@stackbone/cli 0.1.0-alpha.9 → 0.2.0

package/CHANGELOG.md

@@ -7,7 +7,207 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
-## [0.1.0-alpha.9] - 2026-06-12
+## [0.2.0] - 2026-07-02
+
+### Added
+
+- **`stackbone add deep-agent <name>` (alias `add agent`) scaffolds a deep
+  agent.** A single `deep-agents/<name>/index.ts` default-exporting
+  `defineDeepAgent(...)`, added offline into the already-linked workspace.
+
+### Changed
+
+- **Scaffolding targets the deep-agents runtime.** `init` and `add` now generate
+  `deep-agents/<name>/index.ts` (deepagents/LangGraph on the standard
+  OpenAI/Anthropic chat wire) instead of the old `agent/` shell with eve channels
+  and hand-written HMAC auth. `stackbone dev` serves agents over the standard chat
+  endpoints with a playground, durable sessions (PostgresSaver checkpointer),
+  tool-level HITL and connector tools.
+
+### Removed
+
+- **The subtreed-starters system is gone.** `starter-loader` / `starter-renderer`
+  and the `starters-pull` script were removed; scaffolding now renders from the
+  bundled templates only.
+
+### Fixed
+
+- **`stackbone init` / `link` install skills on Windows.** `npx` is a `.cmd` shim
+  there, so spawning it without a shell failed with `spawn npx ENOENT`. The CLI now
+  spawns `npx` through the shell on win32 (the argv is fixed constants) and maps
+  cmd.exe's "is not recognized" to the same npx-not-found hint.
+
+## [0.1.0] - 2026-07-01
+
+### Added
+
+- **First stable release.** Drops the `-alpha` prerelease line: `npm i -g @stackbone/cli` now resolves the stable build through the `latest` dist-tag (previously pinned to the very first alpha).
+
+- **`stackbone` warns when a newer CLI is published, and validates the project's
+  SDK version.** On any command the CLI now runs a cached (24h TTL), non-blocking
+  npm registry check and prints a discreet stderr banner when a newer
+  `@stackbone/cli` exists on the channel matching the installed version (`alpha`
+  for prereleases, `latest` for stable) — silenced under `CI`, non-TTY, `--json`
+  and `STACKBONE_NO_UPDATE_CHECK`. It also reads the control plane's
+  `x-stackbone-min-cli-version` header to warn softly, and turns a `426 Upgrade
+Required` response into a clear `cli_upgrade_required` error. Generated projects
+  now pin `@stackbone/sdk` EXACTLY (no caret); `stackbone dev` warns and
+  `stackbone publish` blocks when the project's installed SDK is incompatible with
+  the CLI's bundled SDK line.
+
+## [0.1.0-alpha.15] - 2026-07-01
+
+### Changed
+
+- **`stackbone init` now links every workspace to the control plane, and `add`
+  no longer touches the network.** The workspace is the linked unit: `init`
+  registers the workspace's identity (an agent named after the workspace) and
+  writes the single `.stackbone/project.json` for EVERY `--with` kind — `empty`
+  and `workflow` included — that `dev`, `publish` and the management commands
+  read. Because of this, `init` now needs a signed-in session for every kind (the
+  fully-offline `init` path is gone — run `stackbone login` first). In exchange,
+  `stackbone add agent` and `add workflow-agent` are now 100% offline: the pieces
+  you add are members of the already-linked workspace, not separate control-plane
+  registrations, so they no longer register an agent or write `project.json`.
+  This fixes the confusing flow where `init` left a workspace unlinked and
+  `stackbone dev` then failed with "No project linked — run `stackbone init`".
+
+## [0.1.0-alpha.14] - 2026-06-25
+
+### Removed
+
+- **`stackbone trigger` and `stackbone queues` are gone.** Both belonged to the
+  deprecated classic `/invoke` runtime. `stackbone trigger` posted to the dev
+  emulator's `/api/invoke`, which the eve runtime retired (replaced by the
+  `/api/agents/:name/chat` proxy), so the command hit a route that no longer
+  exists. `stackbone queues` drove the classic agent-jobs BullMQ dispatcher,
+  which `stackbone dev` no longer boots (it answers `503 queues_unavailable`);
+  in the new runtime async work is modelled as durable workflow runs. To
+  exercise a trigger-driven workflow locally, start it by name with
+  `stackbone workflows start <name>`. The `STACKBONE_EMULATOR_URL` env var
+  (only `stackbone trigger` read it) is removed too. The `client.queues` SDK
+  capability and Studio's Queues inspector are unaffected.
+
+### Fixed
+
+- **`stackbone init` no longer scaffolds a workspace that fails `pnpm install`.**
+  The generated `package.json` pinned `@stackbone/workflow-world` — our durable
+  Workflow SDK World fork, a platform internal that is never published to npm —
+  so the post-init `pnpm install` 404'd ("could not resolve
+  @stackbone/workflow-world"). The CLI now ships its own esbuild-bundled copy of
+  the World and provisions it into the workspace's `node_modules` at `stackbone
+dev` boot (the same mechanism already used for `@stackbone/agent-server`), and
+  the scaffold no longer declares the package. The public `@workflow/world` is
+  kept as the single shared base.
+
+## [0.1.0-alpha.13] - 2026-06-25
+
+### Changed
+
+- **`stackbone dev`'s eve workspace boot now runs on the shared `@stackbone/workspace-runtime`
+  engine** (a pure refactor — no behaviour change). The supervisor that spawns the
+  N eve agents, the per-agent session Worlds, the main workspace World, and their
+  ordered teardown moved out of `apps/cli/src/dev/` into the new
+  `libs/runtime/workspace-runtime` lib, behind a single `bootWorkspaceRuntime`
