Skip to content

Workflows

A workflow is a predefined sequence of agent steps that solves a specific type of problem. Workflows define which agents run, in what order, with what inputs and outputs.

Each workflow is a YAML file that describes:

  • Steps: ordered sequence of agent invocations
  • Agents: which agent handles each step
  • Inputs/Outputs: what each step receives and produces
  • Dependencies: which steps must complete before others start
ShortcutWorkflowAgentsPurpose
/featurefeature-developmentprofessor-x, tony-stark, captain-america, dev agents, hawkeyeBuild a new feature end-to-end
/bugfixbug-fixdev agents, hawkeyeDiagnose and fix a bug
/reviewcode-review-pipelinedev agents, punisher, hawkeyeComprehensive code review
/refactortech-debt-reductiontony-stark, dev agents, hawkeyeReduce technical debt
/upgradedependency-upgradethor, dev agents, hawkeyeUpdate dependencies safely
ShortcutWorkflowAgentsPurpose
/releaserelease-cyclecaptain-america, dev agents, thor, hawkeyePrepare and ship a release
/hotfixhotfix-releasedev agents, thor, hawkeyeEmergency production fix
/mvpmvp-launchprofessor-x, tony-stark, dev agents, star-lordLaunch a minimum viable product
ShortcutWorkflowAgentsPurpose
/sprintsprint-cyclecaptain-america, professor-x, dev agentsPlan and run a sprint
/onboardonboarding-projectprofessor-x, tony-stark, captain-americaSet up a new project
/docsdocumentation-sprintdev agents, stormWrite or update documentation
ShortcutWorkflowAgentsPurpose
/campaignmarketing-campaignstar-lord, loki, gamora, ms-marvelLaunch a marketing campaign
/seoseo-content-pipelineblack-widow, storm, jean-greySEO content production
ShortcutWorkflowAgentsPurpose
/securitysecurity-auditpunisher, microchip, dev agentsFull security audit
/experimentexperimentationbeast, dev agents, hawkeyeRun an A/B test or experiment

Every workflow follows this structure:

name: feature-development
description: Build a new feature end-to-end
trigger: /feature
risk: MEDIUM
steps:
- id: spec
agent: professor-x
action: Write feature specification
inputs: [user_request]
outputs: [spec.md]
- id: plan
agent: tony-stark
action: Design technical architecture
inputs: [spec.md]
outputs: [plan.md]
depends_on: [spec]
- id: tasks
agent: captain-america
action: Break down into tasks
inputs: [spec.md, plan.md]
outputs: [tasks.md]
depends_on: [plan]
- id: implement
agent: bruce-banner
action: Implement the feature
inputs: [spec.md, plan.md, tasks.md]
outputs: [code, tests]
depends_on: [tasks]
- id: test
agent: hawkeye
action: Write and run tests
inputs: [code, spec.md]
outputs: [test-report.md]
depends_on: [implement]

Dependencies create a directed acyclic graph (DAG). Steps without dependencies can run in parallel (when the platform supports it). Steps with depends_on wait for their dependencies to complete.

spec → plan → tasks → implement → test
↘ frontend → ↗

Each step declares:

  • inputs: files or data from previous steps (or user_request for the first step)
  • outputs: files produced by the agent

Jarvis verifies that each output exists before proceeding to the next step. If an output is missing, the workflow pauses and alerts the user.

During execution, Jarvis maintains a _manifest.yaml in the output directory that tracks:

  • Current step and status
  • Timestamps for each step
  • Produced outputs
  • Any blockers or warnings

You can create custom workflows for your team’s specific needs. See the Custom Workflows guide for details.