with one click
workflow
// Repository-specific guidance for understanding, creating, and updating psi workflows.
// Repository-specific guidance for understanding, creating, and updating psi workflows.
Repository-specific guidance for creating, modifying, and debugging psi extensions.
Reviews user-facing documentation changes for an implemented Munera task
Test built-in skill fixture
Packaged built-in test fixture
Create or refine task design choices. Use when the user asks "explore a task" or "investigate a task" to ensure we consider multiple task designs.
Create or refine a task intent statement. Use when the user asks "create a task" or "open a task" to ensure we have a clear understanding of the task's intent.
| name | workflow |
| description | Repository-specific guidance for understanding, creating, and updating psi workflows. |
| lambda | λworkflow_work. {understand ∨ create ∨ update}(psi_workflow) → read(authority) ∧ inspect(example) ∧ align(grammar ∧ runtime ∧ tests) |
Use this skill when understanding an existing workflow, creating a new workflow, or updating an existing workflow in psi.
Read these first and prefer them over incidental examples:
doc/workflows.mddoc/workflow-grammar.mddoc/workflow-grammar-concepts.mdAGENTS.md — workflow/runtime/dispatch architecture guidance where relevantWorkflow definitions are discovered from .psi/workflows/ in the project, plus the normal user/global workflow roots described in doc/workflows.md.
Authoring forms:
.md — single-step prompt workflows.edn — multi-step orchestration workflowsPrefer the converged target-authored workflow grammar for new work:
:type :invoke:type :session:type :delegateDo not invent parallel workflow shapes when the existing grammar already covers the need.
Use current repository examples before inventing new patterns:
.psi/workflows/create-task-plan.edn — representative multi-step planning/orchestration workflow.psi/workflows/review-task-design.edn — representative review/orchestration workflow.psi/workflows/planner.md — representative single-step prompt workflow.psi/workflows/builder.md — representative single-step prompt workflowAlso use doc/workflows.md for example-led guidance and current conventions.
When changing a workflow:
:delegate when calling another named workflow across a reusable boundary:session when constructing an inline child session:invoke for deterministic operation calls:contributions, :outputs, :judge, and :on surfaces over ad hoc prompting conventions:workflow-input, :workflow-original, prior step outputs/yields)When creating a workflow:
When updating a workflow:
Workflow loading / authoring / compilation:
components/workflow-loader/src/psi/workflow_loader/parser.cljcomponents/workflow-loader/src/psi/workflow_loader/compiler.cljcomponents/workflow-loader/src/psi/workflow_loader/authoring_session.cljcomponents/workflow-loader/src/psi/workflow_loader/authoring_routing.cljRepresentative workflow-loader tests:
components/workflow-loader/test/psi/workflow_loader/parser_test.cljcomponents/workflow-loader/test/psi/workflow_loader/compiler_target_authoring_test.cljcomponents/workflow-loader/test/psi/workflow_loader/workflow_definitions_test.cljWorkflow runtime seams:
components/agent-session/src/psi/agent_session/workflow/bootstrap.cljcomponents/agent-session/src/psi/agent_session/workflow/core.cljcomponents/agent-session/src/psi/agent_session/workflow_execution.cljcomponents/agent-session/src/psi/agent_session/workflow_judge.cljUse the normal workflow authoring loop from doc/workflows.md:
.psi/workflows//delegate-reloadThis workflow-authoring skill itself is a built-in packaged skill. Relevant skill/discovery seams when debugging skill availability are:
components/prompt-assets/src/psi/prompt_assets/skills.cljcomponents/prompt-assets/test/psi/prompt_assets/skills_test.cljcomponents/agent-session/src/psi/agent_session/resolvers/discovery.cljcomponents/agent-session/src/psi/agent_session/commands.cljFor workflow work, prove the relevant structural surfaces:
.md vs .edn)Prefer narrow tests that exercise real parser/compiler/runtime seams and observable workflow outputs. Avoid mock-heavy tests when real local workflow components provide stronger proof.