drafted 1.12.5 → 1.12.7

package/agent-instructions/global.md

@@ -14,9 +14,9 @@ Operating principle:
 
 Before working:
 - Check whether Drafted MCP tools are available. If not authenticated, use auth(action="get_link") or auth(action="login") and give the user the sign-in link.
-- Select the correct organization first. Use get_org(action="get") and project(action="list") to identify the target org/project. Opening a project switches org context automatically; for wiki/skill-only work, use get_org(action="switch", orgId=...).
-- Check get_org(action="get") for googleDrive.connected. When Google Drive is connected for the active org, Google Workspace frames are available and should be the strong default for documents, spreadsheets, and presentations.
-- Never assume the current active Drafted project or organization is correct. Verify the returned project/org before writing.
+- Identify the correct organization and project first. Use get_org(action="get") and project(action="list"). Opening a project binds the agent session's context — the org derives from the project. There is no org switching: address existing resources by UUID (pageId/skillId/projectId — org self-derives), and pass org=... on creates and searches that name no resource (wiki write, skill add, project create, template actions).
+- Check get_org(action="get") for googleDrive.connected. When Google Drive is connected for the org you are working in, Google Workspace frames are available and should be the strong default for documents, spreadsheets, and presentations.
+- Never assume the current Drafted project or org context is correct. Verify the project/org echoed in each response before writing.
 
 Wiki rules:
 - Search the org wiki before substantive work: wiki(action="search") with relevant keywords.

package/mcp/server.mjs

@@ -187,7 +187,7 @@ const server = new McpServer({
 
 An org contains projects. Each project has a zoomable canvas with frames (HTML files) organized as /{layer}/{lane}/{filename}. Layers are predefined categories (wireframes, designs, brand-assets, etc.), lanes are groups within a layer, and frames are the individual design files.
 
-WORKFLOW: project(action="list") → project(action="open") → ls / → read/write/edit. Projects span all orgs -- opening a project auto-switches org context. Every response includes a "project" field showing which project you're operating on -- always verify it matches your intent before writing. To switch orgs without opening a project (for wiki/skill work in an org with no projects), call get_org(action="switch", orgId=...).
+WORKFLOW: project(action="list") → project(action="open") → ls / → read/write/edit. Projects span all orgs -- opening a project binds this agent session's context, and the org derives from the project. Every response includes a "project" field showing which project you're operating on -- always verify it matches your intent before writing. There is no org switching: address resources by UUID (org self-derives) and pass org=... on creates/searches that name no resource (wiki write, skill add, project create).
 
 SKILLS: Drafted has a skill library -- reusable agent instructions stored as SKILL.md files. When a user says "use the X skill", call skill(action="search") to find it, then skill(action="load") to get its instructions. Skills can cover anything: UX guidelines, copywriting rules, brand voice, coding standards, review checklists, etc.
 
@@ -200,7 +200,9 @@ IMPORTANT: Any URL containing /f/{uuid} is a Drafted frame link — ALWAYS use r
 }, {
   // Initialize instructions: the agent-identity contract, so an agent learns its own
   // tab name + the right way to read it WITHOUT having to "think to" call a tool.
-  instructions: `SESSION IDENTITY — read this first: you run as a NAMED session tab visible to the user on their Drafted surface. Your session has a human-readable name (a Greek term, e.g. "Nous") — that name is how the user matches YOUR window to the tab they see, so identify yourself by it when it matters which agent you are. Read it from get_org (response field "session.name") or from the whoami tool; call whoami to refresh after a reconnect.`,
+  instructions: `SESSION IDENTITY — read this first: you run as a NAMED session tab visible to the user on their Drafted surface. Your session has a human-readable name (a Greek term, e.g. "Nous") — that name is how the user matches YOUR window to the tab they see, so identify yourself by it when it matters which agent you are. Read it from get_org (response field "session.name") or from the whoami tool; call whoami to refresh after a reconnect.
+
+DUAL REGISTRATION IS NORMAL: a separate "Drafted" remote connector (https://drafted.live/mcp, managed by claude.ai) may appear alongside this local stdio server in MCP listings. It is NOT a duplicate and NOT broken — it serves claude.ai web/mobile/Cowork, where a local stdio process cannot run. Never advise removing it; a "needs authentication" state on it is fixed by signing in from claude.ai and does not affect this stdio session.`,
 });
 
 const layerKeys = Object.keys(LAYERS);
@@ -223,8 +225,8 @@ const TOOL_ANNOTATIONS = {
   whoami:              { title: 'Session identity',    readOnlyHint: true,  destructiveHint: false, openWorldHint: false, description: 'Return THIS agent session\'s identity: its server-assigned human-readable name (the correlation key between an agent window and its web-app session tab), sessionId, userId, orgId, active projectId, editor label, server URL, and surfaced/alive state. Read-only. Use this — not guesses from the host environment — to report which session you are.' },
 
