@critical-path/svelte 0.18.20 → 0.19.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +22 -0
- package/dist/index.js +1112 -635
- package/dist/webmcp-state.svelte.d.ts +8 -0
- package/dist/webmcp-state.svelte.d.ts.map +1 -1
- package/package.json +8 -8
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,27 @@
|
|
|
1
1
|
# @critical-path/svelte
|
|
2
2
|
|
|
3
|
+
## 0.19.0
|
|
4
|
+
|
|
5
|
+
### Minor Changes
|
|
6
|
+
|
|
7
|
+
- e2f997c: Help agents find and use the right tools.
|
|
8
|
+
|
|
9
|
+
- **Toolsets:** tools are grouped into `core`, `collaboration`, `scheduling`, `analytics` and `guidance`. `createCriticalPathMcpServer`, `registerWebMcpTools`, `useWebMCP` and `createWebMcpState` accept `toolsets` and `readOnly` alongside `tools`; the CLI takes `--toolsets`, `--tools` and `--read-only`. `selectTools()` and `TOOLSETS` are exported.
|
|
10
|
+
- **Server instructions:** the MCP server sends `instructions` on connect (data model, id flow, which tool fits which request, skill index), built from the exposed tools by `buildAgentInstructions()`. WebMCP returns them as `handle.instructions`; `useWebMCP` and `WebMcpState` expose `instructions`.
|
|
11
|
+
- **Agent skills:** six bundled playbooks (`plan-project`, `triage-tasks`, `analyze-schedule`, `report-status`, `plan-capacity`, `investigate-task`) in the Agent Skills `SKILL.md` format. Agents load them with the new `get_skill` tool or the `criticalpath://skills/{name}` resources; `npx @critical-path/mcp skills install` writes them to `.claude/skills`. `SKILLS` and `getSkill()` are exported and the files ship in the package.
|
|
12
|
+
- **Descriptions** for every tool say when to use it, what it returns and which sibling to use instead; every argument is described. Unit tests lint descriptions and skills, and `pnpm --filter @critical-path/mcp eval:tools` measures tool choice with Claude.
|
|
13
|
+
- **Structured results:** every tool declares an `outputSchema` and returns `structuredContent`. Errors carry recovery hints (for example, find valid ids with `list_tasks`).
|
|
14
|
+
- **Annotations:** tools send `readOnlyHint`, `destructiveHint`, `idempotentHint` and `openWorldHint`.
|
|
15
|
+
- `useWebMCP` no longer re-registers its tools on every render when `tools` is passed as an inline array.
|
|
16
|
+
|
|
17
|
+
**BREAKING:** `list_projects`, `list_deliverables` and `list_comments` return `{ projects }`, `{ deliverables }` and `{ comments }` instead of arrays. Lookups that find nothing throw instead of returning `null`. Unknown names in `tools` throw. `ToolDefinition.annotations` uses the standard MCP names (`requiresConfirmation` is now `destructiveHint`), and `defineTool` requires `toolset` and `outputZodSchema`. See UPGRADING.md.
|
|
18
|
+
|
|
19
|
+
### Patch Changes
|
|
20
|
+
|
|
21
|
+
- Updated dependencies [e2f997c]
|
|
22
|
+
- Updated dependencies [202a61d]
|
|
23
|
+
- @critical-path/mcp@0.19.0
|
|
24
|
+
|
|
3
25
|
## 0.18.20
|
|
4
26
|
|
|
5
27
|
### Patch Changes
|