Version Packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@dawn-ai/langchain@1.0.0

Major Changes

• ad17e85: Upgrade @langchain/core (0.3 → 1.x), @langchain/langgraph (0.2 → 1.x), @langchain/openai (0.3 → 1.x), and zod (3 → 4). Removes the dual-zod-version cast workaround in tool-converter.ts; DynamicStructuredTool now accepts Standard Schema directly. Downstream consumers must align on the new peer ranges (@langchain/core >=1.1.0).

Minor Changes

• 34e615b: Add the first phase-3 harness capability: planning. A plan.md file in a route directory now opts the agent into a built-in write_todos tool, a todos state channel, a Dawn-locked planning prompt fragment, and a plan_update SSE event. Introduces CapabilityMarker and applyCapabilities in @dawn-ai/core — the autowiring spine that all later phase-3 capabilities (skills, subagents, etc.) will reuse.

• affeb46: Capability tools can now mutate state channels via a Dawn-native {result, state} wrapped return shape — result becomes the agent-visible ToolMessage; state is a partial channel update applied via reducers. The langchain bridge translates this into a LangGraph Command({update}) internally; capability authors don't import from @langchain/langgraph. Plain tool returns (anything not matching the strict wrapper shape) work unchanged.

  Planning's write_todos adopts the new shape, fixing the previously-documented re-emission loop: the todos state channel now actually reflects the agent's writes between turns, so the agent stops re-calling write_todos with the same content. The plan_update stream transformer also reads defensively from both legacy and Command-shaped tool outputs so the SSE event keeps firing.

• 1005b3a: Add provider-aware agent materialization. Agent configs can now carry an optional provider, and the LangChain runtime infers providers for known model families or lazy-loads the explicit provider integration package for built-in provider IDs.

• e8462db: agent({...}) now accepts an optional reasoning: { effort } field. Maps to OpenAI's reasoningEffort parameter (none | minimal | low | medium | high | xhigh). Non-reasoning models silently ignore it. Useful for tool-use-heavy agents that aren't following directives at the default reasoning depth.

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.
• Updated dependencies [82dd52f]
• Updated dependencies [dd242ac]
• Updated dependencies [34e615b]
• Updated dependencies [2ba0773]
• Updated dependencies [affeb46]
• Updated dependencies [12ee95f]
• Updated dependencies [1005b3a]
• Updated dependencies [e8462db]
  • @dawn-ai/core@1.0.0
  • @dawn-ai/sdk@1.0.0

@dawn-ai/cli@1.0.0

Minor Changes

• ad17e85: Upgrade @langchain/core (0.3 → 1.x), @langchain/langgraph (0.2 → 1.x), @langchain/openai (0.3 → 1.x), and zod (3 → 4). Removes the dual-zod-version cast workaround in tool-converter.ts; DynamicStructuredTool now accepts Standard Schema directly. Downstream consumers must align on the new peer ranges (@langchain/core >=1.1.0).

• dd242ac: Add the agents-md built-in capability: Dawn now auto-injects <workspace>/AGENTS.md into every agent's system prompt under a # Memory heading on every model turn. Always-on (no opt-in marker). Preserves the feedback loop — the agent updates its memory via writeFile and the next turn sees the change automatically. Re-reads the file each turn (64 KiB cap; oversize, empty, or unreadable files render empty or a one-line notice).

• 34e615b: Add the first phase-3 harness capability: planning. A plan.md file in a route directory now opts the agent into a built-in write_todos tool, a todos state channel, a Dawn-locked planning prompt fragment, and a plan_update SSE event. Introduces CapabilityMarker and applyCapabilities in @dawn-ai/core — the autowiring spine that all later phase-3 capabilities (skills, subagents, etc.) will reuse.

• 2ba0773: Add the phase-3 skills capability. A route with src/app/<route>/skills/<name>/SKILL.md files now exposes them to the agent via:

  • An always-on # Skills section in the system prompt listing each skill's name + description
  • A readSkill({ name }) tool the agent calls to load a skill's full body on demand

  Each SKILL.md requires YAML frontmatter with description; name defaults to the directory name and can be overridden. The body lives in conversation history after readSkill returns it (not re-injected each turn) — matches the deepagents / Claude Code convention. Typegen includes readSkill in RouteTools when a route has skills. The chat example ships two seeded skills (workspace-conventions, recover-from-failure).

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.
• 13bc466: Fix SSE event payload double-wrap. toSseEvent used to emit data: {"data": <value>} for the built-in chunk event and for capability-contributed events like plan_update, when it should emit data: <value> directly. The shaped events (tool_call, tool_result, done) are unchanged.
• 36552c1: docs: rebrand "LangGraph Platform" → "LangSmith" in user-visible CLI strings, README, and comments. The langgraph.json artifact format is unchanged.
• Updated dependencies [82dd52f]
• Updated dependencies [ad17e85]
• Updated dependencies [dd242ac]
• Updated dependencies [34e615b]
• Updated dependencies [2ba0773]
• Updated dependencies [affeb46]
• Updated dependencies [12ee95f]
• Updated dependencies [1005b3a]
• Updated dependencies [e8462db]
  • @dawn-ai/core@1.0.0
  • @dawn-ai/langchain@1.0.0
  • @dawn-ai/langgraph@1.0.0
  • @dawn-ai/permissions@0.1.8

@dawn-ai/core@1.0.0

Minor Changes

• dd242ac: Add the agents-md built-in capability: Dawn now auto-injects <workspace>/AGENTS.md into every agent's system prompt under a # Memory heading on every model turn. Always-on (no opt-in marker). Preserves the feedback loop — the agent updates its memory via writeFile and the next turn sees the change automatically. Re-reads the file each turn (64 KiB cap; oversize, empty, or unreadable files render empty or a one-line notice).

• 34e615b: Add the first phase-3 harness capability: planning. A plan.md file in a route directory now opts the agent into a built-in write_todos tool, a todos state channel, a Dawn-locked planning prompt fragment, and a plan_update SSE event. Introduces CapabilityMarker and applyCapabilities in @dawn-ai/core — the autowiring spine that all later phase-3 capabilities (skills, subagents, etc.) will reuse.

• 2ba0773: Add the phase-3 skills capability. A route with src/app/<route>/skills/<name>/SKILL.md files now exposes them to the agent via:

  • An always-on # Skills section in the system prompt listing each skill's name + description
  • A readSkill({ name }) tool the agent calls to load a skill's full body on demand

  Each SKILL.md requires YAML frontmatter with description; name defaults to the directory name and can be overridden. The body lives in conversation history after readSkill returns it (not re-injected each turn) — matches the deepagents / Claude Code convention. Typegen includes readSkill in RouteTools when a route has skills. The chat example ships two seeded skills (workspace-conventions, recover-from-failure).

• affeb46: Capability tools can now mutate state channels via a Dawn-native {result, state} wrapped return shape — result becomes the agent-visible ToolMessage; state is a partial channel update applied via reducers. The langchain bridge translates this into a LangGraph Command({update}) internally; capability authors don't import from @langchain/langgraph. Plain tool returns (anything not matching the strict wrapper shape) work unchanged.

  Planning's write_todos adopts the new shape, fixing the previously-documented re-emission loop: the todos state channel now actually reflects the agent's writes between turns, so the agent stops re-calling write_todos with the same content. The plan_update stream transformer also reads defensively from both legacy and Command-shaped tool outputs so the SSE event keeps firing.

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.

• 12ee95f: Two fixes surfaced by live LLM smoke testing the chat example end-to-end:

  • Planning write_todos now declares a real zod schema. Previously the tool's schema field was undefined; the LangChain bridge fell back to z.record(z.string(), z.unknown()), which produced JSON Schema without properties. OpenAI strict-mode tool calling rejected the tool with 400 Invalid schema for function 'write_todos': object schema missing properties. Now the planning marker exports an explicit zod schema for { todos: Array<{ content, status }> }. Adds zod as a runtime dependency of @dawn-ai/core.
  • # Memory block now includes orientation text. The agents-md prompt fragment used to inject # Memory\n\n<content> only. With both planning and memory loaded, the model often called listDir and readFile to look at AGENTS.md even though Dawn had already injected its contents. The fragment now opens with a short paragraph telling the agent the block IS the memory file, re-rendered each turn, and that the way to update it is writeFile. Existing unit tests still pass — the # Memory heading and content substrings are unchanged.

• Updated dependencies [82dd52f]

• Updated dependencies [1005b3a]

• Updated dependencies [e8462db]

  • @dawn-ai/sdk@1.0.0
  • @dawn-ai/permissions@0.1.8
  • @dawn-ai/workspace@0.1.8

@dawn-ai/sdk@1.0.0

Minor Changes

• 1005b3a: Add provider-aware agent materialization. Agent configs can now carry an optional provider, and the LangChain runtime infers providers for known model families or lazy-loads the explicit provider integration package for built-in provider IDs.
• e8462db: agent({...}) now accepts an optional reasoning: { effort } field. Maps to OpenAI's reasoningEffort parameter (none | minimal | low | medium | high | xhigh). Non-reasoning models silently ignore it. Useful for tool-use-heavy agents that aren't following directives at the default reasoning depth.

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.

@dawn-ai/vite-plugin@1.0.0

Minor Changes

• ad17e85: Upgrade @langchain/core (0.3 → 1.x), @langchain/langgraph (0.2 → 1.x), @langchain/openai (0.3 → 1.x), and zod (3 → 4). Removes the dual-zod-version cast workaround in tool-converter.ts; DynamicStructuredTool now accepts Standard Schema directly. Downstream consumers must align on the new peer ranges (@langchain/core >=1.1.0).

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.
• Updated dependencies [82dd52f]
• Updated dependencies [dd242ac]
• Updated dependencies [34e615b]
• Updated dependencies [2ba0773]
• Updated dependencies [affeb46]
• Updated dependencies [12ee95f]
  • @dawn-ai/core@1.0.0

@dawn-ai/config-biome@1.0.0

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.

@dawn-ai/config-typescript@1.0.0

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.

create-dawn-ai-app@1.0.0

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.
• Updated dependencies [82dd52f]
  • @dawn-ai/devkit@1.0.0

@dawn-ai/devkit@1.0.0

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.

@dawn-ai/langgraph@1.0.0

Patch Changes

• 82dd52f: Correct package README links and CLI/runtime examples, export the SDK reasoning type, and fix dawn build agent deployment entry generation.
• Updated dependencies [82dd52f]
• Updated dependencies [1005b3a]
• Updated dependencies [e8462db]
  • @dawn-ai/sdk@1.0.0

@dawn-example/chat-server@0.0.1

Patch Changes

• Updated dependencies [82dd52f]
• Updated dependencies [13bc466]
• Updated dependencies [ad17e85]
• Updated dependencies [dd242ac]
• Updated dependencies [34e615b]
• Updated dependencies [2ba0773]
• Updated dependencies [affeb46]
• Updated dependencies [12ee95f]
• Updated dependencies [1005b3a]
• Updated dependencies [e8462db]
• Updated dependencies [36552c1]
  • @dawn-ai/cli@1.0.0
  • @dawn-ai/core@1.0.0
  • @dawn-ai/langchain@1.0.0
  • @dawn-ai/sdk@1.0.0

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
dawnai	Ready	Preview, Comment	May 21, 2026 9:38pm