@wfgraph/plugins@3.1.1

chore: version packages (#170)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

@wfgraph/plugins@3.1.0

chore: version packages (#152)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

@wfgraph/core@3.1.1

chore: version packages (#170)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

@wfgraph/core@3.1.0

Minor Changes

• #154 6a0cd6e Thanks @alandotcom! - Add durable workflow version history, semantic publication review, and restore as draft.

Patch Changes

• #154 6a0cd6e Thanks @alandotcom! - Prevent stale workflow viewport fitting from hiding the current canvas and reduce comparison and version lookup work.

• #147 a72fa25 Thanks @alandotcom! - Validate non-Effect step and action outputs through ~standard.validate, so undeclared keys no longer pass through when the library strips them and a mismatched answer fails the node once.

@wfgraph/client@3.1.1

Patch Changes

• #156 db14e3f Thanks @alandotcom! - Keep workflow nodes visible while switching between Draft, Runs, and Changes,
  and keep the viewport stable while dragging Lifecycle.

@wfgraph/client@3.1.0

Minor Changes

• #154 6a0cd6e Thanks @alandotcom! - Move Draft, Runs, and Changes into a persistent workflow workspace switcher, with stable loading transitions and matching canvas, inspector, and status-strip behavior.

• #154 6a0cd6e Thanks @alandotcom! - Add durable workflow version history, semantic publication review, and restore as draft.

Patch Changes

• #154 6a0cd6e Thanks @alandotcom! - Make workflow menus hand off on hover, add workflow actions to the command palette, and let node context menus enable or disable eligible steps.

• #154 6a0cd6e Thanks @alandotcom! - Prevent stale workflow viewport fitting from hiding the current canvas and reduce comparison and version lookup work.

• #153 8c52eeb Thanks @alandotcom! - Redesign the editor Runs panel as an Execution Inspector with event-led run identities, outcome summaries, a canvas-synchronized node journey, friendly action results, and a docked raw-payload console.

@wfgraph/plugins@3.0.0

chore: version packages (#144)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

@wfgraph/core@3.0.0

Major Changes

• #145 708ac73 Thanks @alandotcom! - Replace the persisted Group exitNodeId field with exitNodeIds, and support lookup groups whose exits share one downstream endpoint.

Minor Changes

• #128 c2ae17b Thanks @alandotcom! - Add a build agent to the workflow editor: a chat panel that reads the extension
  catalog and the open workflow, then edits the canvas as it answers.

  A host turns it on with a new agent option on createWfGraphApp, carrying an
  OpenAI API key and optionally a model id and an OpenAI-compatible endpoint.
  Absent or blank, the agent is off: no model is called and the editor shows no
  panel, so an adopter who wants no AI in their editor writes nothing.

  The agent has fourteen tools. Seven read (search the action catalog, describe one
  action's config and output fields, list Events and integrations, read the graph,
  list the template tokens a step may reference, and validate the workflow) and
  seven write (add, update and delete a step, connect and disconnect steps, declare
  the Lifecycle Rules, and write a Condition's test). They run on the server against
  the graph the editor sends with each turn, and the resulting graph streams back
  so the canvas redraws as the agent works. The edit lands through the editor's own
  save path, so it is one undo step and the autosave persists it.

  GET /api/extensions now answers agent: { enabled } beside catalog, which is
  how the editor knows whether to offer the panel. agent.chat is the first
  streaming RPC procedure in the contract, declared as an oRPC event iterator.

  The effect catalog moves to 4.0.0-rc.110, which the @effect/ai-openai
  provider names as its peer.

Patch Changes

• #151 1da1baf Thanks @alandotcom! - Improve build-agent authoring for Wait steps, Event Split outlets, condition reference fields, and draft publication blockers.

• #128 c2ae17b Thanks @alandotcom! - Prevent the build agent from applying workflow edits that violate graph topology rules.

• #148 01da8e0 Thanks @alandotcom! - Stop sending start and result payloads on the run-list procedures.

  getExecutions polls every two seconds while the Runs tab is open, and
  getExecutionsGlobal pages the dashboard. Neither list paints input or
  output, yet both selected those JSONB columns and redacted them on every
  answer. Payloads stay on getExecutionLogs, which is fetched for the one open
  run.

@wfgraph/client@3.0.0

Minor Changes

• #128 c2ae17b Thanks @alandotcom! - Add a build agent to the workflow editor: a chat panel that reads the extension
  catalog and the open workflow, then edits the canvas as it answers.

  A host turns it on with a new agent option on createWfGraphApp, carrying an
  OpenAI API key and optionally a model id and an OpenAI-compatible endpoint.
  Absent or blank, the agent is off: no model is called and the editor shows no
  panel, so an adopter who wants no AI in their editor writes nothing.

  The agent has fourteen tools. Seven read (search the action catalog, describe one
  action's config and output fields, list Events and integrations, read the graph,
  list the template tokens a step may reference, and validate the workflow) and
  seven write (add, update and delete a step, connect and disconnect steps, declare
  the Lifecycle Rules, and write a Condition's test). They run on the server against
  the graph the editor sends with each turn, and the resulting graph streams back
  so the canvas redraws as the agent works. The edit lands through the editor's own
  save path, so it is one undo step and the autosave persists it.

  GET /api/extensions now answers agent: { enabled } beside catalog, which is
  how the editor knows whether to offer the panel. agent.chat is the first
  streaming RPC procedure in the contract, declared as an oRPC event iterator.

  The effect catalog moves to 4.0.0-rc.110, which the @effect/ai-openai
  provider names as its peer.

Patch Changes

• #146 04f4fd1 Thanks @alandotcom! - Improve canvas selection, connection feedback, issue counts, keyboard shortcut discovery, and graph accessibility.

• #150 60caf9a Thanks @alandotcom! - Restore a run's pinned graph and running animation after navigating back to it. A same-workflow hydrate was clearing the overlay, so the canvas stayed on the draft while the Runs panel still showed the open run.

• #148 01da8e0 Thanks @alandotcom! - Stop sending start and result payloads on the run-list procedures.

  getExecutions polls every two seconds while the Runs tab is open, and
  getExecutionsGlobal pages the dashboard. Neither list paints input or
  output, yet both selected those JSONB columns and redacted them on every
  answer. Payloads stay on getExecutionLogs, which is fetched for the one open
  run.

• #149 384e8ca Thanks @alandotcom! - Replace the dashboard run-history status chips with a searchable, filterable, virtualized table.

@wfgraph/plugins@2.5.0

chore: version packages (#143)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>