ACP Skills

The ACP Skills tab is used to monitor and manage skill runs executed through the ACP backend. Unlike ACP Chat's continuous dialogue, ACP Skills is designed for one-off or periodically executed skill tasks.

Interface Overview

The ACP Skills panel is divided into the following main areas:

┌─────────────────────────────────────┐
│  Banner: Task Title / Status / Backend   │
├─────────────────────────────────────┤
│  ← Run Drawer  │  Main Content Area  │  Details → │
│               │  Transcript View            │
│  Running      │  Plan Component             │
│  └─ backend1  │  Prompt Component           │
│     ├─ run A  │  Reply Area                 │
│     └─ run B  │                             │
│  Completed    │                             │
│  └─ backend1  │                             │
│     └─ run C  │                             │
└─────────────────────────────────────┘

Banner

The Banner area displays meta-information and action buttons for the currently selected run:

• Task Title: The skill name of the run
• