Agents & subagents

Use the Agents surface to launch, monitor, and interact with Subagents, Claude teams, and provider-backed agent threads.

Happier now treats agent work as a first-class session surface.

In one session, you can have:

Subagents: review, plan, and delegate runs started by Happier
Claude team agents: Claude teammates created inside a Claude team
Provider-backed agent threads: monitor-only task/agent sidechains emitted by providers

The Agents surface is the place to launch them, watch their status, reopen their transcript, and take actions when the runtime supports them.

Plan run vs plan mode

Happier intentionally separates:

a plan run started from the Agents surface, and
a session mode such as plan on the current provider session

Use Start plan run when you want a separate planner or planning-style subagent working in parallel.

Use the session Mode picker when you want the current session itself to switch into a provider-supported mode.

These are different workflows:

a plan run creates a separate agent row in the Agents surface
a session mode changes how the current session behaves

If a provider does not support session modes, Happier simply does not offer the mode control for that session.

Where to find agents in a session

There are two primary entry points:

Session header
- Use the Agents button in the session header to open the Agents surface quickly.
- The badge shows how many agents are currently active in the session.
Right sidebar → Agents
- The right sidebar has a dedicated Agents tab.
- This is the main control surface for agent launch, monitoring, and quick actions.

The session transcript still keeps the original tool cards where an agent was launched. The Agents tab does not replace that history. It gives you a live roster so you do not need to scroll back to find a running agent.

What appears in the Agents tab

The tab is organized into:

LAUNCH: actions for starting new agent work from the current session
ACTIVE: agents that are currently running
RECENT: agents that finished, failed, or were terminated

Each row shows the most important facts for that agent:

name
status
type
provider
intent, when relevant
latest activity preview, when Happier has loaded that transcript

Clicking a row opens the agent in the details pane first, so you can inspect the transcript without leaving the session.

The same agent/run should also stay reachable from its original transcript entry, so you can open details either from the roster or from the place where it first appeared in the conversation.

Agent types in Happier

Subagents

These are subagents started by Happier itself.

Current presets are:

Review
Plan
Delegate

Use them when you want Happier to create a bounded or long-lived subagent that works inside the current session’s context.

What each one is for:

Review: parallel review of the current worktree or change set
Plan: separate planner-style reasoning without changing the current session mode
Delegate: a helper run that you can steer while it executes work in parallel

User-facing behavior:

launch from the Launch Subagents card
reopen from the Agents list
send steering/follow-up messages while the run is still running
stop/cancel the run from the Agents list or details view
open advanced run details when you need lower-level execution-run metadata

Claude team agents

In Claude sessions, Happier can also surface Claude’s team-style subagents as first-class agents.

You can:

create a Claude team
launch a new teammate
launch additional teammates into an existing team
message a teammate directly
broadcast to the team
terminate a teammate
delete a whole team

Claude team agents appear in the same Agents roster as Subagents, but their controls are Claude-specific.

Provider-backed task sidechains

Some providers emit task/agent sidechains that Happier can normalize into agent rows.

These rows are mainly for:

monitoring
reopening the transcript
understanding what a provider-created side thread is doing

They are monitor-only unless the provider/runtime exposes a supported control plane.

That means they may not offer send/stop/delete actions even though they still show up as first-class rows in the Agents surface.

Where subagent transcripts appear

For supported agents and runs, Happier keeps transcript access available from both places:

the Agents surface,
the original tool entry in the main session transcript.

This applies to:

Happier-managed execution runs such as review / plan / delegate,
Claude teammates surfaced from Claude Agent Teams,
provider-backed task sidechains that Happier can normalize into agent rows.

If a row is controllable, opening either entry point should land you on the same underlying transcript/details surface.

Launching new agent work

Launch Subagents

Use Launch Subagents when you want Happier to start a review, plan, or delegate run.

The launcher supports:

preset intents for quick starts
backend selection
permission mode selection
instructions
an Advanced… path when you need the full launcher/details flow

Backend selection here uses the same backend catalog as the rest of Happier. That means the launcher can target:

built-in providers such as Claude, Codex, Gemini, or Kiro
configured Custom ACP backends that you added in ACP Backends

If the session already has agents, the launch section stays collapsed by default so the roster remains visible first.

Launch Claude teams and teammates

In Claude sessions, the sidebar also shows Claude-specific launch actions.

