const updates = await configureUpdates(dirname(state.paths.record), scope, state.record, clientVersion, enabled);
     await save(state.paths.record, { version: 3, scope, client, ...ids, serverHash: serverHash(written), skillHash: hash(skill), ruleHash: hash(rule), updates, installedAt: new Date().toISOString() });
@@ -509,7 +519,7 @@ async function runMain(args, { root = process.cwd(), print = console.log, printE
   } catch (error) {
     if (!(error instanceof Stop)) throw error;
     await telemetry.fail(error);
-    if (options.json) print(JSON.stringify({ ok: false, action, code: Object.keys(exitCodes).find(k => exitCodes[k] === error.exit), message: error.message, next: error.next }, null, 2));
+    if (options.json) print(JSON.stringify({ ok: false, action, code: Object.keys(exitCodes).find(k => exitCodes[k] === error.exit), message: error.message, next: error.next, ...error.details }, null, 2));
     else printError([error.message, error.next].filter(Boolean).join('\n'));
     return error.exit;
   }

package/dist/terminal/src/scout-instructions.js

@@ -83,6 +83,26 @@ Compare quality evidence, reference cost/time and watch-outs. Prefer the cheapes
 usable option the evidence supports, without claiming benchmarks prove
 application accuracy. Use the project's actual SDK and protocol.
 
+## Measuring candidates
+
+Each task also has an \`evaluation\`: the checks (per model call, agent run or
+end-to-end outcome), labelled examples per candidate and when to stop a losing
+candidate. Offer to run it before the user commits to a model. Use the evals the
+project already has, such as datasets and experiments in its eval or tracing
+platform, or its own eval scripts and fixtures: add the shortlisted candidates as
+extra runs of those evals. Do not build a second eval or copy their datasets.
+If the project has no eval for the task, propose a minimal one built from its
+existing traces and the decisions people already make, and create it only where
+and how the user agrees. Ask before spending on model calls or writing to an
+eval platform.
+
+Report what you measured with \`scout_models\`: the same \`conversationId\` and
+\`results\`, one aggregate per model and metric (\`level\`, \`task\` title, \`scope\`,
+\`model\` endpoint or model ID, \`metric\`, \`value\`, \`higherIsBetter\`,
+\`samples\`, \`source\`). Send aggregates only, never inputs, outputs or records.
+Scout ranks them above public benchmarks for this user only; models whose
+\`quality.measured\` is true rest on them.
+
 ## Boundaries
 
 Ask before adding providers or credentials, spending, deploying or external
@@ -93,7 +113,9 @@ own model.
 If Scout fails or has no suitable candidate/evidence for the required deployment,
 say so and pause the model decision; do not substitute a remembered recommendation
 or SDK default. Continue only model-independent work. Local/browser models are not
-exempt when the catalogue lacks coverage. If it reports that the device key is missing,
+exempt when the catalogue lacks coverage. When an install or login waits for approval (exit code
+10), show the user the approval link and its UTC expiry in your reply, not only in
+command output, then rerun the same command once they approve. If it reports that the device key is missing,
 expired or revoked, run \`npx -y openmerit@latest login\` and then
 \`npx -y openmerit@latest install\` again. Never ask the user to paste a key.
 `;

package/dist/terminal/src/scout.js

@@ -3,7 +3,7 @@
 // the same scout_models MCP tool with this device's key, so every client gets the
 // same clarifying questions and per-task shortlists.
 import { readFile } from 'node:fs/promises';
-import { join } from 'node:path';
+import { join, resolve } from 'node:path';
 import { credentialsPath, readCredentials } from "./scout-credentials.js";
 const policyFile = '.openmerit/scout.json';
 export const components = ['input', 'work', 'output', 'criteria', 'constraints'];
@@ -11,12 +11,17 @@ const help = `OpenMerit · Scout
 
   openmerit scout "<the work models need to do>" [--all-endpoints] [--json]
   openmerit scout --conversation <id> [--brief <component>=<text> ...] [--skip] ["<follow-up>"] [--json]
+  openmerit scout --conversation <id> --results <file.json> [--json]
 
 Asks Scout, the agent behind openmerit.ai/scout. Scout may first return a task
 brief with clarifying questions: answer them with --brief (components: input,
 work, output, criteria, constraints) in the same conversation, or --skip to let
 Scout assume. It then splits the work into tasks and recommends up to five
-models for each. A turn usually takes one to three minutes.
+models for each, with an evaluation: how to measure the candidates with your
+own evals. Report what you measured with --results: a JSON list of aggregate
+results ({level, task?, scope, model, metric, value, higherIsBetter, samples,
+source}), never inputs or outputs. Scout ranks them above public benchmarks, for
+your recommendations only. A turn usually takes one to three minutes.
 
