Skip to main content

agent-page

The agent's configuration surfaces: Settings › Agent pages (Instructions, Memory, Skills, Files, Sub-agents) and the full-page `AgentTabsPage` (resources, Snapshots, Agent integrations, Automations, MCP). Use when adding an agent resource group, surfacing context transparency, MCP servers, A2A agents, recurring jobs, or external-client connect flows in the UI.

Source facts

Repository
BuilderIO/agent-native
Last source activity
September 29, 2026 at 12:05
Detected SKILL.md language
English
Stars
7,065
Forks
640

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
agent-page
description
The agent's configuration surfaces: Settings › Agent pages (Instructions, Memory, Skills, Files, Sub-agents) and the full-page `AgentTabsPage` (resources, Snapshots, Agent integrations, Automations, MCP). Use when adding an agent resource group, surfacing context transparency, MCP servers, A2A agents, recurring jobs, or external-client connect flows in the UI.
scope
dev
metadata
{"internal":true}
# Agent Page `AgentTabsPage` (from `@agent-native/core/client`, source `packages/core/src/client/agent-page/`) is the canonical full-page surface for everything that can influence the agent. Design principles: - The page answers "what can influence this agent, and why" — it is not a generic admin console. - Capability and access stay separate: **Connections** is what this agent can reach; **Access** is who can reach this agent. - It is a thin shell that re-hosts existing components (ResourcesPanel, MCP hooks, AgentsSection, context X-Ray, jobs actions) — do not re-implement those surfaces inside it. - Context must be inspectable, attributable, and governable — provenance and governance tiers, not just a token meter. ## Tabs `defaultTab` is `files`. Resource tabs re-host `ResourcesPanel` views; the rest re-host existing components. | Tab | Contents | | --- | --- | | `files`, `instructions`, `agents`, `memory`, `skills`, `learnings`, `remote-agents` | `ResourcesPanel` views (group "resources"). | | `snapshots` | Scope preview, token budget, provenance-grouped system sections (governance tiers), and the latest live-thread snapshot. Backed by `context-preview-get` / `context-manifest-get` (see `context-xray`). Old `#context` links land here. | | `connections` | "Agent integrations": MCP server management (both scopes, admin-gated org writes). | | `jobs` | **Automations**: personal and organization Scheduled/Event tasks with pause/resume/delete. The `jobs` hash is stable for compatibility (see `automations` and `recurring-jobs`). | | `settings` | The agent settings panel (model, keys, limits, voice). | | `access` | "MCP": copyable MCP URL and A2A agent-card URL, per-client connect steps from `packages/core/src/shared/mcp-connect-content.ts` (shared with `/mcp/connect`; edit the shared module, never fork copy), static-token fallback link. | ## Settings Resource Pages Settings › Agent › Instructions, Memory, Skills, and Files (`packages/core/src/client/settings/shell/pages/`) render the same `ResourcesPanel` with `settingsGroups`: Personal, {Org name}, and From Dispatch groups (Memory swaps From Dispatch for Learnings). Add a group or an add action there, not a second panel. Row read-only state comes from the scope: organization rows for members and Dispatch rows for everyone; the server still enforces it. Resource trees refetch on `action` change events, which is how an agent's `save-memory` or `resources` write appears without a reload. Settings › Agent › Sub-agents (`pages/sub-agents.tsx`) is the one place that lists every agent the main agent can hand work to, each once: - **{Org} apps**: first-party apps, from their seeded `remote-agents/<id>.json` manifests plus workspace apps only discovery knows, each "Reachable · {URL}" from the batched `/_agent-native/agents/probe`. - **External agents**: every other `remote-agents/` manifest. The header's **Connect agent** (owners and admins) opens the directory dialog: Foundry, Gemini Enterprise, Anthropic Managed Agents, and any A2A agent by URL. - **Custom agents**: `agents/*.md` profiles through `ResourcesPanel` `settingsGroups`. Anyone can add a personal one. Build on `useRemoteAgents()` and the `AgentAddForm` / `AgentEditForm` exports of `AgentsSection.tsx`, not a second fetch. `?connect=<a2a|anthropic-managed-agents|manual>` and the peer register-back `f_agent*` params still open the connect form, and legacy ids (`agent:agents`, `agent:directory`, `agent:resources:agents`, `agent:resources:remote-agents`, section `a2a`) redirect to this page. Hidden and removed first-party ids (`shared/first-party-agents.ts`) stay unlisted, the same rule discovery applies. ## Mounting In A Template First-party templates don't mount `AgentTabsPage`. Their `/agent` route redirects into Settings with `buildLegacyAgentSettingsRoute(hash, search)`, so `/agent#files`, `#jobs`, `#connections`, and `#access` land on the matching Settings page, and they pass `agentPageHref="/settings/agent"` to `AgentSidebar` (Settings › Agent › Model). New agent-configuration UI belongs on a Settings page, not a new tab here. For an app that wants the whole surface on one page: 1. Mount `AgentTabsPage` from `@agent-native/core/client/agent-chat` on its own route. CSR is fine; keep the app shell in `root.tsx` so navigation does not remount it (`client-side-routing` skill). 2. Pass that route as `agentPageHref` to `AgentSidebar`. 3. App-specific additions go in `extraTabs` (same `SettingsTabItem` shape as the settings page) or `extraTabFactories`; hide built-ins only with `hiddenTabs` when the app genuinely lacks the underlying capability. ## Scope The page currently passes personal `scope` (and `canManageOrg`) to tabs via `AgentPageTabProps`; it does not expose a page-level Personal/Organization toggle. A tab may render its own scoped sections where the underlying actions support them. The Automations tab shows personal and organization sections for both Scheduled and Event triggers. Organization event automations remain creator-run: administrators can manage them, but cannot retarget the creator's identity.
View on GitHub