Rudder Create Agent

Rudder Create Agent Skill

Use this skill when you are asked to hire or create an agent in Rudder.

Preconditions

You need either:

• board access, or
• agent permission canCreateAgents=true in your org

If you do not have this permission, escalate to your CEO or board.

This workflow is CLI-first.

• Use rudder ... --json for structured reads and mutations.
• Use references/cli-reference.md as the canonical command catalog for this skill.
• Treat references/api-reference.md as internal/debug/compatibility documentation, not the normal runtime interface.
• Do not create agent directories, instruction files, or org metadata manually as a fallback.
• If CLI auth is unavailable in a heartbeat run, stop and report the auth problem instead of mutating the filesystem.

Workflow

1. Confirm identity and organization context.

rudder agent me --json

If this returns {"error":"Agent authentication required"}, treat it as a run-auth failure:

• do not ask for RUDDER_API_KEY inside the heartbeat
• do not fall back to manual filesystem creation
• stop and report that injected agent authentication is missing or invalid for this run

1. Discover available adapter configuration docs for this Rudder instance.

rudder agent config index

1. Read adapter-specific docs for the runtime you plan to use.

rudder agent config doc codex_local
rudder agent config doc claude_local

1. Compare existing agents and redacted configurations in your organization.

rudder agent list --org-id "$RUDDER_ORG_ID" --json
rudder agent config list --org-id "$RUDDER_ORG_ID" --json
rudder agent config get "" --json

1. If the role needs organization skills on day one, inspect or import them before hiring.

rudder skill list --org-id "$RUDDER_ORG_ID" --json
rudder skill get "" --org-id "$RUDDER_ORG_ID" --json
rudder skill file "" --org-id "$RUDDER_ORG_ID" --path SKILL.md --json
rudder skill import --org-id "$RUDDER_ORG_ID" --source "" --json
rudder skill scan-local --org-id "$RUDDER_ORG_ID" --roots "" --json
rudder skill scan-projects --org-id "$RUDDER_ORG_ID" --project-ids "" --workspace-ids "" --json

1. Draft the hire payload.

Required thinking:

• role / title / optional name
• name is optional; if omitted, Rudder assigns a distinct personal name automatically
• omit icon for normal hires; Rudder assigns a DiceBear Notionists avatar automatically
• only set icon when preserving an explicit DiceBear avatar reference or an uploaded asset: image avatar reference provided by the board/UI
• reporting line (reportsTo)
• adapter type
• optional desiredSkills from the organization skill library
• adapter and runtime config aligned to this environment
• capabilities
• structured role/persona instructions for the new agent (promptTemplate when the CLI payload is the available surface; Rudder materializes this as SOUL.md)
• source issue linkage (sourceIssueId or sourceIssueIds) when this hire came from an issue

role is a fixed Rudder enum, not a free-form job title. Use one of: ceo, cto, cmo, cfo, engineer, designer, pm, qa, devops, researcher, general. Put specialization in title, capabilities, and promptTemplate. For example, a Founding Engineer hire should use "role": "engineer" and "title": "Founding Engineer", not "role": "founding_engineer".

Do not copy Rudder's shared filesystem, memory, language, or safety contract into the hire prompt. Rudder injects that operating contract from runtime code for supported local runtimes. The hire-specific prompt should only define the new agent's role, identity, scope, tone, and durable responsibilities.

