patchcord 0.5.130 → 0.5.132

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "patchcord",
   "description": "Cross-machine agent messaging. Messages from other agents land in the inbox and wake the agent to reply.",
-  "version": "0.5.130",
+  "version": "0.5.132",
   "author": {
     "name": "ppravdin"
   },

package/README.md

@@ -23,6 +23,9 @@ npx patchcord@latest --token <token> --server https://patchcord.yourdomain.com
 - **Stop hook** — checks inbox between turns, notifies of pending messages
 - **Slash commands** — `/patchcord` and `/patchcord-wait` for Codex and Gemini CLI
 - **MCP config** — per-project or global config depending on tool
+- **Cursor CLI** — bearer-only MCP endpoint, `patchcord:*` pre-allowed in
+  `~/.cursor/permissions.json`, and inbox/wait skills installed when
+  `cursor-agent` is detected
 - **Kimi CLI** — global MCP config plus skills in `~/.kimi/skills/`
 
 ## How it works

package/bin/patchcord.mjs

@@ -74,72 +74,76 @@ if (
   cmd === "help" || cmd === "--help" || cmd === "-h" ||
   process.argv.slice(3).some((a) => a === "--help" || a === "-h")
 ) {
-  console.log(`patchcord — cross-agent messaging + team orchestration for AI coding agents
-
-Two layers: THIS AGENT (a per-project identity, by bearer token) and your ACCOUNT
-(user-level; provisions agents, runs teams + schedules). Run \`patchcord login\`
-once to authenticate the CLI; account/team/schedule commands then use that token
-(they never open a browser on their own).
-
-SETUP (per agent / project)
-  patchcord                                  Set up this project's agent (browser)
-  patchcord --token <t> [--server <url>]     Headless / self-hosted setup
-  patchcord --full                           Same + full statusline
-  patchcord --rename <name> [--agent-type <t>]   Rename this agent
-
-THIS AGENT — identity + messaging
-  patchcord whoami                           Who am I — this agent's identity + project
-  patchcord whoami --propose "<text>"        Set your self-description (two-shot confirm)
-  patchcord agents [name]                    List agents in your namespace (or show one)
-  patchcord subscribe [interval]             Start the realtime listener (Kimi: bg poll)
-  patchcord upload <file> [--mime <t>] [--as <name>]   Share a file as an attachment
-
-ACCOUNT — log in to the CLI (commands below log in on demand)
-  patchcord login                            Authenticate the CLI (your account)
-
-ORCHESTRATOR — set up the orchestrator agent (the shepherd) in this folder
+  console.log(`usage: patchcord [command] [options]
+
+Cross-agent messaging + team orchestration for AI coding agents.
+
+Two layers: THIS AGENT (a per-project identity, by bearer token) and your
+ACCOUNT (user-level — provisions agents, runs teams + schedules). \`patchcord
+login\` authenticates the CLI once; account/team/schedule commands then use
+that token and never open a browser on their own.
+
+setup:
+  patchcord                                 set up this project's agent (browser)
+  patchcord --token <t> [--server <url>]    headless / self-hosted setup
+  patchcord --full                          same, with full statusline
+  patchcord --rename <name> [--agent-type <t>]   rename this agent
+
+agent:
+  patchcord whoami                          this agent's identity + project
+  patchcord whoami --propose "<text>"       set self-description (two-shot confirm)
+  patchcord agents [name]                   list agents in your namespace, or show one
+  patchcord subscribe [interval]            start the realtime listener
+  patchcord upload <file> [--mime <t>] [--as <name>]   share a file as an attachment
+
+account:
+  patchcord login                           authenticate the CLI (your account)
+
+orchestrator:
   patchcord orchestrator [--namespace <ns>] [--tool <harness>] [--project <name>]
-                                             Provision the orchestrator here (always a
-                                             lead — see PROJECTS below); launch it to
-                                             adopt this project or build a new team
-
-TEAM — the orchestrator's tools (run from the project root)
-  patchcord provision <agent> --tool <X> --role <Y> --namespace <ns> [--dir <sub/>] \\
-        [--project <name>] [--lead]
-                                             Create a worker agent (identity + config).
-                                             --lead marks it as a cross-namespace lead
-                                             (--role lead does the same); --project
-                                             assigns its namespace (see PROJECTS below)
+                                             provision the orchestrator here (always
+                                             a lead); launch it to adopt this project
+                                             or build a new team
+
+team (run from the project root):
+  patchcord provision <agent> --tool <X> --role <Y> --namespace <ns>
+        [--dir <sub/>] [--project <name>] [--lead]
+                                             create a worker agent (identity + config)
   patchcord pull <agent> --tool <X> --namespace <ns> [--dir <sub/>]
-                                             Place an EXISTING agent into a folder
-                                             (same identity, new token; nothing overwritten)
+                                             move an existing agent into a folder
+                                             (same identity, new token)
   patchcord provision revoke <agent> --namespace <ns>
-  patchcord team list [--namespace <ns>]     Provisioned agents (server view, deduped)
-  patchcord team status                      Reconcile folder ↔ identity ↔ tmux ↔ token
-  patchcord team launch                      Launch each worker in mux
-
-PROJECTS — cloud-only. Group your OWN namespaces so their leads may cross-message
-  (same user + same project + BOTH ends provisioned as leads → agent@namespace works;
-  anything else is always denied, no exceptions). Namespace names are unique across
-  ALL your projects — you can't reuse a name in a different project.
-  patchcord namespace set-project <namespace> <project>
-                                             Assign a namespace to a project
-  patchcord namespace set-project <namespace>
-                                             Clear it (namespace becomes isolated again)
-
-SCHEDULES — timed / recurring messages (user-level)
-  patchcord schedule create <name> --namespace <ns> --to <agent> --content "..." \\
-        (--at <ISO> | --cron "<expr>" | --every <sec>) \\
+                                             revoke a worker agent
+  patchcord team list [--namespace <ns>]    provisioned agents (server view)
+  patchcord team status                     reconcile folder ↔ identity ↔ tmux ↔ token
+  patchcord team launch                     launch each worker in mux
+
+projects (cloud-only — a free-text label grouping your own namespaces):
+  patchcord namespace set-project <namespace> <project>   assign a project label
+  patchcord namespace set-project <namespace>              clear it
+
+cross-namespace links (cloud-only, no CLI — leads request/approve via MCP:
+  request_namespace_link / respond_namespace_link / list_my_namespace_links.
+  Both namespaces need a --lead agent; links are opt-in, not project-based.)
+
+schedules:
+  patchcord schedule create <name> --namespace <ns> --to <agent> --content "..."
+        (--at <ISO> | --cron "<expr>" | --every <sec>)
         [--timezone <tz>] [--thread <slug>] [--max-runs N] [--expires <ISO>]
   patchcord schedule list [--namespace <ns>]
   patchcord schedule cancel|test|pause|resume <id>
 