+  entry point. The CLI is now just the dev SHELL around it (docker, tunnel,
+  watchers, local config, and the `eve build` / workflow compile it injects); the
+  same engine will boot cloud BYOC + self-host so the three environments run the
+  identical workspace. The engine assumes pre-compiled `.output/` and never
+  compiles — `stackbone dev` keeps its build-then-start behaviour by injecting
+  `runEveBuild`. `stackbone dev` over a v3 workspace boots identically (same
+  ports, health, workflows, chat).
+
+### Added
+
+- **`stackbone publish` now blocks an incomplete eve trace.** A new per-agent build
+  guardrail (alongside the native-deps + SDK-inline gates) detects when `eve build`
+  shipped a PARTIAL externalized `eve` into an agent's `.output/` — the symptom of an
+  upstream bug where eve's bundler (nitro + `@vercel/nft`) drops a file reachable only
+  through eve's `#`-subpath imports (e.g. `connections/errors.js`), which crash-loops
+  the agent at boot with `ERR_MODULE_NOT_FOUND`. Publish aborts with the fix: add
+  `"eve*"` to the agent's `defineAgent({ build: { externalDependencies } })`, which
+  forces nitro's full-trace of eve (and sidesteps eve's own self-filter). Agents that
+  fully inline eve are unaffected, so they never carry the full-trace's extra weight.
+- **`stackbone publish` now packages a v3 workspace.** When the current directory
+  has a `stackbone.config.ts` (`defineWorkspace`), publish compiles every eve
+  agent (`eve build` → its Nitro `.output/`) and the workflows on THIS host, then
+  packs `stackbone.config.ts` + each `.output/` + `.well-known/` + a derived
+  workspace manifest into a single tar with a verifiable SHA-256, written to
+  `dist/stackbone/`. Compiling at publish (not at boot) makes cold starts fast and
+  deterministic and turns a build error into a publish failure instead of a
+  production crash-loop. Each agent is scanned for native (`.node`) dependencies
+  and for an inlined `@stackbone/sdk` — either aborts with an actionable message
+  naming the offending package/agent before any tar is produced. (The
+  control-plane upload + build-pointer registration is the provisioning saga's
+  job; publish produces and records the artifact locally for now.)
+
+## [0.1.0-alpha.12] - 2026-06-17
+
+### Added
+
+- **Full agent-runtime command surface.** The CLI can now drive a running
+  agent end to end against the Studio endpoints, with one command group per
+  capability: `runs` and `config`, `queues`
+  (`list`/`jobs`/`retry`/`discard`/`peek`/`purge`/`push`), `logs`, `storage`,
+  `rag`, `secrets`, `hitl`, `openrouter`, `db`, `prompts`, `playground`,
+  `agents`, and `contract` (`show`/`schema`/`capabilities`/`validate`).
+- `stackbone organization use` switches the active organization for subsequent
+  commands.
+- `stackbone db migrate up`/`status` now prompt to create a `drizzle-kit`
+  rename when run on an interactive terminal, instead of silently guessing.
+- `stackbone playground` streams a run's steps live as they happen during an
+  invocation.
+
+### Changed
+
+- Studio commands now route to the **local emulator** for `kind=local` agents
+  (resolved from the dev statefile) instead of hitting the control plane, which
+  has no boot bundle for local installs.
+- Unified the cross-cutting conventions (flags, output, error handling)
+  across every agent-runtime surface so they behave consistently.
+- The HTTP adapter gained `put`/`delete` plus a Studio-target resolver, and
+  transfer helpers for SSE, multipart uploads and presigned downloads.
+
+### Removed
+
+- `stackbone db studio` — the embedded drizzle-kit studio command.
+
+### Fixed
+
+- `stackbone db migrate up`/`status` discover the dev Postgres URL instead of
+  failing to connect.
+- `queues peek`/`purge` are now scoped to a single handler (logical queue)
+  rather than the whole physical queue.
+
+## [0.1.0-alpha.11] - 2026-06-12
+
+### Fixed
+
+- `stackbone dev` (and the self-host runtime) no longer crash on boot with
+  `BUNDLE_IMPORT_FAILED` when an agent splits its code across files and uses
+  idiomatic extensionless relative imports (e.g. `import { x } from './schema'`).
+  The dev/self-host runner imports the creator's raw TypeScript through Node's
+  native type-stripping, whose ESM resolver never adds a file extension — so a
+  multi-file agent that ran fine once published would crash locally. The
+  generated dev entry now registers a module-resolution hook that mirrors
+  esbuild's TypeScript resolution (the same `stackbone publish` relies on): it
+  adds the missing `.ts`/`.tsx`/`.mts`/`.cts` extension (and the directory
+  `index` form), and applies the NodeNext `./x.js` → `./x.ts` rewrite. Dev and
+  self-host now resolve modules exactly like a published agent.
+
+## [0.1.0-alpha.10] - 2026-06-12
+
+### Fixed
+
+- `stackbone dev` no longer crashes on boot with
+  `ERR_MODULE_NOT_FOUND: @stackbone/agent-server` in a real creator project
+  outside this monorepo. The dev agent runs as a separate `node` process that
+  imports `@stackbone/agent-server` — a private package a creator never installs
+  — and the published CLI had no way to provide it (it could only bundle it from
+  workspace source, which isn't present in an installed CLI). The CLI build now
+  ships its own pre-bundled copy under `runtime-deps/`, and `stackbone dev`
+  copies it into the project's `node_modules` at boot whenever no workspace
+  source is found. In-monorepo development is unchanged.
 