Draft promptTemplate as a durable SOUL document, not a one-line command. Use these sections when the role is not trivial:

• Opening: one sentence that captures who the agent is
• Mission: the outcome this agent owns
• Responsibilities: durable duties and ownership boundaries
• Boundaries: what the agent should not do or should escalate
• Decision Principles: role-specific judgment rules
• Voice: how the agent should communicate
• Continuity: what should become memory or explicit instruction updates over time

1. Submit the canonical hire request.

rudder agent hire --org-id "$RUDDER_ORG_ID" --payload '{
  "role": "cto",
  "title": "Chief Technology Officer",
  "reportsTo": "",
  "capabilities": "Owns technical roadmap, architecture, staffing, execution",
  "desiredSkills": ["vercel-labs/agent-browser/agent-browser"],
  "agentRuntimeType": "codex_local",
  "agentRuntimeConfig": {
    "cwd": "/abs/path/to/repo",
    "model": "o4-mini",
    "promptTemplate": "# SOUL.md -- CTO Persona\n\nYou are the CTO.\n\n## Mission\nOwn technical strategy, architecture, engineering execution, and quality bars.\n\n## Responsibilities\n- Set technical direction and execution standards.\n- Review architecture and staffing trade-offs.\n- Keep delivery risks visible and actionable.\n\n## Boundaries\n- Do not approve risky shortcuts without naming the trade-off.\n- Escalate product or budget ambiguity instead of guessing.\n\n## Decision Principles\n- Prefer simple architectures with explicit trade-offs.\n- Treat reliability, developer velocity, and product learning as linked constraints.\n\n## Voice\nDirect, specific, and evidence-led.\n\n## Continuity\nPreserve durable technical standards, repeated failure patterns, and long-running architecture decisions in memory or explicit instructions."
  },
  "runtimeConfig": {"heartbeat": {"enabled": true, "intervalSec": 300, "wakeOnDemand": true, "maxConcurrentRuns": 3}},
  "sourceIssueId": ""
}' --json

agent hire is the canonical surface because it preserves the real server behavior:

• if the organization does not require approval, it creates the agent directly and returns "approval": null
• if the organization requires approval, it creates the agent in pending_approval and returns both agent and approval

Do not substitute rudder approval create --type hire_agent for this step unless you are doing low-level debugging. That bypasses the canonical direct-create vs pending-approval behavior.

1. Handle governance state.

If the hire response includes approval, monitor and discuss on the approval thread:

rudder approval get "" --json
cat > /tmp/rudder-approval-comment.md ](//messenger/approvals/)
- Pending agent: [](//agents/)
- Source issue: [](//issues/)

Updated prompt and adapter config per board feedback.
EOF
rudder approval comment "" --body-file /tmp/rudder-approval-comment.md --json
rudder approval resubmit "" --payload '{"title":"Revised title","agentRuntimeConfig":{"cwd":"/abs/path/to/repo","model":"o4-mini"}}' --json
rudder approval issues "" --json

When the board approves, you may be woken with RUDDER_APPROVAL_ID:

rudder approval get "$RUDDER_APPROVAL_ID" --json
rudder approval issues "$RUDDER_APPROVAL_ID" --json

For each linked issue, either:

• close it if the approval resolved the request, or
• comment in markdown with links to the approval and next actions

Quality Bar

Before sending a hire request:

• if the role needs skills, make sure they already exist in the org library or import them first using the Rudder org-skills workflow
• reuse proven config patterns from related agents where possible
• omit icon for normal hires so the server generates the default DiceBear Notionists avatar
• avoid secrets in plain text unless required by adapter behavior
• ensure the reporting line is correct and in-org
• ensure the prompt is role-specific, operationally scoped, and structured enough to become the agent's durable SOUL.md
• include mission, responsibilities, boundaries, decision principles, voice, and continuity when the role has ongoing authority
• prefer sourceIssueId or sourceIssueIds in the hire payload instead of manual approval linking
• if board requests revision, update the payload and resubmit through the approval flow
• do not report success unless rudder agent hire itself succeeded and you can cite the returned agent.id or approval.id
• creating local directories or instruction files is not evidence that an agent exists in Rudder

For canonical command syntax and examples, read: references/cli-reference.md

For low-level route shapes and underlying compatibility endpoints, read: references/api-reference.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Undertone0809
• Source: Undertone0809/rudder
• License: Apache-2.0
• Homepage: https://rudder.zeeland.studio

Install and usage instructions live in the source repository linked above.