@open-everywhere/create 0.1.4

package/template/claude/skills/upgrade-open-everywhere/SKILL.md

@@ -0,0 +1,160 @@
+---
+name: upgrade-open-everywhere
+description: Upgrade this Open Everywhere app to a newer release — read the release notes between the installed and the target version, run `oe upgrade`, review the roster.lock.json diff, open a pull request, and confirm the deployment after it ships. Use when the user wants to upgrade, update, or take a new version of Open Everywhere, or asks what is in a newer release.
+---
+
+# Upgrade Open Everywhere
+
+One procedure, two drivers. You run it here, in a checkout of this app. The agent runs the same
+steps on itself, in its own sandbox, against this same repository. Nothing below differs between
+the two.
+
+The procedure is `npx oe upgrade`. It is deterministic: it reads the release record that ships
+inside `@open-everywhere/core`, bumps every package together, installs, applies the config
+codemods the notes name, rebuilds the roster and proves the tree still builds.
+
+## Golden rules
+
+- **`oe upgrade` writes and installs.** Show the command, get a yes, then run it. `--plan` writes
+  nothing, so run that first, every time.
+- **Never merge the pull request yourself.** An admin merges it, and the merge deploys.
+- **Never hand-edit `roster.lock.json`.** `oe build --accept` writes it. Your job is to READ the
+  diff: every line is a capability decision, and a new tool that nobody looked at is how an
+  upgrade quietly widens what the agent can do.
+- **Never edit `UPGRADING.md`.** It is generated from `upgrading.json` beside it.
+- **No secret reaches the chat, the branch or the pull request.** An upgrade never needs one. A
+  release that adds an env NAME tells the admin to set the value in the deployment's environment.
+- Every command runs from this directory (the app root).
+
+## 1. Where you are, and where you are going
+
+```
+node -p "require('./node_modules/@open-everywhere/core/package.json').version"  # installed
+node -p "require('./package.json').dependencies['@open-everywhere/core']"       # pinned
+npm view @open-everywhere/core version                                          # latest published
+```
+
+**Read the installed one, not the pin.** `oe upgrade` writes the pins before it installs, so an
+upgrade that stopped part way leaves a manifest that names a version the app is not running.
+
+The nine packages publish in lockstep, so one version answers for all of them. Agree the target
+with the user before you go on. Taking the latest is the normal choice.
+
+## 2. Read the notes between the two versions
+
+```
+npx oe upgrade --to <target> --plan
+```
+
+It prints the window — every release above the installed version, up to the target — and writes
+nothing. Each entry names:
+
+| Field | What it means for this deployment |
+| --- | --- |
+| BREAKING | behaviour that changes. Read it before anything else. |
+| CONFIG | a change to this app's own config files. A codemod applies some; the rest are yours. |
+| ENV + / ENV - | an env NAME the deployment starts or stops needing. The admin sets the value. |
+| MANUAL | a step outside the files — a dashboard change, a URL to re-paste, a key to rotate. |
+
+The installed copy of the record cannot know about a release published after it, so `--plan`
+says plainly which crossed versions it has no entry for. The full window is printed again in
+step 3, after the install puts the target's own copy on disk. The same text is in
+`node_modules/@open-everywhere/core/UPGRADING.md`.
+
+**Stop and ask the user** when the window holds a breaking change, an env change or a manual step.
+Those are decisions, not steps.
+
+## 3. Run the upgrade
+
+```
+npx oe upgrade --to <target>
+```
+
+In order, and it stops at the first step that fails:
+
+1. bump every `@open-everywhere/*` package to the target, renaming a dependency on the retired
+   unscoped `create-open-everywhere` to `@open-everywhere/create` as it goes;
+2. pin `eve` to the version the target core declares as its peer;
+3. `npm install` — its `postinstall` runs `oe-eve-patch`, which the app refuses to boot without;