   // Projects
-  project:             { title: 'Projects',            readOnlyHint: false, destructiveHint: false, openWorldHint: false, widgetUri: 'ui://widget/drafted-canvas-overview.html', description: 'Manage projects: list (start here), open (switch active project), create, update, move to another org.' },
-  get_org:             { title: 'Organization',        readOnlyHint: false, destructiveHint: false, openWorldHint: false, description: 'Get the active organization (action="get", default), Google Drive availability, or switch to a different org (action="switch", orgId=...). Use switch when you need wiki/skill work in an org that has no projects — opening a project also switches, but is unavailable in empty orgs. When googleDrive.connected is true, strongly prefer Google Workspace frames for documents, sheets, and slides.' },
+  project:             { title: 'Projects',            readOnlyHint: false, destructiveHint: false, openWorldHint: false, widgetUri: 'ui://widget/drafted-canvas-overview.html', description: 'Manage projects: list (start here), open (bind this agent session to a project — org derives from it), create (org= names where it is born), update, move to another org.' },
+  get_org:             { title: 'Organization',        readOnlyHint: true,  destructiveHint: false, openWorldHint: false, description: 'List your orgs, the default org, and Google Drive availability (action="get", default), or fetch installed MCP update instructions (action="update_mcp"). There is no org switching — org derives from the resource you address; creates/searches take org=. When googleDrive.connected is true, strongly prefer Google Workspace frames for documents, sheets, and slides.' },
 
   // Templates
   template:            { title: 'Templates',            readOnlyHint: false, destructiveHint: true,  openWorldHint: false, description: 'Manage project templates: list, create, update, delete, fork.' },
@@ -1235,21 +1237,21 @@ async function getCurrentOrgContext() {
 }
 
 // Project-less mutations (wiki, skills) have no resource to derive the org from.
-// If the agent never bound an org this session (no project opened, no get_org
-// switch) and passed no explicit org, refuse to GUESS when the user belongs to
-// more than one org — otherwise the write silently lands in whatever org the
-// session happened to inherit (the "wiki write into the wrong org" failure).
+// If the agent never bound this session to a project and passed no explicit org,
+// refuse to GUESS when the user belongs to more than one org — otherwise the
+// write silently lands in whatever org the session happened to inherit (the
+// "wiki write into the wrong org" failure).
 // Single-org users are unambiguous and proceed untouched. (DRAFT-36 Phase 4.)
 async function requireBoundOrgForProjectlessMutation(explicitOrg) {
   if (explicitOrg) return;
   const sess = getSessionState();
-  if (sess.boundOrgId) return;            // bound via project open / get_org switch
+  if (sess.boundOrgId) return;            // bound via project open (org = project's org)
   if (getState().projectId) return;        // an active project implies its org
   // Remote/web sessions are isolated per connection: each one gets its OWN
   // server-side session row with its own org_id set on connect, so the session
   // org IS this session's binding — there's no shared, long-lived session to
   // confuse here the way stdio has. Adopt that org automatically instead of
-  // forcing the agent to call get_org switch / project open before a project-less
+  // forcing the agent to open a project before a project-less
   // wiki or skill write. The DRAFT-36 Phase 4 "refuse to guess" guard below
   // therefore applies to stdio only.
   if (isRemote) {
@@ -1267,11 +1269,11 @@ async function requireBoundOrgForProjectlessMutation(explicitOrg) {
   }
   if (count > 1) {
     throw new Error(
-      `Refusing to guess the org: no project is open and no org was bound this ` +
-      `session, but you belong to ${count} orgs — a project-less wiki/skill write ` +
-      `would land in whichever org the session inherited. Call ` +
-      `get_org(action="switch", orgId=...) or project(action="open") first ` +
-      `(skill calls may also pass org=...).`
+      `Refusing to guess the org: no project is open and no explicit org was ` +
+      `given, but you belong to ${count} orgs — a project-less wiki/skill write ` +
+      `would land in whichever org the session inherited. Pass org=... on this ` +
+      `call (org name or id), or open the relevant project first ` +
+      `(project(action="open") — the org derives from the project).`
     );
   }
 }
@@ -1342,8 +1344,8 @@ function normalizeWikiPath(input) {
   return input.replace(/^\/+/, '').replace(/\.md$/, '');
 }
 