-MISC
-  patchcord update                           Update to the latest version
-  patchcord --version                        Show installed version
-  patchcord --help                           Show this help
+misc:
+  patchcord update                          update to the latest version
+  patchcord --version                       show installed version
+  patchcord --help                          show this help
 
-First install: npx patchcord@latest`);
+examples:
+  patchcord provision codex --tool codex --role maker --namespace myproj
+  patchcord schedule create standup --namespace myproj --to lead \\
+        --content "status?" --cron "0 9 * * *"
+  patchcord team status
+`);
   process.exit(0);
 }
 
@@ -1004,6 +1008,19 @@ if (cmd === "login" || cmd === "orchestrator" || cmd === "teamlead" || cmd === "
         o.mcp.patchcord = { type: "remote", url: `${baseUrl}/mcp`, headers: hdr };
       });
     }
+    if (tool === "cursor") {
+      // Cursor IDE + Cursor CLI (cursor-agent) share the same project config —
+      // same file, same schema (see patchcord.dev/console/connect writer).
+      const cdir = join(dir, ".cursor"); mkdirSync(cdir, { recursive: true });
+      return writeJson(join(cdir, "mcp.json"), (o) => {
+        o.mcpServers = o.mcpServers || {};
+        o.mcpServers.patchcord = {
+          type: "streamable-http",
+          url: `${baseUrl}/mcp/bearer`,
+          headers: hdr,
+        };
+      });
+    }
     if (tool === "antigravity" || tool === "agy") {
       // Antigravity CLI: PROJECT-scoped config at .agents/mcp_config.json (one
       // project = one namespace = one agent). Remote HTTP uses `serverUrl`.
@@ -1071,11 +1088,13 @@ if (cmd === "login" || cmd === "orchestrator" || cmd === "teamlead" || cmd === "
     const ns = flagVal("namespace");
     const subdir = flagVal("dir", arg);
     const project = flagVal("project", "");
-    // --lead marks this token as a project-scoped cross-namespace lead (server
+    // --lead marks this token as eligible for cross-namespace lead chat (server
     // trusts ONLY this explicit, main-token-authenticated request — see
-    // patchcord-cloud's api_provision). `--role lead` also sets it server-side
-    // (role==="lead" is treated the same as is_lead:true), but `--lead` is the
-    // explicit, discoverable way to ask for it without overloading the
+    // patchcord-cloud's api_provision). Reaching another namespace's lead also
+    // requires an approved namespace_links pair (via request_namespace_link /
+    // respond_namespace_link — no CLI, agent-to-agent only). `--role lead` also
+    // sets is_lead server-side (role==="lead" is treated the same), but `--lead`
+    // is the explicit, discoverable way to ask for it without overloading the
     // free-text --role field.
     const lead = process.argv.includes("--lead");
     if (!ns) { console.error("--namespace <project-namespace> required"); process.exit(1); }
@@ -1100,12 +1119,12 @@ if (cmd === "login" || cmd === "orchestrator" || cmd === "teamlead" || cmd === "
         writeFileSync(tj, JSON.stringify(man, null, 2) + "\n");
       }
     } catch {}
-    // --project assigns this namespace to a project — the cross-namespace lead
-    // chat boundary (namespaces in the same project may link leads; different
-    // projects, or no project, never can). Separate call: provisioning creates
-    // the agent identity, project assignment groups the NAMESPACE — safe to
-    // retry independently, and a namespace already in a project just gets
-    // reassigned (last call wins, server-side upsert).
+    // --project assigns this namespace to a project — an organizational label
+    // ONLY, does not affect cross-namespace messaging (that's namespace_links,
+    // agent-to-agent, no CLI). Separate call: provisioning creates the agent