 A new conversation is limited to the endpoint IDs in ${policyFile}
 (set with openmerit project). --all-endpoints searches the whole catalogue and is
@@ -114,7 +119,7 @@ export function renderScout(result) {
         const tasks = result.tasks ?? [];
         tasks.forEach((task, i) => {
             if (tasks.length > 1)
-                lines.push(`${i + 1}. ${task.title}`, `   ${task.work}`);
+                lines.push(`${i + 1}. ${task.title}${task.agent ? ` (${task.agent}${task.callsPerRun && task.callsPerRun > 1 ? `, ${task.callsPerRun} calls per run` : ''})` : ''}`, `   ${task.work}`);
             // As on the page, the default cheapest-first order is unlabelled; another priority is named with the value it sorts by.
             const priority = task.priority ?? 'cost';
             if (priority !== 'cost')
@@ -123,14 +128,20 @@ export function renderScout(result) {
                 const price = m.cost ? `${money(m.cost.inputPerMillion)} in / ${money(m.cost.outputPerMillion)} out per 1M via ${m.cost.route} (${m.cost.endpointId})` : 'no priced route';
                 const metric = priority === 'latency' ? ` · ${m.latency ? `${m.latency.seconds.toFixed(1)}s per task` : 'time unknown'}`
                     : priority === 'recency' ? ` · ${m.recency ? `released ${m.recency.released}` : 'release date unknown'}` : '';
-                lines.push(`   ${m.rank}. ${m.name}${metric} · quality ${m.quality.score}/100 · ${price}`, `      ${m.rationale}`);
+                lines.push(`   ${m.rank}. ${m.name}${metric} · quality ${m.quality.score}/100${m.quality.measured ? ' (measured)' : ''} · ${price}`, `      ${m.rationale}`);
             }
-            lines.push(`   ${task.summary}`, '');
+            lines.push(`   ${task.summary}`);
+            if (task.evaluation)
+                lines.push(`   Measure with your evals (${task.evaluation.samples} labelled examples per candidate; ${task.evaluation.stopEarly}):`, ...task.evaluation.checks.map(c => `     - ${c.level}: ${c.metric}. ${c.how}`));
+            lines.push('');
         });
         if (result.assumptions?.length)
             lines.push('Assumptions:', ...result.assumptions.map(a => `  - ${a}`), '');
         lines.push(`Refine: openmerit scout --conversation ${result.conversationId} "<follow-up>"`);
-        lines.push('Quality is Scout\'s estimate from public benchmarks, not measured accuracy for your application.');
+        lines.push(`Report measured results: openmerit scout --conversation ${result.conversationId} --results <file.json>`);
+        lines.push(tasks.some(t => t.models.some(m => m.quality.measured))
+            ? 'Quality marked (measured) rests on the results you reported; the rest is Scout\'s estimate from public benchmarks.'
+            : 'Quality is Scout\'s estimate from public benchmarks, not measured accuracy for your application.');
     }
     else if (result.status === 'working')
         lines.push(`Scout is still working. Read the outcome with: openmerit scout --conversation ${result.conversationId}`);
@@ -164,6 +175,19 @@ export async function main(args, { root = process.cwd(), print = console.log } =
             request.allEndpoints = true;
         else if (arg === '--json')
             json = true;
+        else if (arg === '--results') {
+            const file = value();
+            let results;
+            try {
+                results = JSON.parse(await readFile(resolve(root, file), 'utf8'));
+            }
+            catch {
+                throw new Error(`Could not read results from ${file}: it must be a JSON list.`);
+            }
+            if (!Array.isArray(results))
+                throw new Error(`${file} must hold a JSON list of results.`);
+            request.results = results;
+        }
         else if (arg.startsWith('-'))
             throw new Error(`Unknown option for scout: ${arg}`);
         else
@@ -175,6 +199,8 @@ export async function main(args, { root = process.cwd(), print = console.log } =
         print(help);
         return 1;
     }
+    if (request.results && !request.conversationId)
+        throw new Error('--results continues a conversation: pass --conversation <id> from the recommendation you measured.');
     const result = await scoutTurn(root, request);
     print(json ? JSON.stringify(result, null, 2) : renderScout(result));
     return result.status === 'error' ? 1 : 0;

package/docs/agent-evaluations.md

@@ -1,6 +1,6 @@
 # Agent-evaluation CLI retirement
 
-**npm 0.3.0, 4 November 2026.** The standalone `openmerit eval` CLI is retired in full, including `init`, `doctor`, `plan`, `run`, `watch` and `report`. Older installed packages keep it until they are upgraded. The old Scout JSON recommendation API is retired in unreleased source and goes offline only with a separate Scout deployment.
+**npm 0.3.0, 4 November 2026.** The standalone `openmerit eval` CLI is retired in full, including `init`, `doctor`, `plan`, `run`, `watch` and `report`. Older installed packages keep it until they are upgraded. The old Scout JSON recommendation API went offline with the Scout serving deploy of 5 November 2026.
 
 After an approved Scout release, `/scout/api/recommend` and its legacy alpha/versioned paths return 404, regardless of authentication. There is no replacement frozen-plan protocol. Use the [Scout conversational agent or MCP](scout.md) for provisional model guidance and your own representative task checks for evaluation. The Inference Terminal is [retired as well](inference-terminal.md); the Pi adapter remains a native Scout MCP integration, not an evaluation runner.
 

package/docs/commands.md

@@ -4,6 +4,6 @@
 
 The Pi 0.87 `/openmerit` slash commands were removed with that adapter and have no replacement.
 