+4. run the config codemods the target's notes name, and print the migrations that are yours;
+5. `oe build --accept` — this rewrites `roster.lock.json`;
+6. `npm run typecheck`;
+7. `npm run build`.
+
+A failed step prints which one and leaves the tree where it stopped. Fix the cause and run
+`oe upgrade` again: it RESUMES, from whichever step stopped it.
+
+It can do that because it writes `.open-everywhere/upgrade-resume.json` just BEFORE the install
+and deletes it only when the run completes. Before, not after: a failing `npm install` can still
+leave the new packages on disk, because npm writes the tree first and runs `postinstall` after.
+That file records the version the run started from, so the retry reads the SAME release notes,
+runs the same codemods and takes the same refusals — a retry never quietly forgets a migration
+you still owe. The directory is gitignored, so the marker never reaches a pull request, and it
+survives `rm -rf node_modules && npm ci`.
+
+Never delete the marker to make `oe upgrade` say there is nothing to do. It is the only record
+that a run began taking the target and never proved the tree.
+
+**If it refuses with "the release notes name codemods this CLI does not hold":** the codemod ships
+in a newer core than the one you are running. Upgrade in ONE step to a version that holds it, or
+apply those migrations by hand. Do not work around the refusal — it exists because a half-applied
+migration that reports success is worse than a stop.
+
+## 4. Review the diff
+
+```
+git diff roster.lock.json
+git diff package.json
+git status
+```
+
+Read `roster.lock.json` line by line. A new mounted tool, channel, schedule or connection is a new
+thing the agent can do. If a line surprises you, say so to the user before the pull request — this
+is the review surface, and it is the only one.
+
+Then run the app's own checks once more if anything above was fixed by hand:
+`npm run typecheck && npm run build`.
+
+## 5. Open a pull request
+
+Commit on a branch, never on the default branch. The body carries:
+
+- the versions: `<installed> → <target>`, and the `eve` pin;
+- the release-note window, as `oe upgrade` printed it;
+- what changed in `roster.lock.json`, in your own words;
+- **the env names the deployment needs before this merges**, names only;
+- the manual steps, and who does each one.
+
+Then stop. The admin reviews the lock diff and merges. The merge deploys.
+
+## 6. Confirm the deployment after it ships
+
+A green build is not a working deployment.
+
+**The status endpoint is not built yet.** OPE-2230 adds one: an unauthenticated, non-sensitive
+JSON route that reports the deployment's version, whether the database is reachable and which
+provider it uses, the sandbox backend, whether egress is enforced, whether the model route
+answers, and the Slack posture. That ticket proposes `GET /eve/v1/oe/status`. Do not assume the
+route exists, and never read some other route as if it were this one. Check the deployment's own
+routes first; when the endpoint is there, read it and report each field, in this order:
+
+1. **version** — it reads the target. If it still reads the old version, the deploy has not
+   finished, or it failed. Nothing below is meaningful until this matches.
+2. **database reachable**, and the provider it names.
+3. **sandbox backend**, and **egress enforced**.
+4. **model route answers.**
+5. **Slack** — credentials present, workspace installed, events verified.
+
+**Until that endpoint ships**, confirm what you can and say plainly what you could not:
+
+- the deploy finished, from the host's own deployment log;
+- `npx @open-everywhere/create --check --json --host https://<host>` — its `build-deploy` step
+  passes only on the deployed Slack route's own `401 unauthorized`, which proves boot and
+  routing;
+- Slack's **Event Subscriptions** still shows **Verified**.
+
+A fact that went from good to bad across the upgrade is the upgrade's fault until proven
+otherwise: say so, and offer to roll the pull request back rather than patching forward on a live
+deployment.
+
+Last, ask the user to send the app a direct message. An answer is the only proof that holds.

