Skip to content

Workers

Worker Types

TypePurpose
CodeWorkerImplements features, fixes bugs, refactors
ContentWorkerDrafts content, maintains voice consistency
ResearchWorkerAnalysis, market research, competitive intel
OpsWorkerReports, monitoring, automation
PersonaWorkerA worker that runs as a chat bot, such as the Setup bot
LibraryShared utilities (not directly executable)

HQ ships 43 workers in core/workers/public/. The full list, with each worker’s type and description, is core/workers/registry.yaml. HQ regenerates that file from each worker.yaml, so don’t edit it by hand.

Dev Team (18 Workers)

A development team at core/workers/public/dev-team/.

WorkerTypeSkills
project-managerOpsWorkerPRD lifecycle, issue selection, learning aggregation
task-executorOpsWorkerAnalyze issues, route to workers, validate completion
pipeline-conductorOpsWorkerRouting rules for /run-project --pipeline
architectCodeWorkerSystem design, API design, architecture decisions
backend-devCodeWorkerAPI endpoints, business logic, backend implementation
frontend-devCodeWorkerReact/Next components, pages, UI implementation
database-devCodeWorkerSchema design, migrations, query optimization
infra-devCodeWorkerCI/CD pipelines, deployment configurations, monitoring setup
dev-qa-testerCodeWorkerTesting, browser automation, accessibility, demo accounts
code-reviewerCodeWorkerPR review, merge management, code quality gating
reality-checkerCodeWorkerFinal quality gate that cross-validates QA findings
codex-coderCodeWorkerCode generation through the OpenAI Codex SDK
codex-debuggerCodeWorkerDebugging and root-cause analysis through the OpenAI Codex SDK
codex-reviewerCodeWorkerSecond-opinion code review through the OpenAI Codex SDK
context-managerCodeWorkerDiscover, maintain, and audit project context
motion-designerCodeWorkerAnimations, transitions, visual polish, image generation
knowledge-curatorCodeWorkerProcess learnings, update knowledge bases, maintain documentation
product-plannerCodeWorkerPRD creation, technical specs, API contracts

Content Team

Multi-perspective content review at core/workers/public/content-*/.

WorkerFocusSkills
content-brandVoice & toneVoice analysis, tone check, messaging alignment
content-salesConversionConversion copy, CTAs, value props
content-productAccuracyFeature accuracy, technical claims verification
content-legalComplianceRegulatory compliance, claim verification
content-sharedShared codeTypeScript library for analysis, scoring, and reporting. It has no worker.yaml, so it is not in the registry or the count of 43

Other Teams

TeamWorkersPath
Socialsocial-strategist, social-reviewer, social-publisher, social-verifier, social-sharedcore/workers/public/social-*/
Gardenergarden-scout, garden-auditor, garden-curatorcore/workers/public/gardener-team/

Standalone Workers

WorkerPurpose
setupThe Setup bot that gets a new HQ working
security-scannerSecurity scanning and vulnerability detection
accessibility-auditorWCAG 2.2 AA auditing
qa-testerAdversarial website testing with screenshot evidence
performance-benchmarkerCore Web Vitals, load testing, capacity planning
exec-summaryExecutive summaries with strict word limits
frontend-designerUI generation
paper-designerPaper Desktop design bridge
site-builderBuild and deploy local business websites from a template and config
ontologyMaintains a company’s ontology and signals
knowledge-taggerClassifies knowledge docs against the ontology
pretty-mermaidMermaid diagram rendering
ascii-artistASCII banner images for social posts and OG images

Worker YAML Schema

Every worker is defined by a worker.yaml:

worker:
id: my-worker
name: My Worker
type: CodeWorker
version: "1.0"
execution:
mode: on_demand # most bundled workers use on_demand; the template uses on-demand
max_runtime: 30m
context:
base:
- core/knowledge/public/Ralph/ # Knowledge to load
dynamic: []
exclude: []
verification:
post_execute:
- check: typescript
command: npm run typecheck
approval_required: false
tasks:
source: companies/my-company/projects/my-project/prd.json
output:
destination: workspace/reports/
format: markdown
instructions: |
Multi-line instructions for the worker...

Creating Custom Workers

Use /newworker to scaffold:

/newworker

The wizard asks for:

  1. Worker ID and name
  2. Type (CodeWorker, ContentWorker, etc.)
  3. Skills (name + description for each)
  4. Knowledge context paths
  5. Execution rules

It then asks whether the worker belongs to one company or ships with HQ, and generates it at companies/{co}/workers/{id}/ or core/workers/public/{id}/. The default is company scope. Choose core/ only for a worker meant to ship with HQ releases: /update-hq replaces core/, so a worker you add there for yourself is lost on the next update. The generated folder:

{id}/
├── worker.yaml
├── skills/
│ ├── skill-one.md
│ └── skill-two.md
└── README.md