mcpField
A
agentic-workflow-designer

agentic-workflow-designer

github.com/sylphxaiAgent workflows149 installsUpdated today
Repository

About agentic-workflow-designer

agentic-workflow-designer is a skill for designing reliable, auditable agent workflows. It helps teams turn goals and constraints into a workflow specification with review gates, recovery paths, and validation evidence.

Key features

  1. 01Workflow goals and risks
  2. 02Deterministic and judgment steps
  3. 03Memory and failure recovery planning
  4. 04State machine and tool contracts
  5. 05Validation and approval gates

Use cases

  1. 01Designing auditable agent workflows
  2. 02Adding approvals to high-risk actions
  3. 03Defining recovery and handoff paths

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/sylphxai/skills --skill agentic-workflow-designer

After installing, type/agentic-workflow-designerin Claude Code to confirm it loaded.

Bundled filesagents/openai.yamlreferences/agentic-workflow-systems.md

Details

From the project README

Agentic Workflow Designer

Use this skill to design reliable agent workflows that produce auditable work instead of uncontrolled automation.

Workflow

  1. Identify goal, user, risk, autonomy level, tools, data, output artifact, and acceptance evidence.
  2. Read references/agentic-workflow-systems.md.
  3. Split workflow into deterministic steps, judgment steps, tool steps, review gates, and human approval gates.
  4. Define memory/context, failure recovery, observability, evals, and handoff artifacts.
  5. Produce a workflow spec with state machine, roles, tool contract, and validation plan.

When not to use

  • Do not use for generic advice the base model already handles without this skill's specific artifact contract.

Guardrails

  • Do not automate irreversible, payment, legal, secret, production, or customer-impacting actions without explicit approval gates.
  • Do not rely on hidden chain-of-thought or tribal memory as workflow state.
  • Require machine-readable outputs where downstream agents depend on them.

Output format

Workflow goal:
Autonomy level:
Inputs/tools:

Workflow states:
- <state> -> action, evidence, owner, failure path

Approval gates:
- <gate> -> why, approver, artifact

Evals/observability:
- <check or metric>

Comments

Markdown supported