package/template/gitignore

@@ -0,0 +1,42 @@
+# Dependencies
+node_modules/
+
+# Secrets. Keep .env.example: it carries variable names only.
+# `.env*` and `.vercel` are the exact lines the Vercel CLI adds, so `vercel link` leaves this file alone.
+.env*
+!.env.example
+
+# The Vercel project link
+.vercel
+
+# eve build output and local durable state
+.eve/
+.output/
+.workflow-data/
+
+# Local stores: the embedded operational database, the memory store, the wizard state
+.open-everywhere-data/
+.open-everywhere-data.lock
+*.pglite.lock
+.open-everywhere-memory/
+.open-everywhere/
+*.sqlite3
+*.sqlite3-wal
+*.sqlite3-shm
+
+# The eve slot tree is `oe build` output, generated from roster.config.ts + roster.lock.json.
+# Commit those two files, never the tree.
+/agent/tools/
+/agent/channels/
+/agent/hooks/
+/agent/schedules/
+/agent/skills/
+/agent/connections/
+/agent/lib/providers.generated.ts
+
+# Misc
+*.log
+.DS_Store
+*.tsbuildinfo
+*.pem
+*.key

package/template/package.json

@@ -0,0 +1,62 @@
+{
+  "name": "{{APP_NAME}}",
+  "version": "0.1.0",
+  "private": true,
+  "type": "module",
+  "engines": {
+    "node": "24.x"
+  },
+  "scripts": {
+    "postinstall": "oe-eve-patch",
+    "oe:build": "oe build",
+    "predev": "oe build",
+    "dev": "eve dev",
+    "prebuild": "oe build",
+    "build": "eve build",
+    "prestart": "oe build",
+    "start": "eve start",
+    "pretypecheck": "oe build",
+    "typecheck": "tsc --noEmit -p tsconfig.json",
+    "check": "create-open-everywhere --check",
+    "ops": "node --import tsx scripts/ops-mcp.ts",
+    "env:pipe": "node scripts/env-pipe.mjs"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.29.0",
+    "@open-everywhere/adapters": "{{OE_VERSION}}",
+    "@open-everywhere/core": "{{OE_VERSION}}",
+    "@open-everywhere/provider-discord": "{{OE_VERSION}}",
+    "@open-everywhere/provider-github": "{{OE_VERSION}}",
+    "@open-everywhere/provider-linear": "{{OE_VERSION}}",
+    "@open-everywhere/provider-notion": "{{OE_VERSION}}",
+    "@open-everywhere/provider-slack": "{{OE_VERSION}}",
+    "@open-everywhere/provider-teams": "{{OE_VERSION}}",
+    "eve": "0.54.3",
+    "just-bash": "^3.1.0"
+  },
+  "devDependencies": {
+    "@open-everywhere/create": "{{OE_VERSION}}",
+    "@types/node": "^24.0.0",
+    "tsx": "^4.23.9",
+    "typescript": "^7.0.2"
+  },
+  "overrides": {
+    "zod": "4.5.4",
+    "@agent-browser/eve": {
+      "zod": "4.5.4"
+    },
+    "eve": {
+      "undici": "8.11.2",
+      "@ai-sdk/provider-utils": {
+        "undici": "7.30.0"
+      }
+    },
+    "guarded-fetch": {
+      "undici": "7.30.0"
+    }
+  },
+  "allowScripts": {
+    "esbuild": true,
+    "@mongodb-js/zstd": true
+  }
+}

package/template/roster.config.ts

@@ -0,0 +1,39 @@
+import { defineRoster } from "@open-everywhere/core/agent/scaffold/roster-config.js";
+
+// THE ROSTER — what this agent can do. `oe build` reads this file and roster.lock.json, and writes
+// the eve slot tree (agent/tools/, agent/channels/, agent/hooks/, …). The tree is build output and
+// is not committed. This file and the lock are committed: they are the review surface.
+//
+// The lock is the REVIEWED roll-call. When an upgrade adds a contribution, `oe build` refuses until
+// you decide about it here, then `npx oe build --accept` records the decision. Review the
+// roster.lock.json diff like any other code change.
+//
+// `oe build` loads this file with Node's native type stripping, so keep it ERASABLE syntax only: no
+// `enum`, no `namespace`, and `import type` for type-only imports.
+
+export default defineRoster({
+  // eve's BUILT-INS. Absence KEEPS a built-in. Every line here takes a capability AWAY.
+  framework: {
+    disable: {
+      // Core re-provides these four per session grain (task sessions mount them, standing seats do
+      // not). The sentinel is what makes that provision authoritative: a dynamic tool can only
+      // shadow a static built-in, never remove it.
+      bash: "re-provided per session grain by core (tools/sandbox_tools.ts)",
+      read_file: "re-provided per session grain by core (tools/sandbox_tools.ts)",
+      web_fetch: "re-provided per session grain by core (tools/sandbox_tools.ts)",
+      write_file: "re-provided per session grain by core (tools/sandbox_tools.ts)",
+    },
+  },
+
+  // The portable runtime and the reference Vercel adapters: mount everything they offer.
+  core: true,
+  adapters: true,
+
+  // One provider bundle per surface. Add a line to add a surface (`discord`, `teams` and `linear`
+  // are installed too), then `npx oe build --accept`.
+  providers: {
+    slack: true,
+    notion: true,
+    github: true,
+  },
+});