+    // identity, project assignment groups the NAMESPACE — safe to retry
+    // independently, and a namespace already in a project just gets reassigned
+    // (last call wins, server-side upsert).
     if (project) {
       const { status: pStatus, json: pJson } = await accountCall("POST", "/api/namespace/set-project", { namespace_id: ns, project });
       if (pStatus !== "200") {
@@ -1119,22 +1138,22 @@ if (cmd === "login" || cmd === "orchestrator" || cmd === "teamlead" || cmd === "
   }
 
   if (cmd === "namespace") {
-    // Cloud-only: group namespaces into a "project" — the boundary
-    // cross-namespace lead chat is gated on. Same user + same project + both
-    // ends are provisioned leads (--lead / --role lead) → their leads may
-    // message each other with `agent@namespace`. Different project, or no
-    // project, or a non-lead agent → always denied, no exceptions.
+    // Cloud-only: group namespaces into a "project" — a free-text organizational
+    // label only. Does NOT gate cross-namespace lead chat — that's decided by
+    // per-pair namespace_links, approved lead-to-lead via the MCP tools
+    // request_namespace_link / respond_namespace_link (no CLI for that; see
+    // --help's CROSS-NAMESPACE LINKS section).
     const sub = process.argv[3];
     if (sub === "set-project") {
       const ns = process.argv[4];
       const project = process.argv[5];
-      if (!ns) { console.error("Usage: patchcord namespace set-project <namespace> [project-name]\n  Omit project-name to clear (un-groups the namespace, reverting it to isolated)."); process.exit(1); }
+      if (!ns) { console.error("Usage: patchcord namespace set-project <namespace> [project-name]\n  Omit project-name to clear the label."); process.exit(1); }
       const { status, json } = await accountCall("POST", "/api/namespace/set-project", { namespace_id: ns, project: project || "" });
       if (status !== "200") { console.error(`set-project failed (HTTP ${status}): ${json?.error || ""}`); process.exit(1); }
       if (json.status === "cleared") {
-        console.log(`✓ ${ns} un-grouped — no project, isolated (no cross-namespace lead peers)`);
+        console.log(`✓ ${ns} un-grouped — no project label`);
       } else {
-        console.log(`✓ ${ns} assigned to project "${json.project}"`);
+        console.log(`✓ ${ns} assigned to project "${json.project}" (label only — does not affect cross-namespace messaging)`);
       }
       process.exit(0);
     }
@@ -1188,8 +1207,10 @@ if (cmd === "login" || cmd === "orchestrator" || cmd === "teamlead" || cmd === "
     const project = flagVal("project", "");
     const hostname = run("hostname -s") || run("hostname") || "unknown";
     // is_lead: true always — the orchestrator IS the one identity per namespace
-    // meant to coordinate across teams. Harmless without --project: a lead with
-    // no project assignment has zero cross-namespace peers (fails closed).
+    // meant to coordinate across teams. Harmless by itself: a lead only reaches
+    // another namespace's lead after an approved namespace_links pair (agent-to-
+    // agent via request_namespace_link/respond_namespace_link — fails closed
+    // until both leads explicitly consent).
     const { status, json, m } = await accountCall("POST", "/api/provision", { namespace_id: ns, agent_id: "orchestrator", tool, role: "orchestrator", label: "orchestrator:self", is_lead: true });
     if (status !== "200" || !json?.token) { console.error(`could not set up orchestrator (HTTP ${status}): ${json?.error || ""}`); process.exit(1); }
     const base = String(json.url || `${m.baseUrl}/mcp`).replace(/\/mcp$/, "");
@@ -1722,9 +1743,14 @@ if (!cmd || cmd === "install" || cmd === "agent" || cmd?.startsWith("--")) {
     }
   }
 