-**Retired in npm 0.3.0:** `openmerit dash` ([details](inference-terminal.md)) and all `openmerit eval` commands, including `plan`. Existing local files are untouched. Unreleased Scout source also retires `/scout/api/recommend`; use Scout's conversational agent or MCP. Hosted Workers change only through separate approved deployments. See [retirement details](agent-evaluations.md).
+**Retired in npm 0.3.0:** `openmerit dash` ([details](inference-terminal.md)) and all `openmerit eval` commands, including `plan`. Existing local files are untouched. The Scout serving deploy of 5 November 2026 retired `/scout/api/recommend`; use Scout's conversational agent or MCP. Hosted Workers change only through separate approved deployments. See [retirement details](agent-evaluations.md).
 
 The [Scout setup commands](scout.md#commands) remain available, beginning with `npx -y openmerit@latest install --client claude-code`.

package/docs/getting-started.md

@@ -34,4 +34,4 @@ See [the Pi extension guide](pi-extension.md) for installed paths, migration fro
 
 ## Evaluate application changes
 
-The standalone evaluation CLI is retired in npm 0.3.0, and the legacy JSON recommendation API in unreleased Scout source. Scout guidance is not measured application quality; use your own task checks and provider budget controls. See [retirement details](agent-evaluations.md).
+The standalone evaluation CLI is retired in npm 0.3.0, and the legacy JSON recommendation API went offline with the Scout serving deploy of 5 November 2026. Scout guidance is not measured application quality; use your own task checks and provider budget controls. See [retirement details](agent-evaluations.md).

package/docs/index.md

@@ -1,6 +1,6 @@
 # OpenMerit documentation
 
-These maintained Markdown guides distinguish current source work from published behavior. OpenMerit **0.1.12** added native Scout support for Pi **1.5.0**, replacing the old Pi 0.87 evaluation extension. OpenMerit **0.3.0** retires the Inference Terminal (`openmerit dash`) and standalone evaluation CLI. Unreleased source also retires the legacy recommendation API and the unpublished docs website shell. The policy, protected memo, Scout service and npm package are deployed independently; dated receipts describe the versions actually tested.
+These maintained Markdown guides distinguish current source work from published behavior. OpenMerit **0.1.12** added native Scout support for Pi **1.5.0**, replacing the old Pi 0.87 evaluation extension. OpenMerit **0.3.0** retires the Inference Terminal (`openmerit dash`) and standalone evaluation CLI. The Scout serving deploy of 5 November 2026 retired the legacy recommendation API; unreleased source also retires the unpublished docs website shell. The policy, protected memo, Scout service and npm package are deployed independently; dated receipts describe the versions actually tested.
 
 ## Discover model candidates
 

package/docs/roadmap.md

@@ -2,7 +2,7 @@
 
 ## Current status
 
-The Pi **1.5.0** adapter in OpenMerit **0.1.12** provides only native Scout MCP and shared guidance, not evaluation coordination. OpenMerit **0.3.0** retires the Inference Terminal and the standalone evaluation CLI; unreleased Scout source retires the legacy recommendation API. Passing implementation tests do not establish that an application has better models, trustworthy evaluations, or reliable unattended automation.
+The Pi **1.5.0** adapter in OpenMerit **0.1.12** provides only native Scout MCP and shared guidance, not evaluation coordination. OpenMerit **0.3.0** retires the Inference Terminal and the standalone evaluation CLI; the Scout serving deploy of 5 November 2026 retired the legacy recommendation API. Passing implementation tests do not establish that an application has better models, trustworthy evaluations, or reliable unattended automation.
 
 ## Known limitations
 

package/docs/scout-data-policy.md

@@ -1,6 +1,6 @@
 # Scout data policy
 
-Updated 29 October 2026.
+Updated 5 November 2026.
 
 OpenMerit operates Scout. This policy covers its hosted website and connected clients.
 
@@ -9,6 +9,7 @@ OpenMerit operates Scout. This policy covers its hosted website and connected cl
 ## Data we use
 
 - **Service data:** Your submitted task and related information are processed by an AI service provider, even with analytics off. We also retain account, authorization and device details needed to provide access.
+- **Measured results:** Aggregate evaluation results you report on a Scout conversation, such as a model's score on your own eval, are stored with that conversation and used only for its recommendations. Send aggregates only, never the inputs, outputs or records they were measured on.
 - **Optional analytics:** When enabled, we collect usage, recommendations, feedback, setup activity, device/browser categories, performance and error information to improve Scout and fix problems. Account analytics may include your name, verified email and sign-in provider, and may link activity before and after sign-in.
 - **Shared content:** When prompt sharing is enabled, analytics also includes task text and analysis. We use this for product improvement, not foundation model training. There is no session recording.
 

package/docs/scout.md

@@ -2,7 +2,7 @@
 
 Scout is OpenMerit's model search agent. Describe your task and get provisional model guidance through the conversational page or MCP. Public catalogue evidence is not proof of application quality.
 