package/template/roster.lock.json

@@ -0,0 +1,182 @@
+{
+  "schemaVersion": 1,
+  "generator": 1,
+  "framework": {
+    "bash": "disabled",
+    "read_file": "disabled",
+    "web_fetch": "disabled",
+    "write_file": "disabled"
+  },
+  "packages": {
+    "@open-everywhere/adapters": {
+      "version": "{{OE_VERSION}}",
+      "contributions": {
+        "hook/sandbox-lifecycle": "mounted",
+        "tool/extend_workspace_time": "mounted"
+      }
+    },
+    "@open-everywhere/core": {
+      "version": "{{OE_VERSION}}",
+      "contributions": {
+        "channel/aws-sso-callback": "mounted",
+        "channel/control-mcp": "mounted",
+        "channel/control-read": "mounted",
+        "channel/credential-intake-page": "mounted",
+        "channel/oe-oauth-callback": "mounted",
+        "channel/oe-status": "mounted",
+        "channel/policy-pages": "mounted",
+        "channel/test-wake": "mounted",
+        "channel/transcript-relay": "mounted",
+        "channel/unified": "mounted",
+        "channel/webhook-reveal-page": "mounted",
+        "channel/webhooks": "mounted",
+        "connection/notion": "mounted",
+        "hook/approval-raised": "mounted",
+        "hook/approval-settlement": "mounted",
+        "hook/binding-bridge": "mounted",
+        "hook/connector-audit": "mounted",
+        "hook/egress": "mounted",
+        "hook/eve-patch-gate": "mounted",
+        "hook/fire-carriage": "mounted",
+        "hook/fire-toolset": "mounted",
+        "hook/job-posture": "mounted",
+        "hook/lean-seat": "mounted",
+        "hook/learn": "mounted",
+        "hook/link-claim": "mounted",
+        "hook/log-flush": "mounted",
+        "hook/loop-breaker": "mounted",
+        "hook/memory": "mounted",
+        "hook/native-input-audit": "mounted",
+        "hook/own-account-ask": "mounted",
+        "hook/requester-identity": "mounted",
+        "hook/requester-messages": "mounted",
+        "hook/routine-outcome": "mounted",
+        "hook/routine-silence": "mounted",
+        "hook/session-park": "mounted",
+        "hook/spawn-taint": "mounted",
+        "hook/spend": "mounted",
+        "hook/stall-guard": "mounted",
+        "hook/telemetry": "mounted",
+        "hook/turn-anchor": "mounted",
+        "hook/wake-accounting-turn": "mounted",
+        "hook/wake-turn-start": "mounted",
+        "hook/window-routing": "mounted",
+        "schedule/directory-sync": "mounted",
+        "schedule/dispatch": "mounted",
+        "schedule/dream-mode3": "mounted",
+        "schedule/tidy": "mounted",
+        "skill/scoped": "mounted",
+        "tool/add_label": "mounted",
+        "tool/add_routine_rule": "mounted",
+        "tool/agent_version": "mounted",
+        "tool/assign": "mounted",
+        "tool/aws": "mounted",
+        "tool/browser_tools": "mounted",
+        "tool/cancel_routine": "mounted",
+        "tool/capture_events": "mounted",
+        "tool/check_pr_status": "mounted",
+        "tool/checkout_repo": "mounted",
+        "tool/clear_connector_override": "mounted",
+        "tool/clear_spend_limit": "mounted",
+        "tool/complete_job": "mounted",
+        "tool/complete_thread": "mounted",
+        "tool/confirm_link": "mounted",
+        "tool/connector": "mounted",
+        "tool/create_agent": "mounted",
+        "tool/create_routine": "mounted",
+        "tool/create_skill": "mounted",
+        "tool/create_ticket": "mounted",
+        "tool/declare_webhook": "mounted",
+        "tool/delete_skill": "mounted",