Use them to:

Create team
Launch teammate

When a Claude team already exists in the session, the team group in the Agents roster also exposes a direct Launch teammate action for that existing team.

This avoids forcing you back through a generic global launcher when you already know which team you want to extend.

Backend-aware launch defaults

Because launchers now operate on real backend choices rather than only built-in provider ids, backend-aware defaults can follow the selected target more precisely.

That includes things like:

compatible profiles
default permission mode
default model or mode where supported

This is especially useful when you use multiple configured ACP backends that represent different agent roles on top of the same underlying CLI.

Opening an agent

When you click an agent row in the Agents list:

Happier opens the details pane first
the transcript is shown using the same transcript rendering stack used elsewhere in the app
the top of the view shows a compact summary card with agent metadata

This is important for consistency:

tool-backed agent transcripts use the same transcript components as normal focused tool details
Subagent transcripts stream and update through the same session data flow
full-page open is still available when you want more space

Sending messages to agents

Subagents

While a Subagent is running, you can send follow-up instructions from:

the main composer, when the agent is selected as the recipient
the agent details pane
the full details route

Depending on the run type, Happier uses the appropriate execution-run delivery mode under the hood.

Running execution runs can also appear as recipients in the main session composer, so you do not always need to open the details pane first.

See:

Participant routing & live run messaging

Claude teammates

For Claude team agents, Happier supports routed messages through the session composer.

The recipient chip lets you send to:

Lead
a specific teammate
Broadcast

Important current limitation:

routed Claude teammate messages do not support attachments
routed Claude teammate messages do not support review comment drafts

Use a normal lead message when you need those.

Regular non-routed session messages still support Happier attachments as usual.

See:

When agents disappear from the roster

The Agents roster follows the live control state, not just historical transcript entries.

Examples:

a running execution run appears while it can still be controlled,
after stop/cancel or completion, it moves out of the active recipient lifecycle,
a Claude teammate that has been terminated is removed from the active recipient list.

Historical transcript entries remain available, but the recipient picker only shows targets that are still valid to message.

Monitor-only provider sidechains

If the row is a provider-backed monitor-only agent, Happier shows the transcript but does not expose send controls.

This is intentional: the row exists for visibility, not for fake control that the provider/runtime cannot actually honor.

Agent lifecycle in the UI

Active vs Recent

ACTIVE shows work that is currently running
RECENT shows completed, failed, cancelled, or terminated agents

This makes long sessions much easier to reason about because you can separate “what still needs attention” from “what already finished”.

Claude teammate termination

When a Claude teammate is terminated:

it is removed from active/sendable targets
it remains visible in RECENT for history and traceability

This avoids the confusing state where an agent looks available even though Claude has already shut it down.

Running transcript visibility

You do not need to scroll back to the launch point to monitor a running agent.

The Agents tab is the canonical live roster for:

what is running now
what recently finished
how to reopen the relevant transcript

Direct sessions, inactive sessions, and resumable sessions

Agent control depends on what kind of session you are in.

Synced active sessions

These support the full agent control surface.

Inactive but resumable sessions

If the session is inactive but resumable and the machine is reachable:

Happier can still surface the Happier-subagent launcher
starting a Subagent resumes the session first, then launches the subagent

That keeps the UI honest: if Happier offers the launcher, the start flow should work.

Linked direct sessions

Linked direct sessions are different.

If the direct session is only being observed and the local runner is not active yet:

Happier treats it as observe-only for Subagents
no Happier-subagent launch
no execution-run send/stop controls
no execution-run recipient targeting

Once the direct session is locally controlled again, those controls become available.

Recommended workflows

I want a quick parallel review from the current session

Open Agents → Launch Subagents → Review.

Use this when you want a bounded review pass without losing the main session context.

I want a long-running helper that I can steer

Start a delegate-style Subagent, then keep using the row or details pane to steer it while it runs.

I want Claude teammates working in parallel

In a Claude session:

create a team
launch one or more teammates
send teammate-specific messages or broadcast from the main composer

I only want to monitor provider-created side work

Use the Agents tab as a read-only roster for provider-backed task sidechains.

This is useful when the provider emits side threads but does not expose a supported control surface through Happier.

Mental model

The simplest way to reason about the feature is:

the transcript keeps the causal history of how an agent was launched
the Agents tab is the live roster of what exists now
the details pane is the fastest way to reopen and interact with that agent

If you keep those three roles separate, the UI becomes predictable even in long sessions with many agents.

Agents & subagents

On this page