-**Unreleased retirement, November 2026:** this source removes `/scout/api/recommend`, its alpha/versioned aliases and the complete standalone `openmerit eval` CLI, including `plan`. Retired paths return 404 before authentication, redirects or asset fallback. The agent, canonical `/scout/mcp`, coding-agent setup, catalogue publisher, status and account/privacy capabilities remain. No remote data or existing local evaluation files are deleted. Package publication and Worker deployment require separate approval; neither is implied by this source change.
+**Retirement, November 2026:** npm 0.3.0 removed the complete standalone `openmerit eval` CLI, including `plan`. The Scout serving deploy of 5 November 2026 removed `/scout/api/recommend` and its alpha/versioned aliases; retired paths return 404 before authentication, redirects or asset fallback. The agent, canonical `/scout/mcp`, coding-agent setup, catalogue publisher, status and account/privacy capabilities remain. No remote data or existing local evaluation files are deleted.
 
 Dated receipts and the old search/ranking/JSON protocol descriptions below are historical, not supported interfaces in this source. There is no replacement frozen-plan API. See [evaluation retirement](agent-evaluations.md) and [commands](commands.md) for migration boundaries.
 
@@ -18,6 +18,21 @@ Scout is a separate hosted beta product. Npm 0.1.7 introduced optional Claude Co
 
 **Product wording — unreleased source:** the homepage, Scout page, onboarding and search/social descriptions call Scout “OpenMerit's model search agent,” with “model search” as the shorthand and “Find the right AI models for your task” as its promise. “Model recommendations” describes its output; the web and coding agents are access channels, not separate product categories. This copy is not yet deployed. Dated receipts below retain the wording and behavior of their original releases.
 
+## Evaluation plans and measured results — deployed 5 November 2026, npm 0.3.1
+
+The Scout serving deploy of 5 November 2026 (version `ea371c78-6ccf-4931-a0b8-aa28a6599aaa`) serves this behavior to every client. A live MCP check returned an evaluation plan and agent grouping, and reported results moved a measured model below the quality bar. npm 0.3.1 adds `openmerit scout --results` and the installed instructions for measuring; older installations reach the new `scout_models` fields through MCP but their instructions do not mention measuring until the installer is rerun.
+
+OpenMerit does not run evaluations or host an eval platform. Scout tells you what to measure, your own eval system measures it, and Scout uses the aggregate results:
+
+- **Agents and calls.** A task is still one model call. When the work is an agent that calls models in a loop, a task can name its `agent` and `callsPerRun`; each model's `cost` then also has `usdPer1kAgentRuns`, the same route's cost of 1,000 runs of that agent.
+- **Evaluation plan.** Every task in a recommendation has an `evaluation`: `checks` at the `task` (one model call), `agent` (one agent run) or `environment` (end-to-end outcome) level, labelled `samples` per candidate, and a `stopEarly` rule for a candidate that is clearly losing. The installed instructions tell the coding agent to add the candidates as extra runs of evals the project already has, such as datasets and experiments in its eval or tracing platform, or its own eval scripts, and not to build a second eval. If none exists, it proposes a minimal one and creates it only where and how you agree.
+- **Measured results.** Continue the conversation with `results` (`scout_models`, `POST /scout/api/agent`) or `openmerit scout --conversation <id> --results <file.json>`. Each result is one aggregate: `level`, optional `task` title, `scope`, `model` (catalogue model or endpoint ID), `metric`, `value`, `higherIsBetter`, `samples` and `source`, at most 50 per call and 200 per conversation. Inputs, outputs and records are not accepted. Scout recommends again with measured results ranked above public benchmarks: a candidate you measured on a task must cite that result, cited results appear as evidence named "Your eval", ranked among your results on the same metric, and `quality.measured` is true for models whose estimate rests on them. Since the Scout serving deploy of 5 November 2026 (version `8782d345-a63c-4560-bb08-1a6a95e8c0ae`), models whose estimate rests on your results also rank ahead of unmeasured ones, in the conversation's priority order within each group (cheapest first by default), so a model estimated only from benchmarks no longer leads a shortlist you measured. Before it, measured results could only move a model below the bar.
+- **Scope.** Results are stored with the conversation for your account and inform only your recommendations. They are not shared with other accounts or used to change Scout's catalogue.
+
+The quality bar now considers what an error costs once your environment's controls, such as a person approving every payment, have acted. Cost remains the default ranking among models that reach the bar.
+
+Known limitations: Scout does not connect to eval platforms yet, so results arrive only through the coding agent or the CLI; the web page neither shows the evaluation plan nor accepts results; and Scout cannot verify a reported result.
+
 ## Five-step introduction modal — deployed, 28 October 2026
 
 **Deployment receipt:** the owner approved commit, push and deployment. [PR #28](https://github.com/laz-aslam/openmerit/pull/28) merged as [a70103a](https://github.com/laz-aslam/openmerit/commit/a70103a1c8eb8a701ee262106ae6e6c026cf04b4). All PR and merged-commit CI checks passed. The guarded Scout-serving deploy moved `openmerit-scout-alpha` from `2cf2c2d3-8aef-40f0-9c44-800bcab247bf` to `17ab0166-9a5c-43e9-932b-c13bee040db7` at 20:45 UTC. Live Scout HTML, chat/auth/onboarding JavaScript, styles and Claude screenshot matched the release build. `/` and `/connect/` stayed byte-identical; other Workers were not redeployed. The catalogue remained healthy on snapshot `588855a0bf28212fad48a9513dbf3d5b`.
@@ -45,12 +60,12 @@ Once the brief is settled, Scout splits the work into tasks, one per model it ne
 Coding agents now use the same Scout agent as the page, instead of the older single-call search. Claude Code and Pi reach it through the `scout_models` MCP tool, which runs one conversation turn per call and returns `status`:
 