-async function getTreeAsMap() {
-  const tree = await api('GET', '/api/wiki/tree');
+async function getTreeAsMap(orgHeader = {}) {
+  const tree = await api('GET', '/api/wiki/tree', undefined, orgHeader);
   const pages = tree.pages || [];
   const pathToPage = {};
   for (const p of pages) { pathToPage[p.path] = p; }
@@ -1540,11 +1542,17 @@ async function sessionSurfaceBlock() {
   // No WS ack yet — best-effort identity from /auth/me so callers still get a userId/org.
   const cookieSid = sessionId || getBootstrapSessionId();
   let me = null;
+  let unreachable = false;
   if (cookieSid) {
     try {
       const res = await fetch(`${getServerUrl()}/auth/me`, { headers: { Cookie: `gc_session=${cookieSid}` } });
       if (res.ok) me = await res.json();
-    } catch { /* not yet authenticated */ }
+    } catch {
+      // Transport failure, NOT a sign-out. Conflating the two made agents tell
+      // users "you're signed out" during a network blip and start needless
+      // re-logins — surface the distinction instead.
+      unreachable = true;
+    }
   }
   return {
     sessionId: cookieSid,
@@ -1556,6 +1564,10 @@ async function sessionSurfaceBlock() {
     color: null,
     surfaced: false,
     alive: false,
+    ...(unreachable ? {
+      serverUnreachable: true,
+      note: `Could not reach ${getServerUrl()} — identity UNKNOWN, not signed out. This is a network/transport failure: do not tell the user they are logged out and do not start a new login; retry when connectivity is back.`,
+    } : {}),
   };
 }
 
@@ -1573,13 +1585,13 @@ tool('whoami', 'Return THIS agent session\'s identity: its server-assigned human
 
 // ── Project management tools (direct HTTP) ────────────────────────
 
-tool('project', 'START HERE for project management. Dispatch by `action`: list (lists all projects across all orgs — always call first), open (switch the active project; required before reading/writing frames), create (new project, optionally from a template), update (change name/folder/description/layers), move (transfer to another org). Opening a project auto-switches the org. To change orgs WITHOUT a project (for wiki/skill work in an empty org), use get_org(action="switch", orgId=...). **Skill gate:** projects with attached skills will REJECT all mutations (write, edit, mv, rm, shape, group, connector, layout, layer, asset upload) until you have loaded each attached skill via skill(action="load"). Skills tell you HOW to do the work — they\'re not optional. Open returns the attached skill list and auto-inlines content for projects with ≤3 skills.', {
+tool('project', 'START HERE for project management. Dispatch by `action`: list (lists all projects across all orgs — always call first), open (bind this agent session to a project; required before reading/writing frames — the org derives from the project), create (new project; org= names where it is born), update (change name/folder/description/layers), move (transfer to another org). There is no org switching: for project-less work (wiki/skills) pass org=... on the call. **Skill gate:** projects with attached skills will REJECT all mutations (write, edit, mv, rm, shape, group, connector, layout, layer, asset upload) until you have loaded each attached skill via skill(action="load"). Skills tell you HOW to do the work — they\'re not optional. Open returns the attached skill list and auto-inlines content for projects with ≤3 skills.', {
   action: z.enum(['list', 'open', 'create', 'update', 'move']).describe('Operation to perform.'),
   projectId: z.string().optional().describe('[open|update|move] project ID. Get IDs from action=list.'),
   name: z.string().optional().describe('[create|update] project name'),
   description: z.string().nullable().optional().describe('[create|update] project description'),
   templateSlug: z.string().optional().describe('[create] template slug (e.g. "web-design", "mobile-app", "landing-page")'),
-  org: z.string().optional().describe('[create] org slug or id to create the project in, without switching the active org. Defaults to the active org.'),
+  org: z.string().optional().describe('[create] org slug or id the project is born in. Defaults to this session\'s org (the open project\'s org, else the default org).'),
   folder: z.string().nullable().optional().describe('[update] folder name (null to remove from folder)'),
   layers: z.array(z.object({}).passthrough()).optional().describe('[update] full layers array replacement. Use ls / to read current layers first.'),
   targetOrgId: z.string().optional().describe('[move] destination organization ID. Get org IDs from action=list (each project has an orgId field) or get_org. Both source and target org must include the current user.'),
@@ -1734,7 +1746,7 @@ tool('project', 'START HERE for project management. Dispatch by `action`: list (
   } catch (error) { return err(error); }
 });
 