+        "tool/describe_image": "mounted",
+        "tool/dm_person": "mounted",
+        "tool/edit_reply": "mounted",
+        "tool/edit_routine": "mounted",
+        "tool/enable_routine": "mounted",
+        "tool/explain_permission": "mounted",
+        "tool/get_pr_feedback": "mounted",
+        "tool/grant_tools": "mounted",
+        "tool/halt": "mounted",
+        "tool/link_identity": "mounted",
+        "tool/list_recent_events": "mounted",
+        "tool/list_repos": "mounted",
+        "tool/list_routines": "mounted",
+        "tool/list_skills": "mounted",
+        "tool/list_webhooks": "mounted",
+        "tool/memory_admin_list": "mounted",
+        "tool/memory_admin_read": "mounted",
+        "tool/message_thread_seat": "mounted",
+        "tool/no_reply": "mounted",
+        "tool/open_pull_request": "mounted",
+        "tool/permission_fit_report": "mounted",
+        "tool/post_file": "mounted",
+        "tool/post_message": "mounted",
+        "tool/post_pr_walkthrough": "mounted",
+        "tool/purge_memory": "mounted",
+        "tool/query_analytics": "mounted",
+        "tool/react": "mounted",
+        "tool/read_channel": "mounted",
+        "tool/read_docs": "mounted",
+        "tool/read_file_upload": "mounted",
+        "tool/read_routine_transcript": "mounted",
+        "tool/read_thread": "mounted",
+        "tool/read_thread_seat": "mounted",
+        "tool/read_webhook_history": "mounted",
+        "tool/read_work_channel": "mounted",
+        "tool/read_work_thread": "mounted",
+        "tool/record_follow_up": "mounted",
+        "tool/record_job_state": "mounted",
+        "tool/record_watch_state": "mounted",
+        "tool/remove_routine_rule": "mounted",
+        "tool/reply": "mounted",
+        "tool/report_run_failed": "mounted",
+        "tool/request_access": "mounted",
+        "tool/request_credential_grant": "mounted",
+        "tool/resume": "mounted",
+        "tool/revoke_connector": "mounted",
+        "tool/revoke_credential_grant": "mounted",
+        "tool/revoke_routine_permissions": "mounted",
+        "tool/rotate_webhook": "mounted",
+        "tool/run_routine_now": "mounted",
+        "tool/sandbox_tools": "mounted",
+        "tool/search": "mounted",
+        "tool/session_cost": "mounted",
+        "tool/set_dm_delivery": "mounted",
+        "tool/set_spend_limit": "mounted",
+        "tool/set_state": "mounted",
+        "tool/spawn_thread_session": "mounted",
+        "tool/spend_report": "mounted",
+        "tool/stand_down": "mounted",
+        "tool/subscribe_pr": "mounted",
+        "tool/unlink_identity": "mounted",
+        "tool/update_status": "mounted",
+        "tool/usage_report": "mounted",
+        "tool/walk_pr": "mounted"
+      }
+    },
+    "@open-everywhere/provider-github": {
+      "version": "{{OE_VERSION}}",
+      "contributions": {
+        "channel/github": "mounted"
+      }
+    },
+    "@open-everywhere/provider-notion": {
+      "version": "{{OE_VERSION}}",
+      "contributions": {
+        "channel/notion": "mounted",
+        "schedule/notion-ingress": "mounted",
+        "tool/notion_search": "mounted",
+        "tool/read_notion_comments": "mounted",
+        "tool/read_notion_page": "mounted"
+      }
+    },
+    "@open-everywhere/provider-slack": {
+      "version": "{{OE_VERSION}}",
+      "contributions": {
+        "channel/slack": "mounted",
+        "channel/slack-commands": "mounted",
+        "channel/slack-setup-page": "mounted"
+      }
+    }
+  }
+}

