Conversation
…ol call serialization - Add execute-by-id endpoint to workflow-orchestrator for service-to-service workflow execution - Add run_workflow, get_workflow_execution_status, approve_workflow MCP tools to mastra-agent-tanstack - Add RunWorkflow UI component with form, status polling, and approval gate UI - Fix AI SDK tool call serialization (getter properties don't survive JSON.stringify) - Fix MCP client pool session management to prevent OOM from session churn - Add LogsViewer component, k8s sandbox config, and Dapr pub/sub event publishing - Update function-router registry and credential handling for mastra/execute action type Co-Authored-By: Claude Opus 4.6 <[email protected]>
Mastra wraps AI SDK tool calls in an envelope:
{ type: "tool-call", payload: { toolCallId, toolName, args } }
The previous fix tried accessing tc.toolName directly, which was undefined.
Now unwraps from tc.payload first. Also removes otel deps (from other process)
that caused CJS/ESM import crash on startup.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
…legacy services Remove planner-dapr-agent service and all planner/* action types. All agent actions now route through mastra-agent-tanstack exclusively. Fix orchestrator routing: only `mastra/execute` (async, fire-and-forget) goes through `process_agent_child_workflow`. Sync actions like `mastra/clone` and `mastra/plan` now correctly route through `execute_action` → function-router → mastra-agent-tanstack's proper endpoints. Fix workflow-mcp-server's execute_workflow tool to use the orchestrator's `execute-by-id` endpoint which properly serializes React Flow nodes, fixing approval gate timeouts (was 60s default instead of configured 300s) and missing node labels. Also removes workflow-orchestrator-ts-archived, legacy dapr workflow API routes, planner plugin, and workflow-executor. Updates CLAUDE.md to reflect the current architecture. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Replace mastra-agent-tanstack's in-memory agent loop with a durable Dapr Workflow implementation that survives restarts, has built-in retries, and deterministic replay. Deployed alongside (Phase 1) so both agent/* and durable/* action types work. - Copy @dapr-agents/durable-agent library (workflow activities, state management, orchestration, pub/sub, registry, observability) - Port K8s sandbox, remote filesystem, and sandbox config from mastra-agent-tanstack (standalone, no @mastra/core dependency) - Create workspace tools as DurableAgentTool objects (read/write/edit files, execute commands, clone repos) - Add Express server with same API surface (run, plan, execute-plan, tools, health, dapr subscriptions) - Integrate with orchestrator (durable/* action routing) and function-router (durable-agent service resolution) - Register durable-agent plugin with durable/run action type Co-Authored-By: Claude Opus 4.6 <[email protected]>
- Upgrade ai to ^6.0.0 and @ai-sdk/openai to ^3.0.0
- Use LanguageModel type (replaces LanguageModelV1)
- Use openai.chat() to force Chat Completions API (not Responses API)
- Fix tool-call content parts: use `input` field (not `args`)
- Fix tool-result output: use discriminated union { type: "json", value }
- Build zodToJsonSchema converter for tool input schemas
- Replace pub/sub completion with direct Dapr service invocation to
orchestrator raise_event API (bypasses component scoping mismatch)
- Use DaprWorkflowClient.waitForWorkflowCompletion() for reliable polling
- Handle WorkflowRuntimeStatus as numeric enum (COMPLETED=1)
Co-Authored-By: Claude Opus 4.6 <[email protected]>
…retention The Runs tab lost per-node output data when navigating away because the status polling endpoint updated status/phase/progress but never wrote output to workflow_executions. This adds two complementary persistence paths: 1. Status polling endpoint now writes output on terminal status 2. New persist_results_to_db orchestrator activity writes output directly at workflow completion (belt-and-suspenders for closed-browser case) Co-Authored-By: Claude Opus 4.6 <[email protected]>
- Add maxTurns as UI-editable field on mastra-run, mastra/execute, and durable/run actions, flowing through orchestrator to durable-agent - Increase default maxIterations from 10 to 50 for durable-agent - Accumulate all tool calls across ReAct turns in AgentWorkflowResult - Strengthen execution prompt to ensure agent completes all plan steps - Update CLAUDE.md with durable-agent, node-sandbox, and current architecture Co-Authored-By: Claude Opus 4.6 <[email protected]>
- Add automated kill-and-restart durability test (scripts/test-dapr-durability.ts) that proves durable-agent workflows survive pod kills and resume to completion - Fix crash recovery gap: pass previous tool results through generator to repair Redis conversation state after crashes between saveToolResults and next callLlm - Add safety check that detects broken assistant messages (tool_calls without matching tool results) and truncates conversation to last clean state - Add eager WorkflowRuntime initialization at startup so pending workflows replay immediately on pod restart instead of waiting for first /api/run request - Add user message deduplication guard in callLlm activity Co-Authored-By: Claude Opus 4.6 <[email protected]>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
workflow-spec/v2and tighten lint/decompile/compile behaviorworkflow-controlstep type (schema, linting, UI node/config, orchestrator execution)respond/__workflow_builder_control) and use explicit durable control flowValidation