 - `needs_answers`: Scout's `brief` and `questions`. The agent shows them to the user, then calls again with the `conversationId` and `brief` set to the rewritten components, or `skip: true`.
-- `recommended`: `tasks[]`, each with `title`, `work`, `input`, `output`, `summary`, `workload`, `priority` (`cost` by default, or `quality`, `latency` or `recency`), `qualityBar` and up to five ranked `models`. Coding-agent conversations have no task limit. Each model's `cost` names the cheapest eligible route's `endpointId` and `providerModelId`.
+- `recommended`: `tasks[]`, each with `title`, `work`, `input`, `output`, `summary`, `workload`, `priority` (`cost` by default, or `quality` or `recency`), `qualityBar` and up to five ranked `models`. Coding-agent conversations have no task limit. Each model's `cost` names the cheapest eligible route's `endpointId` and `providerModelId`.
 - `working`: the call ended before the turn did. Calling again with only the `conversationId` reads the outcome.
 
 The tool takes `prompt`, `conversationId`, `brief`, `skip` and, on the first call, `endpointIds`. The inventory limits every search and shortlist in that conversation, whatever filters the agent's model chooses; an empty list allows none, and later turns cannot change it. `POST /scout/api/agent` accepts the same `endpointIds`. `modelIds` is no longer accepted, and the response no longer has `tasks[].slots`. A turn usually takes one to three minutes, much longer than a search. Conversations are keyed by the device key's account. The local `npm run scout` preview has no agent and reports MCP calls as unavailable.
 
-npm 0.2.0 adds `openmerit scout` for agents without MCP and for people at a terminal. It calls the same tool with this device's key and applies the project's saved endpoint inventory. Without `--json` it prints each task's models in ranked order with their quality estimate and list prices. As on the page, the default cheapest-first order is unlabelled; when the user asks for the best quality, the fastest or the newest model, a line names that order, and fastest and newest lists also show each model's time per task or release date (unknown when missing). The installed instructions tell the agent to relay Scout's questions rather than answer them itself.
+npm 0.2.0 adds `openmerit scout` for agents without MCP and for people at a terminal. It calls the same tool with this device's key and applies the project's saved endpoint inventory. Without `--json` it prints each task's models in ranked order with their quality estimate and list prices. As on the page, the default cheapest-first order is unlabelled; when the user asks for the best quality, the fastest or the newest model, a line names that order, and fastest and newest lists also show each model's time per task or release date (unknown when missing). Scout no longer ranks by speed, so the fastest order no longer appears. The installed instructions tell the agent to relay Scout's questions rather than answer them itself.
 
 ## Every model kind in the agent — source only, not deployed
 
@@ -63,7 +78,7 @@ The agent's tools change as follows:
 - **search_catalogue** (new) searches models, benchmarks and leaderboards by free text. By default it combines keyword (BM25) and embedding results by reciprocal rank fusion; it can also use either alone, and filter by result type and model kind. Each result names the field that matched. Without embeddings (no OpenAI key, an embedding error, or a catalogue version published without them) it returns keyword results and says so. Its queries, written by the agent from your task, are embedded with OpenAI, as benchmark search already does.
 - **search_models** filters by model kind and reports the cheapest eligible price in each unit a model is billed in: per million tokens, per image, per second of audio or video, per million characters, per page or per request. Sorting by price compares one unit only (by default the usual unit for the requested kind, otherwise tokens); models not billed in that unit follow, unsorted. A route that lists only zero prices, such as a local runtime or a free tier, is not counted as a price.
 - **get_model_details** and **benchmark_leaderboard** include leaderboard results, and leaderboard IDs work wherever benchmark IDs do.
-- **submit_recommendation** accepts models of any kind. Fewer than five are accepted only when fewer models meet the stated hard constraints, which can now include the model kind. The per-task workload is given in billing units (input and output tokens, images, seconds of audio or video, characters, pages, requests) and cost per 1,000 tasks is computed from catalogue prices on the eligible route that prices most of the workload, then the cheapest. Quantities the route does not price are listed, and such a partial cost never leads on cost. Citations must be the model's own benchmark or leaderboard results.
+- **submit_recommendation** accepts models of any kind. It must assess at least five candidates, fewer only when fewer models meet the stated hard constraints, which can now include the model kind. Fewer than five may reach the quality bar, so the shortlist can be shorter than five instead of the bar dropping to fill it; at least one must reach it. For a task that is an agent's call, every candidate at or above the bar must cite a benchmark tagged tools or agents, or the user's measured result, so domain benchmarks alone cannot qualify a model that calls tools. The per-task workload is given in billing units (input and output tokens, images, seconds of audio or video, characters, pages, requests) and cost per 1,000 tasks is computed from catalogue prices on the eligible route that prices most of the workload, then the cheapest. Quantities the route does not price are listed, and such a partial cost never leads on cost. Citations must be the model's own benchmark or leaderboard results.
 