-tool('template', 'Manage project templates in the org. Dispatch by `action`: list/create/update/delete/fork. A template bundles layer definitions and an optional list of skill slugs that auto-attach to projects created from it.', {
+tool('template', 'Manage project templates in an org. Dispatch by `action`: list/create/update/delete/fork. A template bundles layer definitions and an optional list of skill slugs that auto-attach to projects created from it. Scopes to the open project\'s org by default; pass org=... to target another org (per-request only — nothing is switched).', {
   action: z.enum(['list', 'create', 'update', 'delete', 'fork']).describe('Operation to perform.'),
   templateId: z.string().optional().describe('[update|delete|fork] template ID'),
   name: z.string().optional().describe('[create|update|fork] template name (required for create; optional rename for fork)'),
@@ -1742,12 +1754,14 @@ tool('template', 'Manage project templates in the org. Dispatch by `action`: lis
   layers: z.array(z.object({}).passthrough()).optional().describe('[create|update] array of layer definitions'),
   skillSlugs: z.array(z.string()).optional().describe('[create|update] skill slugs to auto-attach to projects created from this template. Slugs are resolved against the org\'s skills (org-local first, built-in fallback) at project-create time; missing slugs are silently skipped. Pass an empty array on update to clear.'),
   visibility: z.string().optional().describe('[create|update] "org" or "public"'),
+  org: z.string().optional().describe('Org (id or name) to scope this call to — defaults to the open project\'s org. Per-request only.'),
 }, async (args) => {
   try {
     const { action } = args;
+    const orgHeader = args.org ? { 'X-Drafted-Org': args.org } : {};
     switch (action) {
       case 'list': {
-        const tpls = await api('GET', '/api/templates');
+        const tpls = await api('GET', '/api/templates', undefined, orgHeader);
         markSearched(getSessionState().gates, 'template');
         return ok(tpls);
       }
@@ -1757,7 +1771,7 @@ tool('template', 'Manage project templates in the org. Dispatch by `action`: lis
         const body = { name, description, layers };
         if (Array.isArray(skillSlugs)) body.skillSlugs = skillSlugs;
         if (visibility) body.visibility = visibility;
-        return ok(await api('POST', '/api/templates', body));
+        return ok(await api('POST', '/api/templates', body, orgHeader));
       }
       case 'update': {
         const { templateId, name, description, layers, skillSlugs, visibility } = args;
@@ -1769,19 +1783,19 @@ tool('template', 'Manage project templates in the org. Dispatch by `action`: lis
         if (Array.isArray(skillSlugs)) body.skillSlugs = skillSlugs;
         if (visibility) body.visibility = visibility;
         if (Object.keys(body).length === 0) throw new Error('At least one field is required for action=update');
-        return ok(await api('PUT', `/api/templates/${templateId}`, body));
+        return ok(await api('PUT', `/api/templates/${templateId}`, body, orgHeader));
       }
       case 'delete': {
         const { templateId } = args;
         if (!templateId) throw new Error('templateId required for action=delete');
-        return ok(await api('DELETE', `/api/templates/${templateId}`));
+        return ok(await api('DELETE', `/api/templates/${templateId}`, undefined, orgHeader));
       }
       case 'fork': {
         const { templateId, name } = args;
         if (!templateId) throw new Error('templateId required for action=fork');
         const body = {};
         if (name) body.name = name;
-        return ok(await api('POST', `/api/templates/${templateId}/fork`, body));
+        return ok(await api('POST', `/api/templates/${templateId}/fork`, body, orgHeader));
       }
       default:
         throw new Error(`Unknown template action: ${action}`);
@@ -2118,8 +2132,7 @@ async function getMcpUpdateMetadata() {
 
 
 tool('get_org', {
-  action: z.enum(['get', 'switch', 'update_mcp']).optional().describe('Default: "get" returns active org and member info. Use "switch" with orgId to change the active org without opening a project. Use "update_mcp" to get explicit installed stdio MCP update instructions.'),
-  orgId: z.string().optional().describe('[switch] target org ID to switch to. Must be one of the orgs the user is a member of.'),
+  action: z.enum(['get', 'update_mcp']).optional().describe('Default: "get" returns your orgs, the default org, and Google Drive availability. Use "update_mcp" to get explicit installed stdio MCP update instructions. There is no org switching: org derives from the resource you address (projectId/pageId/skillId), and creates/searches take an explicit org param.'),
 }, async (args = {}) => {
   try {
     const action = args.action || 'get';
@@ -2129,28 +2142,6 @@ tool('get_org', {
       return ok(buildInstalledMcpUpdateInstructions(mcpUpdate));
     }
 
