A declared output is written through a capture tool, but the dispatch-time
harness check read only the workflow's needs.captures. A workflow that
declared outputs without also writing needs: {captures: true} could name a
Kit profile whose adapter serves no capture tools; a required output then
parked with "output %q is required" instead of the profile being refused.
WorkflowInterface.Needs now reports captures for a non-empty Outputs beside
the declared needs, and the profile check and Kit launcher read it. The
needs: block keeps its meaning for a workflow that captures without
declaring outputs.
- internal/domain/workflow/task: Needs() derives captures; DeclaredNeeds
carries only the declared block.
- internal/daemon: pinTaskProfile and runKitTask read Needs().
- tests: parse-time derivation, and pinTaskProfile refusing an
outputs-declaring workflow against a Kit adapter with no capture MCP.
3.2 KiB
Workflows
Authority: docs/architecture/agent-system.md for the lifecycle,
docs/prds/event-automation.md for inputs, repository modes and profiles.
Workflow definitions are YAML stored as a control-plane resource. The State Store validates them and the agent compiles them, each against the step vocabulary it registers, so both must see the same step types.
Adding a step type
- Write the factory in
internal/domain/workflowas aStepType. It decodes and validates its settings and returns an error for anything it cannot honour. The factory runs when a definition is saved and again when it is compiled, so a bad definition is refused on save. - Register it through a provider in
internal/infrastructure/workflowsteps/workflowsteps.go. That one list is what every process registers, which is how the validating and executing sides agree. - An agent-driven step builds on
AgentStage; do not start a second path to the agent runner. - A step that needs no repository goes in
repoFreeStepTypesindefinition.go, or workflows declaringrepository: noneoroptionalcannot use it. A step that needs one may assumetc.Dirholds a worktree only when the workflow's repository isrequired. - A step that ends the workflow sets
tc.Outcome. A workflow that runs out of steps without an outcome is parked as a definition bug. - Add a parse test (settings accepted and refused) and a run test with fake
Treesand agent runner.
Changing the definition format
- The fields a caller needs without the engine (inputs, repository, profile)
live in
task.WorkflowInterface, because the dashboard process must not import the workflow engine (cmd/archie-ui/architecture_test.go). Anything a binding or the dashboard checks goes there. - The YAML decoder rejects unknown keys. Existing stored definitions must keep parsing, so a new key is optional with a default that preserves today's behaviour.
- A declared
outputs:entry is written through a capture tool, so it forces the captures need:task.WorkflowInterface.Needsresolves that from the declaration rather than the author also writingneeds.captures. A Kit profile whose harness serves no capture tools is refused before dispatch. - Examples in
examples/workflows/are parsed against the registered vocabulary byTestExampleWorkflowsParse. Add one for a new capability.
Enabling and disabling
The workflow-enablement control-plane resource lists, per org, the workflows
that org has disabled; anything unlisted is enabled. Its readers are the
daemon's binding dispatch (a disabled workflow's binding is skipped like an
unarmed one), the dashboard's binding save (refused) and binding list
(workflow_disabled), and work requests. A new reader of "may this org run
this workflow" goes through task.WorkflowEnablement.Enabled, never a second
list. The dashboard acts in the default org until access resolves a caller's
own.
Starting a workflow from chat
task_spawn accepts workflow and an inputs object. The gateway forwards
inputs through the task creation contract; it does not interpret individual
workflows. The daemon checks inputs against the pinned WorkflowInterface
before acquiring a container and parks invalid requests with the reason.