 In the shortlist, the Input and Output columns keep per-million-token prices; a model billed in another unit shows that unit's price there instead, marked with its unit, and the expanded model shows the estimated cost per 1,000 tasks. Known limitations: per-second audio prices apply to audio seconds, but audio billed in tokens must be estimated as tokens; image prices are per generated image; prices are base-tier list prices from catalogue sources and can be wrong at the source.
 
@@ -83,7 +98,7 @@ In the shortlist, the Input and Output columns keep per-million-token prices; a
 
 The conversational agent is the Scout page at `/scout/`, replacing the search box and its results table. The page stays open to everyone; sending a message requires sign-in, as searching did, and a message typed while signed out is kept while you sign in. The JSON search API (`/scout/api/recommend`), the MCP server (`/scout/mcp`), coding-agent setup at `/connect/` and the catalogue refresh are unchanged, so the CLI and installed coding agents keep their current behaviour. Sections below that describe the search page's prompt examples, results table and loading messages are historical; the Claude Code setup handoff remains available in the new page.
 
-You describe the work; Scout writes a task brief with five components (input, work, output, criteria and constraints), each marked stated, assumed or missing. A component holds only what is known, without listing unknowns. Criteria and constraints are filled only from what you explicitly say, such as "best quality output" or "open weights only"; otherwise they stay empty and missing. Each component is editable in place: correct anything Scout got wrong, or fill in a missing one. All five fields share the same editable background, including Work; the page does not show per-component question text. Once sent, the brief collapses. **Continue** sends only the components you changed and requests recommendations immediately, using stated assumptions for missing details. It also works without edits; there is no separate Skip questions button. Pressing Enter in a brief field does the same; Shift+Enter adds a line break. The message box stays at the bottom of the page. Each turn ends with an editable brief or a shortlist of five models, followed by Scout's summary. While a turn runs, its steps are listed with the current one last; afterwards they stay as a collapsed record of the work. By default each shortlist is the five cheapest models that maintain the task's output quality. Scout sets a 0–100 quality bar for the task and estimates each candidate it assessed (up to ten) on the same scale. Candidates below the bar are dropped, the rest are ranked by estimated cost per task (Scout's estimate of the task's workload at the route's list prices) and the cheapest five are shown, so a costlier model appears only when fewer than five cheaper ones reach the bar. A cost that leaves part of the workload unpriced ranks after complete ones. Only when the user explicitly asks for the best quality, the fastest or the newest model are the qualifying candidates ranked by quality estimate, reference time or release date instead. The default ranking is not labelled on the page; an explicit priority is named in a line above the table. The bar and the estimates are Scout's judgement from benchmark standings, not measured accuracy. The shortlist reports Scout's 0–100 quality estimate with the cited benchmark ranks, input and output prices per million tokens on its cheapest eligible route and release date; a latency ranking adds a time per task column. Expanding a model shows its rationale, benchmark ranks, route and reference latency. A missing release date is shown as unknown. Reply in the same conversation to change constraints or trade-offs; **New** starts over. When Scout is updated while the page is open, a notice at the bottom of the sidebar asks you to refresh, and clicking anywhere on it reloads the page; an open page notices on its next turn or when you return to its tab, and checks `GET /scout/api/version`, which returns the deployed version ID.
+You describe the work; Scout writes a task brief with five components (input, work, output, criteria and constraints), each marked stated, assumed or missing. A component holds only what is known, without listing unknowns. Criteria and constraints are filled only from what you explicitly say, such as "best quality output" or "open weights only"; otherwise they stay empty and missing. Each component is editable in place: correct anything Scout got wrong, or fill in a missing one. All five fields share the same editable background, including Work; the page does not show per-component question text. Once sent, the brief collapses. **Continue** sends only the components you changed and requests recommendations immediately, using stated assumptions for missing details. It also works without edits; there is no separate Skip questions button. Pressing Enter in a brief field does the same; Shift+Enter adds a line break. The message box stays at the bottom of the page. Each turn ends with an editable brief or a shortlist of five models, followed by Scout's summary. While a turn runs, its steps are listed with the current one last; afterwards they stay as a collapsed record of the work. By default each shortlist is the five cheapest models that maintain the task's output quality, or fewer when fewer reach the bar. Scout sets a 0–100 quality bar for the task and estimates each candidate it assessed (up to ten) on the same scale. Candidates below the bar are dropped, the rest are ranked by estimated cost per task (Scout's estimate of the task's workload at the route's list prices) and the cheapest five are shown, so a costlier model appears only when fewer than five cheaper ones reach the bar. A cost that leaves part of the workload unpriced ranks after complete ones. Only when the user explicitly asks for the best quality or the newest model are the qualifying candidates ranked by quality estimate or release date instead. Speed never ranks: few benchmarks measure it, and it depends on the endpoint and the reasoning setting, so reference latency is shown for information only. The default ranking is not labelled on the page; an explicit priority is named in a line above the table. The bar and the estimates are Scout's judgement from benchmark standings, not measured accuracy. The shortlist reports Scout's 0–100 quality estimate with the cited benchmark ranks, input and output prices per million tokens on its cheapest eligible route and release date. Expanding a model shows its rationale, benchmark ranks, route and reference latency. A missing release date is shown as unknown. Reply in the same conversation to change constraints or trade-offs; **New** starts over. When Scout is updated while the page is open, a notice at the bottom of the sidebar asks you to refresh, and clicking anywhere on it reloads the page; an open page notices on its next turn or when you return to its tab, and checks `GET /scout/api/version`, which returns the deployed version ID.
 