-    if (action === 'switch') {
-      if (!args.orgId) throw new Error('orgId is required for action=switch');
-      await api('POST', '/auth/switch-org', { orgId: args.orgId });
-      // Bust the per-session org cache so subsequent calls re-fetch /auth/me.
-      const sess = getSessionState();
-      sess.cachedOrgId = null;
-      sess.cachedOrgIdTime = 0;
-      // Clear active project too — projects are scoped to orgs, so the
-      // previous one isn't valid in the new org.
-      setMcpActiveProject(null, null);
-      // Bind this session to the chosen org so session recovery re-asserts it
-      // (set AFTER the clear above, which would otherwise leave it unchanged).
-      sess.boundOrgId = args.orgId;
-      const me = await api('GET', '/auth/me');
-      const orgs = (await api('GET', '/api/orgs')).orgs || [];
-      const activeOrg = (orgs || []).map(o => ({ id: o.orgId || o.id, name: o.orgName || o.name })).find(o => o.id === me?.orgId) || null;
-      const googleDrive = await getGoogleDriveAvailability();
-      const mcpUpdate = await getMcpUpdateMetadata();
-      const session = await sessionSurfaceBlock();
-      return ok({ switched: true, activeOrg, googleDrive, mcpVersion: PACKAGE_VERSION, mcpUpdate, session, note: 'Active org switched. Wiki and skill calls now target this org. Active project cleared — open a project (or stay org-scoped for wiki/skill). If googleDrive.connected is true, prefer Google Workspace frames for docs, sheets, and slides.' });
-    }
-
     // Source of truth = the org this MCP process scopes requests to (what mutations
     // will actually hit). Each MCP process is independent — multiple agents can run
     // in parallel scoped to different orgs. /auth/me reads sessions.org_id directly.
@@ -2179,7 +2170,7 @@ tool('get_org', {
       mcpVersion: PACKAGE_VERSION,
       mcpUpdate,
       session: await sessionSurfaceBlock(),
-      note: "activeOrg is the org this MCP process scopes its requests to (carried per-request as X-Drafted-Org); mutations target it. `session` is THIS agent's own surface identity — `session.name` is the human-readable tab name the user sees (use it to identify which agent you are); refresh it via whoami. To scope to a different org without opening a project, call get_org(action=\"switch\", orgId=\"...\"). Concurrent MCP sessions and browser tabs for the same user can be scoped to different orgs. If googleDrive.connected is true, strongly prefer Google Workspace frames for docs, sheets, and slides.",
+      note: "Org is derived from the resource you address: opening a project binds this agent session's context (org = the project's org), and UUIDs (pageId/skillId/projectId) self-derive. activeOrg is only the DEFAULT for calls that name no resource and pass no org param — to target a different org, pass org=... on the call (wiki write/search, skill add/update/fork/push/list, project create, template actions), never a switch. `session` is THIS agent's own surface identity — `session.name` is the human-readable tab name the user sees (use it to identify which agent you are); refresh it via whoami. Concurrent MCP sessions can operate on different orgs simultaneously. If googleDrive.connected is true, strongly prefer Google Workspace frames for docs, sheets, and slides.",
     });
   } catch (error) { return err(error); }
 });
@@ -3113,7 +3104,7 @@ tool('skill', 'Manage the Drafted skill library. Skills are reusable prompts/gui
   path: z.string().optional().describe('[read_file|update_file] relative path inside skill directory (e.g. "examples/react.md")'),
   offset: z.number().optional().describe('[search|list] skip N results for pagination; [read_file] start reading at this byte offset (default 0) — for large files (e.g. a >90KB app-frame bundle) read in chunks using the returned nextOffset until truncated=false'),
