@stackbone/cli 0.3.3 → 0.3.4

package/CHANGELOG.md

@@ -7,6 +7,99 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.3.4] - 2026-08-19
+
+### Added
+
+- **`stackbone init` installs the workspace dependencies itself.** After the
+  files are written it runs `pnpm install` when `pnpm` answers on `PATH` and
+  `npm install` otherwise, so the workspace is ready for `stackbone dev` the
+  moment the command returns. The scaffold works under both: its `.npmrc` pins
+  a hoisted `node_modules`, its root `package.json` now declares the
+  `workflows` subpackage in a `workspaces` field as well as in
+  `pnpm-workspace.yaml` (npm does not read the yaml), and neither `build` nor
+  `package` install from the workspace. Best-effort like the coding-agent setup: a failed
+  install prints the package manager's last lines as a warning, the command still
+  exits `0`, and the "Next steps" note keeps the `<pm> install` line — naming
+  the manager that ran, not a hard-coded `pnpm`. `--no-install` skips the step.
+  The JSON envelope gains `dependencies: { status, package_manager, error? }`
+  (`installed` | `failed` | `skipped`).
+
+### Changed
+
+- **`stackbone init` and `stackbone link` ask which coding agents you use, and
+  that one answer installs the skills AND configures the documentation MCP
+  server.** The CLI used to guess: it called `npx skills` with no agent list and
+  let upstream detection decide where to copy. That detection is
+  `existsSync(~/.<agent>)`, so a leftover directory from an old `skills add -g`
+  made every later `init` copy into tools the creator had never opened, and when
+  it detected nothing at all it installed into every known agent. There is now a
+  multiselect of the seven supported agents (Claude Code, Cursor, GitHub
+  Copilot, Gemini CLI, Windsurf, Cline, Codex) with the ones present on disk
+  pre-ticked, and whatever you tick gets both halves: the four agent skills, and
+  a `stackbone-docs` entry in that agent's own MCP configuration. Zero is a
+  valid answer.
+
+  The seven MCP entries do **not** share a shape, and that is registry data
+  rather than branches in code: VS Code keys its map `servers`, not
+  `mcpServers`; Gemini CLI reads `httpUrl` and treats `url` as SSE; Windsurf
+  reads `serverUrl`; Cline spells the transport `streamableHttp` in that exact
+  casing (with `streamable-http` it falls back to SSE and the endpoint answers
+  405); and Codex is TOML. Every wrong shape parses without complaining and then
+  fails to connect, which is the worst failure we could ship because the creator
+  sees a config file that looks right. Every write MERGES: the other MCP servers
+  in the file survive, so do foreign root keys (`inputs` in `.vscode/mcp.json`),
+  and an existing `stackbone-docs` entry is left byte for byte as it is, because
+  the creator may have edited it. A file we cannot parse is left untouched and
+  that one target is marked `failed`; the rest are still written.
+
+  The four project-scoped MCP files (`.mcp.json`, `.cursor/mcp.json`,
+  `.vscode/mcp.json`, `.gemini/settings.json`) are deliberately NOT added to the
+  `.gitignore` block: they are meant to be committed, so a teammate who clones
+  the repository inherits the documentation server without running anything.
+
+  **Breaking.** In non-interactive mode (`--json`, `--yes`, no TTY, or `CI` set
+  in the environment) `init` and `link` now install **nothing** unless you pass
+  `--agents <list>`. `stackbone init --json` used to install the skills for
+  whichever agents were auto-detected; it now installs none. A multiselect fired
+  under `--json` would hang a pipeline forever with nothing on stdout to explain
+  it, so the gate is hard. Any script or CI job that relied on `init --json`
+  leaving the skills in place has to add `--agents claude-code,cursor,…`.
+  `--no-agents` is the explicit switch for setting nothing up.
+
+  The JSON envelope of both commands changes with it: `skills_install` is gone
+  and `agent_setup` takes its place, shaped
+  `{ ok, selected, steps: { skills, mcp } }` where `mcp` carries a per-agent
+  status (`written` | `already` | `failed` | `skipped`). The shape stays constant
+  even when nothing is selected, so an agent reading the output never has to
+  branch on a missing block.
+
+- **`stackbone docs` hands over MCP connection details instead of printing the
+  documentation.** The CLI used to compile a copy of the wiki into the binary
+  and render a page inline (`stackbone docs cli/commands`). That copy was 544 KB
+  of a 2.8 MB bundle, it had no search so the reader had to guess an exact topic
+  key, and an old install could hand out documentation that quietly disagreed
+  with the live site. The wiki now serves the same pages over MCP at
+  `https://docs.stackbone.ai/mcp`, which searches by meaning and always reflects
+  the current deploy, so the command's job is the connection details: the
+  endpoint, the `claude mcp add` one-liner, and the `mcpServers` block every
+  other client accepts. `--json` returns them structured, with
+  `install.mcpServers` ready to write into a client config without parsing the
+  human output.
+
+  **Breaking.** Passing a topic now exits `4` and names the replacement rather
+  than rendering a page. Rejecting the argument beats ignoring it: a script that
+  still passes one finds out, instead of silently reading a help screen where it
+  expected a reference page. The binary lost ~518 KB.
+
+### Removed
+
+- **The wiki-to-CLI documentation sync.** `apps/cli/src/docs/` (including the
+  generated `topics.generated.ts`), the `sync-docs` and `sync-docs:check` Nx
+  targets, the `cli:sync-docs` / `cli:sync-docs:check` workspace scripts, and the
+  `sync-docs` dependency of `cli:build`. Nothing regenerates into the binary any
+  more, so nothing can drift against it.
+
 ## [0.3.3] - 2026-08-11
 
 ### Added

package/README.md

@@ -19,23 +19,23 @@ That's it. `init` scaffolds a new workspace shell (agents + workflows + coding-a
 
 ## Commands
 
-| 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 build`             | Compile the workspace into the bundle your container image bakes in.   |
-| `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.
+| 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 build`             | Compile the workspace into the bundle your container image bakes in.  |
+| `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`              | Show how to reach the documentation over MCP.                         |
+
+Run any command with `--help` for the full flag reference. The full documentation lives at <https://docs.stackbone.ai>, and is also served to coding agents over MCP at <https://docs.stackbone.ai/mcp> — run `stackbone docs` for the connection details.
 
 ## Requirements
 
@@ -46,7 +46,7 @@ Run any command with `--help` for the full flag reference, or `npx @stackbone/cl
 
 - Documentation: <https://docs.stackbone.ai>
 - Marketplace: <https://stackbone.ai>
-- Issues & feedback: report via `stackbone docs feedback` once authenticated, or contact the team directly.
+- Issues & feedback: contact the team directly.
 
 ## License