-  // Cursor
+  // Cursor.  Cursor CLI (cursor-agent) can be installed without creating the
+  // IDE's skills directory, so detect the executable as well as the directory.
+  // The bearer-only endpoint prevents Cursor from incorrectly starting OAuth
+  // discovery when the installer has supplied a static bearer header.
   const cursorSkillsRoot = join(HOME, ".cursor", "skills-cursor");
-  if (existsSync(cursorSkillsRoot)) {
+  const hasCursorAgent = run("which cursor-agent");
+  if (hasCursorAgent || existsSync(cursorSkillsRoot)) {
+    mkdirSync(cursorSkillsRoot, { recursive: true });
     const cursorSkillDir = join(cursorSkillsRoot, "patchcord");
     const cursorWaitDir = join(cursorSkillsRoot, "patchcord-wait");
     let cursorChanged = false;
@@ -1739,6 +1765,27 @@ if (!cmd || cmd === "install" || cmd === "agent" || cmd?.startsWith("--")) {
       cursorChanged = true;
     }
     if (cursorChanged) globalChanges.push("Cursor skills installed");
+
+    if (hasCursorAgent) {
+      const cursorPermissionsPath = join(HOME, ".cursor", "permissions.json");
+      let allowlistChanged = false;
+      const permissionsOk = updateJsonConfig(cursorPermissionsPath, (obj) => {
+        if (obj.mcpAllowlist === undefined) obj.mcpAllowlist = [];
+        if (!Array.isArray(obj.mcpAllowlist)) {
+          console.log(`\n  ${yellow}⚠${r} Skipped Cursor MCP allowlist — ${cursorPermissionsPath} has a non-array mcpAllowlist.`);
+          return;
+        }
+        if (!obj.mcpAllowlist.includes("patchcord:*")) {
+          obj.mcpAllowlist.push("patchcord:*");
+          allowlistChanged = true;
+        }
+      });
+      if (!permissionsOk) {
+        globalChanges.push("✗ Cursor permissions error");
+      } else if (allowlistChanged) {
+        globalChanges.push("Cursor MCP allowlist configured");
+      }
+    }
   }
 
   // Windsurf
@@ -2074,8 +2121,8 @@ if (!cmd || cmd === "install" || cmd === "agent" || cmd?.startsWith("--")) {
     }
   }
 
-  if (!hasClaude && !existsSync(codexConfig) && !hasKimi) {
-    console.log(`${dim}No Claude Code, Codex CLI, or Kimi Code detected — skipping global setup.${r}`);
+  if (!hasClaude && !existsSync(codexConfig) && !hasKimi && !hasCursorAgent && !existsSync(cursorSkillsRoot)) {
+    console.log(`${dim}No Claude Code, Codex CLI, Cursor, or Kimi Code detected — skipping global setup.${r}`);
     console.log(`${dim}  Kimi Code: https://kimi.moonshot.cn/download${r}`);
   }
 
@@ -2142,21 +2189,17 @@ if (!cmd || cmd === "install" || cmd === "agent" || cmd?.startsWith("--")) {
     || (flags.includes("--token") ? flags[flags.indexOf("--token") + 1] : "");
 
