How do I install the Workflow skill?

Run npx skillmds@latest add alirezarezvani/workflow in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Workflow safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Workflow?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Workflow free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Workflow?

Alireza Rezvani (@alirezarezvani) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Workflow by alirezarezvani — Agent Skill

Name: Workflow
Author: Alireza Rezvani

Workflow

Use to run a dynamic, goal-driven workflow — orchestrate a multi-step goal by composing skills, tools, and MCPs, with a quality gate and verification at each step. Trigger on "/workflow", "run the <name> workflow", "orchestrate X", or a multi-step recurring goal ("draft the investor update", "give me my morning brief"). Follows references/sops/dynamic-workflows.md and the wf-*.md cards. Respects the autonomy/cadence gate (auto on reversible/internal, draft-then-approve on external/regulated).

Alireza Rezvani f196f5a 2.4 KB Updated Sep 17, 2026 20.4k repo stars

File contents

Workflow — run a dynamic workflow

Orchestrate a goal end-to-end: plan → route → compose skills/tools/MCPs → gate each step → verify → fail safe. The engine for speed + quality. Pattern: references/sops/dynamic-workflows.md. Named workflows: references/sops/wf-*.md.

How to run

Resolve the goal. Named workflow (/workflow investor-update) → load its wf-*.md card. Free-form goal → first run /structure to get Context · Outcome · Goal · Success criteria, then plan the steps.
Check the autonomy gate per step. Internal/reversible/no-PHI → may run unattended. External/irreversible/regulated → stop and have the user approve (draft-never-send; never to a sensitive recipient without approval).
Execute step by step. For each: run the action (a /skill, a tools/*.py, or an MCP call) → run its gate/verify check → branch on the result (e.g. missing data → flag + ask, don't guess). Reuse existing capabilities; don't rebuild.
Fail safe. On error: read it, retry/repair if trivial, else escalate to the user with what's blocked. Never silently drop a step.
Verify done. Confirm the success criteria are met (evidence, not assertion) before reporting complete.
Deliver. Internal → your sanctioned cloud / chat. External → draft to the user. Log meaningful decisions to decisions/log.md.

Build / refine a workflow

New recurring multi-step goal that composes ≥2 capabilities and has a definable "done"? Author a references/sops/wf-<name>.md card (format in dynamic-workflows.md) and add it to the inventory. Don't workflow-ify one-offs (Simplicity). Don't overwrite a workflow card without asking.

Guardrails

All of CLAUDE.md. Especially: the autonomy gate per step, quality gate before "done", cite-don't-invent, no confidential data persisted, draft-never-send external.

---name: workflowdescription: Use to run a dynamic, goal-driven workflow — orchestrate a multi-step goal by composing skills, tools, and MCPs, with a quality gate and verification at each step. Trigger on "/workflow", "run the <name> workflow", "orchestrate X", or a multi-step recurring goal ("draft the investor update", "give me my morning brief"). Follows references/sops/dynamic-workflows.md and the wf-*.md cards. Respects the autonomy/cadence gate (auto on reversible/internal, draft-then-approve on external/regulated).---# Workflow — run a dynamic workflowOrchestrate a goal end-to-end: **plan → route → compose skills/tools/MCPs → gate each step → verify → fail safe.** The engine for speed + quality. Pattern: `references/sops/dynamic-workflows.md`. Named workflows: `references/sops/wf-*.md`.## How to run1. **Resolve the goal.** Named workflow (`/workflow investor-update`) → load its `wf-*.md` card. Free-form goal → first run `/structure` to get Context · Outcome · Goal · Success criteria, then plan the steps.2. **Check the autonomy gate per step.** Internal/reversible/no-PHI → may run unattended. External/irreversible/regulated → **stop and have the user approve** (draft-never-send; never to a sensitive recipient without approval).3. **Execute step by step.** For each: run the action (a `/skill`, a `tools/*.py`, or an MCP call) → **run its gate/verify check** → **branch** on the result (e.g. missing data → flag + ask, don't guess). Reuse existing capabilities; don't rebuild.4. **Fail safe.** On error: read it, retry/repair if trivial, else escalate to the user with what's blocked. Never silently drop a step.5. **Verify done.** Confirm the success criteria are met (evidence, not assertion) before reporting complete.6. **Deliver.** Internal → your sanctioned cloud / chat. External → draft to the user. Log meaningful decisions to `decisions/log.md`.## Build / refine a workflowNew recurring multi-step goal that composes ≥2 capabilities and has a definable "done"? Author a `references/sops/wf-<name>.md` card (format in `dynamic-workflows.md`) and add it to the inventory. Don't workflow-ify one-offs (Simplicity). Don't overwrite a workflow card without asking.## GuardrailsAll of `CLAUDE.md`. Especially: the autonomy gate per step, quality gate before "done", cite-don't-invent, no confidential data persisted, draft-never-send external.

Frequently asked questions