package/template/scripts/env-pipe.mjs

@@ -0,0 +1,74 @@
+// Write one value to stdout for a pipe, so it reaches `vercel env add` without being printed:
+//
+//   npm run -s env:pipe -- SLACK_BOT_TOKEN | npx vercel env add SLACK_BOT_TOKEN production
+//   npm run -s env:pipe -- OE_PUBLIC_BASE_URL=https://agent.example.com | npx vercel env add OE_PUBLIC_BASE_URL production
+//
+// `NAME` reads NAME from .env.local in the current directory. `NAME=value` writes that literal value.
+// The value goes out with no trailing newline, because a stored newline becomes part of the value.
+//
+// ALWAYS pipe it straight into `vercel env add`. Never run it bare. It writes only to a PIPE: a
+// terminal, a file or anything else is refused. That is not a guarantee that no person or agent sees
+// the value. An agent's shell and Git Bash's mintty terminal both connect stdout to a pipe, so a bare
+// run there passes this check and prints the value. The pipe into `vercel env add` is what keeps it
+// out of sight.
+//
+// It reads only .env.local, never the shell's own environment, so a variable that happens to be
+// exported in this shell is never sent by mistake. It parses the file with `node:util` parseEnv, the
+// parser eve uses for .env.local, so quotes, `export ` prefixes and ` # comment` tails are
+// stripped exactly as the running app strips them.
+
+import { fstatSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+import { parseEnv } from "node:util";
+
+const arg = process.argv[2];
+if (arg === undefined || process.argv.length > 3) {
+  console.error("usage: npm run -s env:pipe -- <NAME | NAME=value> | npx vercel env add <NAME> production");
+  process.exit(2);
+}
+// What counts as a pipe, per platform:
+//   • a FIFO — what a shell `|` gives on Linux and macOS (`isFIFO()`);
+//   • a socket — what Node's own `spawn(…, { stdio: "pipe" })` gives on Linux and macOS (a
+//     socketpair), so a parent process that reads the value directly is a pipe too;
+//   • on Windows, a named pipe. Node reports it with the POSIX S_IFIFO bits in `mode`, but
+//     `fs.constants.S_IFIFO` is undefined there, so `isFIFO()` answers false. Only on Windows
+//     are the mode bits read by hand.
+const S_IFMT = 0o170000;
+const S_IFIFO = 0o010000;
+let isPipe = false;
+try {
+  const st = fstatSync(1);
+  isPipe = st.isFIFO() || st.isSocket() || (process.platform === "win32" && (st.mode & S_IFMT) === S_IFIFO);
+} catch {
+  isPipe = false;
+}
+if (!isPipe) {
+  console.error("env:pipe writes a secret value, so it writes only to a pipe. Pipe it straight into `npx vercel env add <NAME> production`.");
+  process.exit(2);
+}
+
+const eq = arg.indexOf("=");
+const name = eq >= 0 ? arg.slice(0, eq) : arg;
+if (!/^[A-Z][A-Z0-9_]*$/.test(name)) {
+  console.error(`env:pipe: "${name}" is not an environment variable name`);
+  process.exit(2);
+}
+
+let value;
+if (eq >= 0) {
+  value = arg.slice(eq + 1);
+} else {
+  let text;
+  try {
+    text = readFileSync(join(process.cwd(), ".env.local"), "utf8");
+  } catch {
+    console.error("env:pipe: no .env.local in this directory");
+    process.exit(1);
+  }
+  value = parseEnv(text)[name];
+}
+if (value === undefined || value === "") {
+  console.error(`env:pipe: ${name} has no value${eq >= 0 ? "" : " in .env.local"}`);
+  process.exit(1);
+}
+process.stdout.write(value);

package/template/scripts/ops-mcp.ts

@@ -0,0 +1,35 @@
+// The ops / setup MCP server for this app, on stdio. `.mcp.json` registers it as
+// `open-everywhere-ops`, so a coding agent opened in this directory (Claude Code, for one) can
+// read setup status, write `.env.local` values and load secrets into the credential store.
+//
+// Core's server is provider-blind: each surface's app manifest and preflight checks register from
+// its provider bundle. Register the bundles this app's roster mounts, then serve.
+//
+// stdout IS the MCP transport. Diagnostics go to stderr.
+
+import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
+import { buildOpsServer } from "@open-everywhere/core/agent/ops/server.js";
+import { registerSlackOps } from "@open-everywhere/provider-slack/agent/ops.js";
+import { registerGithubOps } from "@open-everywhere/provider-github/agent/ops.js";
+import { registerSurfaceTrustPreflight } from "@open-everywhere/core/agent/identity/trust-preflight.js";
+import { importAppModule, resolveAppRoot } from "@open-everywhere/core/agent/scaffold/app-root.js";
+
+// The app root: `OE_APP_ROOT`, else the nearest directory above the working directory that holds
+// package.json and agent/agent.ts.
+const APP_ROOT = resolveAppRoot();
+// Populate the trust register BEFORE the trust preflight registers: an empty register would report
+// every surface with traffic as undeclared.
+await importAppModule("agent/identity/surfaces.ts", APP_ROOT);
+
+registerSlackOps();
+registerGithubOps();
+// Added a surface to roster.config.ts? Register its ops arm too:
+//   import { registerTeamsOps } from "@open-everywhere/provider-teams/agent/ops.js";
+//   import { registerDiscordOps } from "@open-everywhere/provider-discord/agent/ops.js";
+registerSurfaceTrustPreflight();
+
+const server = buildOpsServer();
+await server.connect(new StdioServerTransport());
+process.stderr.write(
+  "open-everywhere-ops MCP server ready on stdio (preflight · surface_manifest · scaffold_env · scaffold_credential_key · secret_set/list/rm)\n",
+);

package/template/tsconfig.json

@@ -0,0 +1,23 @@
+{
+  "compilerOptions": {
+    "target": "ES2023",
+    "module": "NodeNext",
+    "moduleResolution": "NodeNext",
+    "lib": [
+      "ES2023"
+    ],
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "noEmit": true,
+    "resolveJsonModule": true,
+    "types": [
+      "node"
+    ]
+  },
+  "include": [
+    "agent/**/*.ts",
+    "scripts/**/*.ts",
+    "roster.config.ts"
+  ]
+}