 The agent runs in a Cloudflare Durable Object, one per conversation, using Pi Durable and the Agents SDK. Its model is set by `SCOUT_AGENT_MODEL`: `openai/<model>` calls OpenAI's Responses API directly with the Worker's `OPENAI_API_KEY` (default `openai/gpt-6-sol`), and any other value runs on Workers AI, such as `@cf/zai-org/glm-5.3`. Its tools search benchmark descriptions by embedding (OpenAI embeddings, as in search), read leaderboards and model records from the published catalogue, and use Workers AI web search when the catalogue cannot answer something decisive. Cost, latency and recency figures come from catalogue data, not from the model's text, and a shortlist that cites models or benchmarks outside the catalogue, or breaks a stated hard constraint, is rejected and resubmitted. A turn usually takes one to three minutes.
 
@@ -337,9 +352,9 @@ The prompt contains no key, account data or task draft. Its 10-minute timeout is
 
 If the Claude desktop app isn't installed or your browser blocks the link, paste the copied prompt into Claude Code instead. The command needs `npx` (Node.js 22.19 or newer) and the `claude` CLI on the PATH of that session; the installer runs `claude mcp add-json`. Having the desktop app alone does not establish that those commands are available. See [Claude Desktop](https://code.claude.com/docs/en/desktop).
 
-When the command can open a browser, it waits up to nine minutes for the Approve click. When it can't, for example over SSH, it prints the link and code and returns after 30 seconds with exit code 10 so the agent can show them. Rerunning the same command resumes the same request; it does not issue a new code. Codes expire after ten minutes and can be used once. `OPENMERIT_LOGIN_WAIT` sets the wait in seconds.
+When the command can open a browser, it waits up to nine minutes for the Approve click. When it can't, for example over SSH, it prints the link and code and returns after 30 seconds with exit code 10 so the agent can show them. Rerunning the same command resumes the same request; it does not issue a new code. Codes expire after 30 minutes (ten minutes before the Scout serving deploy of 5 November 2026) and can be used once. From npm 0.3.2 the installer prints the expiry in UTC with the minutes left, the pending message and `--json` result (`approvalUrl`, `userCode`, `expiresAt`) carry the link, and the installed skill tells the agent to show both. `OPENMERIT_LOGIN_WAIT` sets the wait in seconds.
 
-`--client` defaults to the agent running the command (Claude Code sets `CLAUDECODE=1`). `--scope` defaults to `user`, which makes Scout available in every project. `--scope project` writes the skill to `.claude/skills/` and startup rule to `.claude/rules/` in the current project and stores the MCP server in Claude Code's private per-project (`local`) configuration, so the key never enters the repository. `--plan` lists the changes without making them. `--json` prints one result object for agents.
+`--client` defaults to the agent running the command (Claude Code sets `CLAUDECODE=1`). `--scope` defaults to `user`, which makes Scout available in every project. `--scope project` writes the skill to `.claude/skills/` and startup rule to `.claude/rules/` in the current project and stores the MCP server in Claude Code's private per-project (`local`) configuration, so the key never enters the repository. Claude Code keys that configuration by the git repository root, so for a project inside a larger repository (such as a monorepo package) the MCP server applies to every Claude Code session in that repository, while the skill and rule stay in the project; npm 0.3.2 and later say so, and earlier versions report a failed install in this case. `--plan` lists the changes without making them. `--json` prints one result object for agents.
 
 | Exit code | Meaning |
 |---|---|

package/docs/security.md

@@ -2,7 +2,7 @@
 
 **Pi adapter — npm 0.1.12 and later:** Pi 1.5.0 uses native MCP plus shared Scout rule/skill sections, not the former evaluation adapter. It never initiates login, changes the host model/tools/session, or provisions background jobs. Browser-approved credentials are independent per client/scope/project, stored privately outside the repository, and passed as MCP HTTP headers, never instructions. Pi's installed connection JSON contains only the private credential path. See [Pi paths and targeted removal](pi-extension.md). The standalone evaluation CLI remains separate.
 
-**Retirement:** npm 0.3.0 removes the standalone [agent-evaluation runner](agent-evaluations.md), including planning; unreleased Scout source removes `/scout/api/recommend`. Existing local reports, budgets, configuration and host-agent instructions are not deleted; review them separately. No provider billing control or remote data deletion is implied.
+**Retirement:** npm 0.3.0 removes the standalone [agent-evaluation runner](agent-evaluations.md), including planning; the Scout serving deploy of 5 November 2026 removed `/scout/api/recommend`. Existing local reports, budgets, configuration and host-agent instructions are not deleted; review them separately. No provider billing control or remote data deletion is implied.
 
 The Scout agent, MCP, publisher and account analytics/deletion capabilities remain. Historical analytics and stored catalogue/research data are not erased by endpoint retirement. Necessary provider processing for the agent and publisher continues under the [Scout data policy](scout-data-policy.md).
 