   maxBytes: z.number().optional().describe('[read_file] return at most this many bytes from offset (default: whole remaining file). Response reports totalSize/offset/truncated/nextOffset.'),
-  org: z.string().optional().describe('[add] org (id or name) to create the skill in — defaults to the bound/active org; [fork|push|update] resolve/fork into this Drafted org (id or name); scopes the request without switching the active org'),
+  org: z.string().optional().describe('[add] org (id or name) the skill is born in — defaults to the open project\'s org; [list|search|load] scope to this org; [fork|push|update] resolve/fork into this org. Per-request only — nothing is switched.'),
   setup: z.array(z.string()).optional().describe('[add|update] setup command(s) (in order) run on materialize to build a source-only skill, e.g. ["npm ci","npm run build"]'),
   files: z.array(z.object({ path: z.string(), content: z.string() })).optional().describe('[push] source files to push (path + UTF-8 content); server strips artifacts + enforces caps'),
   dir: z.string().optional().describe('[push] local directory to push instead of files[]; walked locally (heavy dirs, .skillinstall/, and .skillignore pre-filtered), server re-enforces. On push the dir\'s .gitignore is auto-updated to exclude .skillinstall/ (the rebuildable bundle).'),
@@ -3131,14 +3122,14 @@ tool('skill', 'Manage the Drafted skill library. Skills are reusable prompts/gui
     }
     switch (action) {
       case 'search': {
-        const { query, tags, scope = 'all', limit, offset, compact } = args;
+        const { query, tags, scope = 'all', limit, offset, compact, org } = args;
         const params = new URLSearchParams();
         if (query) params.set('q', query);
         if (tags?.length) params.set('tags', tags.join(','));
         if (scope) params.set('scope', scope);
         const qs = params.toString();
         const endpoint = query ? '/api/skills/search' : '/api/skills';
-        const result = await api('GET', `${endpoint}${qs ? '?' + qs : ''}`);
+        const result = await api('GET', `${endpoint}${qs ? '?' + qs : ''}`, undefined, org ? { 'X-Drafted-Org': org } : {});
         markSearched(getSessionState().gates, 'skill');
         return ok(shapeSkillCatalog(result, { limit, offset, compact }));
       }
@@ -3147,7 +3138,7 @@ tool('skill', 'Manage the Drafted skill library. Skills are reusable prompts/gui
         if (!skill) throw new Error('skill (ID or slug) is required for action=load');
         const isUuid = /^[a-f0-9-]{36}$/.test(skill);
         const endpoint = isUuid ? `/api/skills/${skill}` : `/api/skills/slug/${skill}`;
-        const result = await api('GET', endpoint);
+        const result = await api('GET', endpoint, undefined, args.org ? { 'X-Drafted-Org': args.org } : {});
         if (result?.id) getSessionState().loadedSkillIds.add(result.id);
         return ok(result);
       }
@@ -3163,11 +3154,11 @@ tool('skill', 'Manage the Drafted skill library. Skills are reusable prompts/gui
         return ok(await api('GET', `/api/skills/${id}/versions`, undefined, extra));
       }
       case 'list': {
-        if (args.scope || args.tags?.length) {
+        if (args.scope || args.tags?.length || args.org) {
           const params = new URLSearchParams();
-          params.set('scope', args.scope || 'all');
+          params.set('scope', args.scope || (args.org ? 'org' : 'all'));
           if (args.tags?.length) params.set('tags', args.tags.join(','));
-          const result = await api('GET', `/api/skills?${params.toString()}`);
+          const result = await api('GET', `/api/skills?${params.toString()}`, undefined, args.org ? { 'X-Drafted-Org': args.org } : {});
           return ok(shapeSkillCatalog(result, { limit: args.limit, offset: args.offset, compact: args.compact }));
         }
 
