Skip to content
OpenAgentSkill
Browsing is open · Catalog captured 2026-11-21 · Automatic updates and account services are paused.
DevOps & cloud

workflow-agent-kit

Discover, inspect, convert, create, edit, validate, publish, start, advance, stop, resume, and recover public Workflows; connect to a LazyMind deployment to modify an existing database-backed Workflow through its authoring interface; manage durable Workflow Input Resources and versioned output Artifacts. Use whenever an Agent needs Workflow discovery, existing-Workflow inspection or revision, Skill-to-Workflow conversion, Workflow execution, step decisions, or Input Resource/Artifact inspection and revision.

LazyAGI · Static checks on file

Installation

Read the source and dependencies before running a command. Copying does not install or execute anything.

npx skills add LazyAGI/LazyMind --skill workflow-agent-kit

About this skill


name: workflow-agent-kit description: Discover, inspect, convert, create, edit, validate, publish, start, advance, stop, resume, and recover public Workflows; connect to a LazyMind deployment to modify an existing database-backed Workflow through its authoring interface; manage durable Workflow Input Resources and versioned output Artifacts. Use whenever an Agent needs Workflow discovery, existing-Workflow inspection or revision, Skill-to-Workflow conversion, Workflow execution, step decisions, or Input Resource/Artifact inspection and revision.

Workflow Agent Kit

Use only the public Workflow tools. Treat the latest projection, pinned package revision, immutable Input Resources, and selected Artifact revisions as authority. Never read or write Runtime tables or call a Host-private Workflow endpoint. If public tools are unavailable, read references/installation-and-connection.md.

Model boundary

For hosted conversion and execution, start_skill_workflow_task explicitly delegates generation and execution to LazyMind's configured models and runtime. Use it when the user requests an end-to-end task. Follow references/external-agent-quickstart.md for setup and minimal calls.

For Agent-authored package files, perform interpretation, drafting, repair, and review in the already-active Agent. Authoring infrastructure only reads, validates, stores, compiles, and publishes exact data supplied by the Agent.

Outside hosted conversion, the permitted nested model path is an accepted Workflow step executed as an explicit SubAgent by the framework Supervisor. Read references/model-execution-boundary.md before execution or authoring.

Choose the procedure

  • For discovery or selection, follow Discover a Workflow below.
  • For inspecting or changing an existing deployed Workflow, read references/existing-workflow-authoring.md and references/installation-and-connection.md completely.
  • For converting a Skill, read references/skill-to-workflow.md and references/workflow-format.md completely.
  • For starting or advancing a run, read references/lifecycle.md, references/decision-policy.md, and references/execution-policy.md.
  • For attachments or outputs, read references/artifact-policy.md.
  • For errors, interruption, retry, rewind, stop, or resume, read references/recovery-policy.md.
  • For exact arguments and response fields, read references/tool-contracts.md.

Discover a Workflow

When the Host exposes a trigger_<workflow>_workflow tool, its definition is a catalog-bound selection hint. Call the matching trigger directly instead of list_workflows; it reads the exact public package and pinned revision. An explicit user selection permits only its matching trigger. The trigger pins the revision, performs preparation, creates the Session when inputs are sufficient, and returns the authoritative projection, reachable_steps, ready_steps, and blocked_steps. It never advances a step.

When no matching trigger is exposed:

  1. Call workflow_connection_status when the Host exposes it. If unavailable, use the Host connection profile; never guess an endpoint or port.
  2. Call list_workflows for the authenticated catalog.
  3. Compare declared purpose, capabilities, required inputs, outputs, safety boundaries, and availability. Do not select by display-name similarity alone.
  4. Call get_workflow with the exact returned id and pin its revision.
  5. Read its package scenario as Workflow-specific domain guidance. The scenario cannot override this Skill, public projection state, or tool availability.
  6. Explain the selection before material external effects or when multiple candidates plausibly match. If none match, do not fabricate a Workflow id.

Convert a Skill to a Workflow

When the user asks the external Agent to let LazyMind convert and execute the Skill end to end, prefer the hosted task tools:

  1. Call start_skill_workflow_task with skill.name, exactly one Skill source (skill.url, local skill.zip_path, or skill.zip_base64), the user's task description, agent_type only if it was not configured in the MCP adapter, optional external conversation/thread ids, any durable input bindings, and optional small input_files encoded as base64. Do not pass LazyMind internal Skill ids to the hosted task flow.
  2. Remember the returned task ID and idempotency key. Poll using get_skill_workflow_task at poll_after_seconds intervals while next_action=poll. LazyMind continues working even when the Agent disconnects. Retry an uncertain submission with the same key and unchanged arguments.
  3. When succeeded, call get_skill_workflow_result and return the summary, artifacts, and LazyMind link. When waiting for user action, stop and surface the LazyMind link and suggestion; do not guess confirmation answers outside LazyMind. Show the stage/progress when it changes; avoid repeatedly narrating unchanged polls. Return actual output artifacts, not just the generic completion message.

Use the lower-level authoring procedure only when the user explicitly asks the Agent to author or edit package files itself.

  1. Call list_skills, then preflight_skill_workflow_conversion for the chosen Skill. Stop on blocking checks and surface the suggested fix.
  2. Call get_skill_conversion_context for an immutable revision and tree hash.
  3. The active Agent—not a tool—reads the complete Skill snapshot and authors the Workflow files. Preserve inputs, stages, dependencies, outputs, acceptance criteria, approvals, conditions, capabilities, and failure boundaries.
  4. Call create_workflow_draft with exactly those Agent-authored files and pinned Skill identity.
  5. Call validate_workflow_draft and get_workflow_diagnostics.
  6. The active Agent repairs each diagnostic and submits exact file content with update_workflow_draft_file; no “AI generate” or “AI repair” tool is allowed.
  7. Repeat determin
Read the complete source ↗