Agent Cards

rensei agent includes a platform-specific extension for working with AgentCards: agent cards, to browse the catalog. It complements the OSS rensei agent surface (documented at donmai.dev/docs/cli/agent).

There is no rensei agent run-card command. An earlier version existed but was removed (2026-07-27): it resolved the card and printed a "dispatched" confirmation without ever actually POSTing a dispatch request - a silent success lie. There is currently no CLI flag to bind a specific AgentCard to a dispatch. Use rensei dispatch (below) to run work against a project; it resolves the agent/workflow to use from the project's subscription, not from a card id you pass on the command line.

Commands

`agent cards list`

List AgentCards visible to the active project. Cards are shown from all scopes: system, org-global, and project-scoped.

rensei agent cards list [--scope <tier>] [--work-type <type>] [--json]

Flag	Description
`--scope`	Filter by scope tier: `system`, `org`, `project`, or `workflow`
`--work-type`	Filter by workType: `research`, `backlog-writing`, `development`, `coordination`, `qa`, `acceptance`, or `other`
`--json`	Emit the full card catalog as a JSON array

rensei agent cards list
rensei agent cards list --json
rensei agent cards list --scope org
rensei agent cards list --work-type development

Example output:

ID                   NAME                       WORK TYPE     SCOPE    TRUST
card_abc123          TypeScript Backend          development   org      verified
card_def456          Security Reviewer           code-review   org      verified
card_ghi789          Fintech QA                  qa            project  verified
card_jkl012          BFSI Research Assistant     research      project  community
card_mno345          Personal Dev Config         development   project  unverified

`agent cards show`

Show full detail for a single AgentCard.

rensei agent cards show <id> [--json]

The <id> argument accepts the platform-generated ID (ag_…) or the stable metadata ID (e.g. research-agent).

rensei agent cards show card_ghi789
rensei agent cards show research-agent --json

Scopes from highest to lowest priority in resolution order: user > project > org > system.

JSON output (one element):

{
  "id": "card_ghi789",
  "name": "Fintech QA",
  "scope": "project",
  "projectId": "proj_xyz",
  "version": "0.9.1",
  "workTypes": ["qa"],
  "description": "End-to-end QA agent with Playwright and Jest toolchains",
  "authModes": ["host-session", "metered"],
  "capabilities": ["playwright", "jest", "typescript"],
  "extends": "card_abc123",
  "updatedAt": "2026-05-28T09:30:00Z"
}

Dispatching work

To dispatch a session, use rensei dispatch - the real /api/cli dispatch path:

rensei dispatch --project <slug> --goal "<free-text goal>" --work-type <research|implement|review> --wait

dispatch does not take an AgentCard id directly. Instead, the platform resolves cards by a most-specific-wins cascade (workflow → project → org → system) filtered by workType - author or promote the card at the scope you want it to win at (Factory → Agent Cards in the project, or promote to org scope; see Agent cards overview), and it is picked up automatically on the next matching dispatch. Monitor a run with rensei agent status <session-id>, rensei session stream <session-id>, or the session detail UI.

AgentCard scope model

AgentCards follow a 4-scope resolution model. When a card uses extends, the platform merges the parent definition with the child's overrides.

user (personal overrides)
  └── project (project-specific configuration)
        └── org (org defaults and shared configurations)
              └── marketplace (published community cards)

Cards at a narrower scope can override any field from a broader-scope card. The capabilities, authModes, and tool allowlist are merged (union), while systemPrompt, modelProfile, and maxCost are overridden.

Authoring AgentCards

AgentCards are defined in YAML and managed via the platform API or the web UI. Here is a minimal example:

# agent-card.yaml
name: Fintech QA
description: End-to-end QA agent for payment processing APIs
version: 0.9.1
scope: project
workTypes:
  - qa
authModes:
  - host-session
  - metered
extends: card_abc123   # inherits TypeScript Backend base

systemPrompt: |
  You are a QA specialist for a PCI-DSS payment processing service.
  Always test for edge cases in monetary calculations.
  Never write tests that store real card numbers.

modelProfile: claude-sonnet-4-5-thinking
capabilities:
  - playwright
  - jest
  - typescript

toolAllowlist:
  - bash
  - computer
  - edit_file
  - browser

maxCost:
  perSession: 5.00
  currency: USD

For the full AgentCard schema, partial inheritance, and versioning model, see Agent cards overview and Agent card partials.

Difference from `rensei agent` (OSS)

The base rensei agent command group (list, status, stop, chat, reconnect, run) is inherited from the OSS donmai CLI. agent cards is the platform-specific addition:

Command	Origin	Description
`rensei agent list`	OSS	List running sessions
`rensei agent status`	OSS	Get session status
`rensei agent stop`	OSS	Stop a running session
`rensei agent chat`	OSS	Interactive chat with a session
`rensei agent reconnect`	OSS	Reconnect to a running session
`rensei agent run`	OSS	Run an agent with inline config
`rensei agent cards list`	Platform	List AgentCard catalog
`rensei agent cards show`	Platform	Show full detail for a card

For the OSS command reference, see donmai.dev/docs/cli/agent.

Agent cards overview - full AgentCard schema, scope cascade, YAML authoring
Agent card partials - Handlebars partials, auto-attach, version pinning
Agent Registry Providers (ARP) - 11+ providers, capability flags
Marketplace browse - discover community-published cards
Cost and caps - maxCost, cost_events, per-(authMode × provider) cap matrix

On this page