   if (tokenFlag) {
-    // --token bypass: need tool picker in terminal — unless --tool=<slug>
-    // already pre-selected one for us.
+    // --token bypass requires --tool=<slug> up front — no terminal picker.
+    // The web connect page (patchcord.dev/console/connect/<platform>) is
+    // where the tool is chosen; it hands back the full npx command with
+    // --tool=<slug> already baked in. If someone lands here without one,
+    // point them at the web flow instead of prompting.
     if (!choice) {
-      console.log(`\n${bold}Which tool are you setting up?${r}\n`);
-      console.log(`  ${cyan}1.${r} Claude Code   ${cyan}5.${r} Gemini CLI   ${cyan}9.${r} OpenClaw`);
-      console.log(`  ${cyan}2.${r} Codex CLI     ${cyan}6.${r} VS Code      ${cyan}10.${r} Antigravity CLI`);
-      console.log(`  ${cyan}3.${r} Cursor        ${cyan}7.${r} Zed          ${cyan}11.${r} Cline`);
-      console.log(`  ${cyan}4.${r} Windsurf      ${cyan}8.${r} OpenCode     ${cyan}12.${r} Kimi Code`);
-      console.log(`  ${cyan}13.${r} Hermes\n`);
-      choice = (await ask(`${dim}Choose (1-13):${r} `)).trim();
-      if (!["1","2","3","4","5","6","7","8","9","10","11","12","13"].includes(choice)) {
-        console.error("Invalid choice.");
-        rl.close();
-        process.exit(1);
-      }
+      console.error(`${bold}--tool=<slug> is required with --token.${r}`);
+      console.error(`${dim}Pick your tool at https://patchcord.dev/console/connect — it gives you the exact command.${r}`);
+      console.error(`${dim}Valid slugs: ${Object.keys(CLIENT_TYPE_MAP).join(", ")}${r}`);
+      rl.close();
+      process.exit(1);
     }
     token = tokenFlag.trim();
     if (!isSafeToken(token)) {
@@ -2506,7 +2549,8 @@ if (!cmd || cmd === "install" || cmd === "agent" || cmd?.startsWith("--")) {
     const cursorConfig = {
       mcpServers: {
         patchcord: {
-          url: `${serverUrl}/mcp`,
+          type: "streamable-http",
+          url: `${serverUrl}/mcp/bearer`,
           headers: {
             Authorization: `Bearer ${token}`,
             "X-Patchcord-Machine": hostname,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "patchcord",
-  "version": "0.5.130",
+  "version": "0.5.132",
   "description": "Cross-machine agent messaging for Claude Code and Codex",
   "scripts": {
     "version": "node scripts/sync-plugin-version.mjs && git add .claude-plugin/plugin.json"

package/skills/inbox/SKILL.md

@@ -184,7 +184,7 @@ Named threads group related messages between a pair of agents. Use them for mult
 - Never show raw JSON to the human - summarize naturally.
 - **Cross-namespace addressing (`agent@namespace`)**: the syntax always exists in `send_message`/`reply`, but what it actually reaches depends on who you are:
   - **Ordinary agents (the default):** `agent@namespace` only ever works for YOUR OWN namespace — same as a bare name. Targeting any other namespace is rejected. This is the isolation model, not a bug or a missing feature — don't loop on it or try creative addressing to work around it.
-  - **Lead agents (cloud only — a small number of specially-provisioned identities per namespace, usually the orchestrator):** may reach a peer lead in a DIFFERENT namespace, but only if a human explicitly grouped both namespaces into the same "project" (a cloud-side setting you cannot set yourself from inside a session) — different project, or no project, is always rejected, no exceptions. If you don't know whether you're a lead, you're not one — this isn't discoverable by trial and error; ask the human.
+  - **Lead agents (cloud only — a small number of specially-provisioned identities per namespace, usually the orchestrator):** may reach a peer lead in a DIFFERENT namespace, but only after that specific pair of namespaces has an `approved` `namespace_links` row. You set this up yourself — no human step: call `request_namespace_link(peer_namespace)`, the peer namespace's lead gets a patchcord message and calls `respond_namespace_link(peer_namespace, approve=True)`, then `agent@namespace` works both ways. Check `list_my_namespace_links()` to see pending/approved/denied status. If you don't know whether you're a lead, you're not one — this isn't discoverable by trial and error; ask the human. (Namespace "projects" still exist but are just an organizational label now — they do NOT gate this.)
   - A rejection here is almost always correct behavior, not a server error — don't retry with variations of the name.
 - **Do not reply to acks.** "ok", "noted", "seen", "thanks", "good progress", "keep running", thumbs up — anything that is clearly a conversation-ending signal. Just read them and move on. If you must close the thread, use `reply(id, resolve=true)` with NO content. Never send a text reply to an ack.
 - **resolve=true with ack-only content is an anti-pattern.** `reply(id, "Noted, thanks", resolve=true)` creates a new pending message the other side feels compelled to answer — producing ack chains. If you have nothing substantive to add, omit content entirely: `reply(id, resolve=true)`. Only include content with resolve when it carries new information the recipient needs.