@@ -3182,13 +3173,13 @@ tool('skill', 'Manage the Drafted skill library. Skills are reusable prompts/gui
             return ok(await api('GET', `/api/projects/${pid}/skills`));
           } catch (e) {
             // 404 / "not found" means the project isn't visible in the
-            // current org (typical after an org switch). Fall through to
+            // session's org (typical after binding to a project in another org). Fall through to
             // org-attached skills rather than bubbling a useless error.
             if (!/not found/i.test(e.message)) throw e;
           }
         }
         const orgId = await getCurrentOrgId();
-        if (!orgId) throw new Error('No active org. Call get_org first or pass projectId.');
+        if (!orgId) throw new Error('No org context. Open a project (project(action="open")), pass projectId, or pass org=.');
         return ok(await api('GET', `/api/orgs/${orgId}/skills`));
       }
       case 'add': {
@@ -3334,11 +3325,11 @@ tool('skill', 'Manage the Drafted skill library. Skills are reusable prompts/gui
 // All 11 actions dispatch from one tool. Read-only actions skip the
 // skill gate; mutations require org-level wiki-maintainer skills loaded.
 
-tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and other agents/humans share maintenance — every edit broadcasts live, and edits from others appear in `recent` and on `read`.\n\n**Wiki always operates on the ACTIVE org.** If the content you need lives in a different org, switch first via `get_org(action="switch", orgId=...)` — don\'t assume "no hits" means the content doesn\'t exist. Verify the active org with `get_org` before searching.\n\nBefore mutating: check `recent` and `search` for relevant existing pages. Before mv/rm: check `links` (or pass `dryRun=true`). After completing a logical session of work, append a `log` entry.\n\nThe tool handles bookkeeping you\'d otherwise forget: `mv` rewrites inbound references via the link index, `read` shows who edited last and when. Use `health` to find unlinked pages and broken links.\n\n**Skill gate:** the org may attach a `wiki-maintainer` skill that you MUST load before mutations. If you get a skill-gate error, run skill(action="load", skill="wiki-maintainer") then retry.', {
+tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and other agents/humans share maintenance — every edit broadcasts live, and edits from others appear in `recent` and on `read`.\n\n**Addressing:** a `pageId` (UUID) self-derives its org — no org arg needed. Path-based and listing actions scope to the open project\'s org by default; pass `org=...` to target another org (there is no org switching). `search` spans ALL your orgs by default — don\'t assume "no hits" means the content doesn\'t exist.\n\nBefore mutating: check `recent` and `search` for relevant existing pages. Before mv/rm: check `links` (or pass `dryRun=true`). After completing a logical session of work, append a `log` entry.\n\nThe tool handles bookkeeping you\'d otherwise forget: `mv` rewrites inbound references via the link index, `read` shows who edited last and when. Use `health` to find unlinked pages and broken links.\n\n**Skill gate:** the org may attach a `wiki-maintainer` skill that you MUST load before mutations. If you get a skill-gate error, run skill(action="load", skill="wiki-maintainer") then retry.', {
   action: z.enum(['ls', 'recent', 'read', 'search', 'links', 'log', 'health', 'write', 'edit', 'mv', 'rm', 'source-register', 'source-list', 'source-get', 'bulk-write']).describe('Operation to perform.'),
   path: z.string().optional().describe('[ls|read|links] wiki path. For ls: default / (root). For read: required. For links: required.'),
   pageId: z.string().optional().describe('[read|edit|mv|rm|links] page UUID (from read/search). UUID-first: addresses the page directly, org auto-derives — no org needed and no path lookup. Preferred over path for an existing page.'),
-  org: z.string().optional().describe('[write] org slug or id to create/target the page in, without switching the active org. [search] scope the search to this org (default: search ALL your orgs). Required to create a page when you belong to more than one org and none is active.'),
+  org: z.string().optional().describe('Org slug or id to scope this call to (per-request only — nothing is switched). [write] the org the page is created in — required when you belong to several orgs and no project is open. [search] restrict to this org (default: ALL your orgs). [ls|recent|read|links|log|health|edit|mv|rm|bulk-write] target this org\'s wiki instead of the open project\'s org. Ignored when a pageId is given (the page self-derives its org).'),
   recursive: z.boolean().optional().describe('[ls] list recursively with depth indicators'),
   limit: z.number().optional().describe('[recent|search] max results (recent default 10, search default 25)'),
   query: z.string().optional().describe('[search] term to search in title, path, and content'),
@@ -3377,9 +3368,19 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
       if (!args.pageId) await requireBoundOrgForProjectlessMutation(args.org);
     }
 
-    // Resolve org context once for the wiki tool. Each MCP session is bound
-    // to one org (parallel agents on different orgs is supported by design).
-    const orgCtx = await getCurrentOrgContext();
+    // Resolve org context for this CALL. An explicit org= override wins (per
+    // request, nothing is switched); otherwise the session's binding (the open
+    // project's org). Resolving the override here keeps the echoed `org` field
+    // and every emitted browser URL truthful about where the call landed.
+    let orgCtx = await getCurrentOrgContext();
+    if (args.org) {
+      const d = await api('GET', '/api/orgs');
+      const list = (d.orgs || d || []).map(o => ({ id: o.orgId || o.id, name: o.orgName || o.name }));
+      const wanted = String(args.org).toLowerCase();
+      const byName = list.filter(o => (o.name || '').toLowerCase() === wanted);
+      const match = list.find(o => o.id === args.org) || (byName.length === 1 ? byName[0] : null);
+      if (match) orgCtx = match;
+    }
     const orgId = orgCtx?.id || null;
     // Mutation responses include `org` so the agent always sees where the
     // write landed — eliminates silent cross-org confusion.
@@ -3404,7 +3405,7 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
       // ── ls ──────────────────────────────────────────────────────
       case 'ls': {
         const { path: lsPath = '/', recursive: lsRecursive = false } = args;
-        const { pages, pathToPage } = await getTreeAsMap();
+        const { pages, pathToPage } = await getTreeAsMap(orgHeader);
         const parent = normalizeWikiPath(lsPath);
 
