Agent Subagents UI

Add, extend, fix, or document the Subagents panel inside the OpenClaw Control UI Agents page. Use when implementing an Agents-page Subagents tab, wiring edit...

Security Scan

The skill is an instruction-only UI development helper that is consistent with its stated purpose and does not request credentials, installs, or unrelated system access.

Purpose & Capability

The name/description describe adding a Subagents panel to the Agents UI; the included SKILL.md and reference source files are all UI components, tests, and implementation notes that match that goal. Nothing requested (no env vars, no binaries, no config paths) is out of scope for a UI change.

Instruction Scope

The runtime instructions are focused on UI wiring, form fields, and config keys (agents.list[i].subagents.*, agents.defaults.subagents.*). The guidance references only repository files and config shapes relevant to the feature and does not instruct the agent to read unrelated files, environment secrets, or contact external endpoints.

Install Mechanism

This is instruction-only (no install spec or code executed by the skill). The validation steps call pnpm and vitest (pnpm --dir ui exec vitest run ... and pnpm --dir ui build) but the skill does not declare required binaries — the commands assume a development toolchain is present. That's expected for a dev helper but worth noting: tests/build will execute project scripts on the host.

Credentials

The skill requests no environment variables, credentials, or config paths. The UI code referenced calls the application's client.request methods to interact with the app backend (e.g., agents.create, agents.files.set) which is consistent with a Control UI implementation and not an unexplained credential request.

Persistence & Privilege

The skill does not request always:true and is user-invocable only. Nothing in the instructions indicates it would modify other skills or system-wide agent settings beyond wiring UI state and calling the application's normal API surfaces.

Assessment

This skill is a coherent, instruction-only developer helper for adding a Subagents panel to the Control UI. It does not require credentials and includes reference code and test instructions. Before using it: (1) run it in a developer sandbox (not a production server); (2) note that validation steps call pnpm/vitest and build scripts — ensure you have pnpm and the project toolchain installed and review package.json scripts to avoid running unexpected install/build hooks; (3) because the skill's UI code will call the app's client.request methods at runtime, only use it against a trusted OpenClaw backend; and (4) the skill contains no hidden network endpoints or secret-exfiltration instructions, but always review proposed file changes before applying them.

Build and maintain the Subagents tab inside the OpenClaw Agents page.

Scope

Use this skill for:

• adding or refining the Subagents tab under Agents
• exposing editable subagent model settings in Control UI
• wiring subagent settings into the existing config form/save path
• keeping the UI honest about what is configurable per-agent vs shared defaults
• documenting runtime precedence for subagent model selection

Core design

Keep the feature aligned with the real runtime behavior.

Subagent model precedence is:

1. agent.subagents.model
2. agents.defaults.subagents.model
3. target agent primary model
4. runtime fallback default

Do not invent backend schema that does not exist. If the current schema only supports a single subagents.model selection and shared concurrency limits, surface that clearly in the UI.

UI pattern

Add Subagents as an Agents-page panel beside:

• Overview
• Files
• Tools
• Skills
• Channels
• Cron

Recommended panel layout:

• Agent overrides
  • subagent model
  • subagent fallback models
  • subagent thinking level
• Shared subagent defaults
  • default subagent model
  • default fallback models
  • default thinking level
  • maxConcurrent
  • maxSpawnDepth
  • maxChildrenPerAgent
• include a short precedence note explaining which value wins

Config wiring rules

Prefer the existing config editing path. Do not create a disconnected store.

Edit through the same config-form/save flow already used by the Agents UI:

• agents.list[i].subagents.model
• agents.list[i].subagents.thinking
• agents.defaults.subagents.model
• agents.defaults.subagents.thinking
• agents.defaults.subagents.maxConcurrent
• agents.defaults.subagents.maxSpawnDepth
• agents.defaults.subagents.maxChildrenPerAgent

If fallback models are represented as part of the existing model object shape, reuse that shape instead of inventing a new field.

Implementation checklist

1. Add/update ui/src/ui/views/agents-panel-subagents.ts
2. Wire "subagents" through ui/src/ui/views/agents.ts
3. Add required state/type wiring in ui/src/ui/app.ts and ui/src/ui/app-view-state.ts
4. Wire change handlers in ui/src/ui/app-render.ts
5. Add/update focused UI tests
6. Build the UI and run the targeted test file

Validation

Prefer these checks after changes:

cd ~/openclaw
pnpm --dir ui exec vitest run --config vitest.config.ts src/ui/views/agents.test.ts
pnpm --dir ui build

If repo-wide typecheck is already noisy, do not claim unrelated errors belong to this feature.

References

Read these before editing:

• references/file-map.md
• references/runtime-precedence.md

For the concrete implementation that originally shipped this feature, read:

• references/implementation-notes.txt
• references/agents-panel-subagents.ts.txt
• references/agents.ts.txt
• references/app.ts.txt
• references/app-view-state.ts.txt
• references/app-render.subagents.txt
• references/agents.test.ts.txt

Use those files as the exact reference implementation when recreating or porting the feature to another OpenClaw tree.