…e-connection-usage

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…, Blocked status

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…points

- StepRequirement: unified skill-based model with native:* namespace
  for built-in capabilities and {provider}:{service} for connections
- StepStatus::Blocked + WorkflowStatus::Blocked for unmet requirements
- StepKind::ToolCall: single tool invocation without full agent loop
- Richer Script: cwd, env, timeout_secs, output_format, shell type
- AgentRun: skills, model override, max_iterations fields
- CheckpointStrategy: Internal (Redis/TTL) and External (tool call)
- is_stuck() detection for blocked workflow state
- Executor capability registration: supports() + available_skills()
- Requirement checking in WorkflowRunner.run_next()
- 37 comprehensive unit tests (up from 10)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- WorkflowStore trait in distri-types with CRUD + discovery methods
  (list_public, star, unstar, clone) following SkillStore patterns
- WorkflowRecord with definition as serde_json::Value (avoids crate dep)
- WorkflowListItem lightweight type for list endpoints
- InMemoryWorkflowStore in distri-stores for testing/OSS
- Workflow CRUD routes in distri-server (GET/POST /workflows, etc.)
- workflow_store field added to InitializedStores

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- DistriStepExecutor: real HTTP executor using Distri client
  - ApiCall: resolves context vars, makes real HTTP requests
  - ToolCall: delegates to server via client.call_tool()
  - Checkpoint: passes through
  - Script/AgentRun/Condition: placeholder (not yet wired)
  - Context template resolution ({context.key} in URLs and bodies)
- Client API: list/get/create/update/delete/push workflow methods
- CLI: workflow run (uses real executor), push, list, status
- Fix deepagent Cargo.toml paths (../../ -> ../)
- Generic test_workflow.json sample (API test, not WritePlace-specific)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Input validation:
- with_input() validates input JSON against input_schema (JSON Schema)
- Uses jsonschema crate for validation
- Rejects missing required fields, wrong types
- No schema = accepts anything (backward compatible)
- Sets status to Running on successful initialization

Events:
- WorkflowEvent enum: WorkflowStarted, StepStarted, StepCompleted,
  StepFailed, WorkflowCompleted
- EventSink trait: receives events during execution
- TracingEventSink: logs events via tracing
- NoopEventSink: default (no events)
- WorkflowRunner::with_events(store, executor, sink) to enable events
- run_all() emits started/completed events
- run_next() emits step started/completed/failed events

44 tests + 1 doc-test, all passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

WorkflowAgent executes a workflow DAG instead of an LLM loop:
- New AgentConfig::WorkflowAgent variant with WorkflowAgentDefinition
- Implements BaseAgent trait (invoke_stream, get_dag, etc.)
- ContextEventSink bridges WorkflowEvents to agent event channel
- ContextStepExecutor handles ApiCall, ToolCall, Checkpoint steps
- Input validation via JSON Schema (input_schema field)
- Streams step progress as text messages via SSE
- DAG visualization via get_dag() returns workflow step graph
- All existing match arms updated for new variant

44 workflow tests passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- WorkflowSession: wraps runner + event channel, public API for apps
  - new(client, workflow) → take_events() → run() / run_with_input()
  - Events emitted via mpsc channel as WorkflowEvent (SSE-compatible JSON)
- ChannelEventSink: bridges WorkflowRunner events to tokio mpsc channel
- Same event contract as server-side WorkflowAgent
- Client apps can: spawn runner in task, forward events as SSE to their
  own clients, distrijs renders them in Chat UI

Usage:
  let mut session = WorkflowSession::new(client, workflow);
  let mut rx = session.take_events().unwrap();
  tokio::spawn(async move { session.run().await });
  while let Some(event) = rx.recv().await {
      // Forward as SSE to frontend
  }

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- test_workflow_event_serialization: verifies SSE-compatible JSON format
  with "event" tag, round-trip serialization
- test_workflow_session_events: end-to-end against live server,
  verifies WorkflowStarted/StepStarted/StepCompleted/WorkflowCompleted
- test_workflow_session_with_input: context variable resolution from input

All 3 tests pass (1 unit + 2 integration against running distri-cloud).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…tion

- resolve.rs: centralized namespace resolution with {input.X}, {steps.X.Y}, {env.X}
  - Full-value references preserve types (arrays, objects, numbers)
  - Backward compat: {context.X} still works
  - resolve_step_input: explicit mapping or full context pass-through
  - 14 unit tests for resolution

- types.rs:
  - Add `input` field to WorkflowStep for explicit input mapping
  - `detect_cycles()` on WorkflowDefinition (DFS cycle detection)
  - All runtime fields default (status, context, current_step, notes, timestamps)
  - StepStatus/WorkflowStatus now Default (Pending)

- store.rs: auto-store step results at steps.<step_id> in structured context

- executor.rs: resolve step input before passing to executor, cycle check in run_all()

- workflow_agent.rs:
  - Save user message to thread on invocation
  - Save summary message on completion
  - Populate env namespace from executor context
  - Use centralized resolve (removed local copies)

- client workflow_executor.rs: use centralized resolve, updated tests

68 workflow tests + 5 client tests passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Removed from: WorkflowDefinition, WorkflowEvent, WorkflowListItem,
WorkflowRecord, NewWorkflow, WorkflowFilter, all stores, all routes,
CLI, client, tests. Tags and name are sufficient for categorization.

68 tests passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>