 ### Added
 

package/README.md

@@ -9,30 +9,33 @@ The CLI for [Stackbone](https://stackbone.ai) — the marketplace and runtime fo
 No install required — invoke the CLI through `npx`:
 
 ```bash
-npx @stackbone/cli@alpha init my-agent
-cd my-agent
-npx @stackbone/cli@alpha dev
+npx @stackbone/cli login
+npx @stackbone/cli init my-workspace
+cd my-workspace && npm install
+npx @stackbone/cli dev
 ```
 
-That's it. `init` scaffolds a new agent project (templates + Claude skill); `dev` runs a local control-plane emulator with Stackbone Studio at `http://localhost:4242` so you can iterate without publishing.
+That's it. `init` scaffolds a new workspace shell (agents + workflows + coding-agent skills) and links it to your organization, so you sign in first; `dev` boots the whole workspace locally — Postgres, Redis, MinIO and Stackbone Studio at `http://localhost:4242` — so you can iterate without publishing.
 
 ## Commands
 
-| Command                      | What it does                                                           |
-| ---------------------------- | ---------------------------------------------------------------------- |
-| `stackbone init [name]`      | Scaffold a new Stackbone project from a template.                      |
-| `stackbone dev`              | Run the local control-plane emulator + Stackbone Studio.               |
-| `stackbone publish`          | Build, push, and publish the current agent to Stackbone runtime.       |
-| `stackbone login` / `logout` | Authenticate the CLI via the device-code flow (RFC 8628).              |
-| `stackbone whoami`           | Show the active user and organization.                                 |
-| `stackbone current`          | Print the organization currently linked to this session.               |
-| `stackbone link`             | Link the current directory to an existing organization + agent.        |
-| `stackbone list`             | List organizations the current user owns.                              |
-| `stackbone metadata`         | Agent-friendly snapshot of the workspace state.                        |
-| `stackbone db <subcmd>`      | Manage the local emulator database (migrations, queries).              |
-| `stackbone docs <topic>`     | Print Stackbone documentation inline. Run with no args to list topics. |
-
-Run any command with `--help` for the full flag reference, or `npx @stackbone/cli@alpha docs cli` for the inline documentation.
+| Command                       | What it does                                                           |
+| ----------------------------- | ---------------------------------------------------------------------- |
+| `stackbone init [name]`       | Scaffold a new workspace shell and link it to your org (needs login).  |
+| `stackbone add <kind> <name>` | Add an agent / workflow / workflow-agent to the workspace (offline).   |
+| `stackbone dev`               | Boot the whole workspace locally + Stackbone Studio.                   |
+| `stackbone workflows <verb>`  | List, inspect the schema of, or start the install's workflows.         |
+| `stackbone publish`           | Compile every agent + workflow and pack the workspace bundle.          |
+| `stackbone login` / `logout`  | Authenticate the CLI via the device-code flow (RFC 8628).              |
+| `stackbone whoami`            | Show the active user and organization.                                 |
+| `stackbone current`           | Print the organization currently linked to this session.               |
+| `stackbone link`              | Link the current directory to an existing organization + agent.        |
+| `stackbone list`              | List organizations the current user owns.                              |
+| `stackbone metadata`          | Agent-friendly snapshot of the workspace state.                        |
+| `stackbone db <subcmd>`       | Manage the local emulator database (migrations, queries).              |
+| `stackbone docs <topic>`      | Print Stackbone documentation inline. Run with no args to list topics. |
+
+Run any command with `--help` for the full flag reference, or `npx @stackbone/cli docs cli` for the inline documentation.
 
 ## Requirements