@@ -30,7 +30,7 @@ Hosted Scout uses Clerk Google/GitHub sign-in. The page and status are public; J
 
 ### Coding-agent device keys
 
-The [coding-agent setup](scout.md#connect-a-coding-agent) in npm 0.1.7 uses hosted device sign-in instead of OAuth or tokens in prompts. The copied prompt contains no secret. A signed-in person approves an eight-character code on `/connect`; codes expire after ten minutes and are single-use. Approval mints one key per device and client, scoped to `scout:recommend` and accepted only on `/scout/mcp`. The server stores only SHA-256 hashes. Keys expire after 90 days without use and can be revoked on `/connect` or with `openmerit remove`/`logout`. Locally the key is in `~/.config/openmerit/credentials.json` (mode 600) and in Claude Code's plain-text MCP configuration; the CLI never prints it. As with any device-code flow, only approve a code you started.
+The [coding-agent setup](scout.md#connect-a-coding-agent) in npm 0.1.7 uses hosted device sign-in instead of OAuth or tokens in prompts. The copied prompt contains no secret. A signed-in person approves an eight-character code on `/connect`; codes expire after 30 minutes (ten before the Scout serving deploy of 5 November 2026) and are single-use. Approval mints one key per device and client, scoped to `scout:recommend` and accepted only on `/scout/mcp`. The server stores only SHA-256 hashes. Keys expire after 90 days without use and can be revoked on `/connect` or with `openmerit remove`/`logout`. Locally the key is in `~/.config/openmerit/credentials.json` (mode 600) and in Claude Code's plain-text MCP configuration; the CLI never prints it. As with any device-code flow, only approve a code you started.
 
 **Automatic instruction updates — version 0.1.10:** updater-enabled setup defaults on with an opt-out, installing asynchronous Claude Code startup/response-end hooks. Checks use npm's public stable registry without credentials, task content or installation IDs. SHA-512 verification checks package bytes against registry metadata; it does not defend against a compromised publisher/registry. The updater reads only a bounded instruction bundle, never executes downloaded code or package scripts, and leaves MCP configuration, keys, privacy settings and project policy unchanged. A supported release contract permits instruction updates; maintainers must change it when permissions or data handling change. This is a publisher assertion, not semantic validation of instructions. Edited files are retained, concurrent writes serialize, and journaled recovery stops on intervening edits. New sessions reliably load new instructions. See [automatic updates](scout.md#automatic-instruction-updates-version-0110) for cadence, offline retries, manual commands and the one-time bootstrap required for older installations. Availability follows npm publication independently of hosted releases.
 

package/docs/testing.md

@@ -1,6 +1,6 @@
 # Testing and validation
 
-**Cleanup:** npm 0.3.0 retires the standalone evaluation CLI and the Inference Terminal; unreleased source also retires the legacy JSON recommendation pipeline and unpublished docs website shell. Dated evaluation/API acceptance records below describe historical releases, not current supported commands. Current verification retains agent, MCP, publisher, status, policy and protected-memo coverage; it does not publish or deploy anything.
+**Cleanup:** npm 0.3.0 retires the standalone evaluation CLI and the Inference Terminal; the Scout serving deploy of 5 November 2026 retired the legacy JSON recommendation pipeline, and unreleased source retires the unpublished docs website shell. Dated evaluation/API acceptance records below describe historical releases, not current supported commands. Current verification retains agent, MCP, publisher, status, policy and protected-memo coverage; it does not publish or deploy anything.
 
 **Pi adapter — npm 0.1.12 and later:** the Pi 1.5.0 adapter uses native Scout MCP and shared guidance only. Dated acceptance records below describe the packages tested at the time; some mention package exports that have since been removed. Standalone evaluation tests remain separate.
 

package/docs/troubleshooting.md

@@ -11,7 +11,7 @@ npx -y openmerit@latest status --client pi --scope user --check
 npx -y openmerit@latest doctor --client pi --scope user
 ```
 
-Use `--scope project` from the original project root for project setup. Check Pi and Node.js versions against [setup](getting-started.md), complete browser device approval, then start a **new session**. Installing the npm package or reconnecting an old session does not load new extension instructions. For SDK embedding, load Pi's `createMcpExtension()` explicitly alongside `createScoutPiExtension`; the Scout factory alone is not an MCP runtime.
+Use `--scope project` from the original project root for project setup. In a project inside a larger git repository, Claude Code project setup up to npm 0.3.1 ends with "Claude Code did not save the openmerit MCP server" although Claude Code saved it under the repository root; Scout still works in that repository, and npm 0.3.2 reads the entry correctly; rerun the install with it. Check Pi and Node.js versions against [setup](getting-started.md), complete browser device approval, then start a **new session**. Installing the npm package or reconnecting an old session does not load new extension instructions. For SDK embedding, load Pi's `createMcpExtension()` explicitly alongside `createScoutPiExtension`; the Scout factory alone is not an MCP runtime.
 
 ## The connection configuration is invalid
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openmerit",
-  "version": "0.3.0",
+  "version": "0.3.2",
   "description": "Evidence-driven model decisions for coding harnesses.",
   "type": "module",
   "bin": {