         if (lsRecursive) {
@@ -3462,7 +3463,7 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
       // ── recent ──────────────────────────────────────────────────
       case 'recent': {
         const { limit: recentLimit = 10 } = args;
-        const tree = await api('GET', '/api/wiki/tree');
+        const tree = await api('GET', '/api/wiki/tree', undefined, orgHeader);
         const recent = (tree.pages || [])
           .filter(p => p.updatedAt)
           .sort((a, b) => new Date(b.updatedAt) - new Date(a.updatedAt))
@@ -3558,7 +3559,7 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
         let existingContent = '';
         let existingId = null;
         try {
-          const logPage = await api('GET', '/api/wiki/page?path=log');
+          const logPage = await api('GET', '/api/wiki/page?path=log', undefined, orgHeader);
           existingContent = logPage.content || '';
           existingId = logPage.id;
         } catch {
@@ -3567,7 +3568,7 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
             path: 'log',
             title: 'Log',
             content: entry + '\n',
-          });
+          }, orgHeader);
           return ok(withOrg({ appended: true, created: true, pageId: created.id, path: 'log', url: wikiBrowserUrl('log') }));
         }
 
@@ -3579,7 +3580,7 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
 
       // ── health ──────────────────────────────────────────────────
       case 'health': {
-        return ok(await api('GET', '/api/wiki/health'));
+        return ok(await api('GET', '/api/wiki/health', undefined, orgHeader));
       }
 
       // ── write ───────────────────────────────────────────────────
@@ -3713,7 +3714,7 @@ tool('wiki', 'Per-org wiki. Markdown pages with paths as hierarchy. You and othe
           type: p.type,
           frontmatter: p.frontmatter,
         }));
-        const result = await api('POST', '/api/wiki/pages/bulk', { pages: payload });
+        const result = await api('POST', '/api/wiki/pages/bulk', { pages: payload }, orgHeader);
         return ok(withOrg({
           created: result.created?.length ?? 0,
           updated: result.updated?.length ?? 0,
@@ -3820,7 +3821,7 @@ tool('collector', {
   fresh: z.boolean().optional().describe('[test_start] start a brand-new run instead of resuming your latest'),
   actionId: z.string().optional().describe('[test_resolve] id of the pending action to resolve'),
   approve: z.boolean().optional().describe('[test_resolve] approve (default true) or reject the pending action'),
-  projectId: z.string().optional().describe('[create|meta] project to bind/scope to (defaults to the active project). Operates on the session’s active org — open the target project first via project(action="open"), or get_org(action="switch") to change org.'),
+  projectId: z.string().optional().describe('[create|meta] project to bind/scope to (defaults to the active project). The org derives from this project — open the target project first via project(action="open") if none is active.'),
   name: z.string().optional().describe('[create|update] collector name'),
   description: z.string().optional().describe('[create|update] one-line description shown to the consumer'),
   enabled: z.boolean().optional().describe('[create|update] whether the collector is live; a disabled collector 404s on its /c/<slug> link. enable/disable set this directly.'),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "drafted",
-  "version": "1.12.5",
+  "version": "1.12.7",
   "description": "Drafted — visual thinking surface for humans and AI agents. Renders HTML, markdown, images, and code as frames on a zoomable canvas, with MCP tools for AI agents and real-time sync for humans.",
   "type": "module",
   "files": [

package/plugin/commands/onboard-drafted.md

@@ -8,7 +8,7 @@ Onboard the user to Drafted — a producibles harness that compounds across thre
 This is the over-arching prime command: orient, then seed all three stores so every later session starts smart.
 
 1. **Orient** — in 3-4 lines explain the loop: you *prime* from the harness (the system makes you search the wiki, and auto-loads the project's attached skills, anchors, and layer rules before you work), you *build*, then you *compound* (deposit what you learned). The more it's used, the less searching and the more stable the work.
-2. **Confirm the org** — call `get_org`. Confirm the active organization is the one to build the harness in. If unsure, ask.
+2. **Confirm the org** — call `get_org` to list the user's orgs. Confirm which one the harness should be built in — if unsure, ask — then name it explicitly (org=...) on every create in the steps below.
 3. **Seed knowledge** — run the `/drafted:ingest` flow: help the user point at existing business materials (folders, docs, past research) or interrogate them for tacit knowledge, and land durable pages in the wiki.
 4. **Seed procedures** — from those materials and the conversation, surface 1-3 candidate SOPs. For the most valuable, run the `/drafted:create-skill` flow.
 5. **Seed the surface** — run the `/drafted:create-project` flow for the user's immediate piece of work (or a reusable template).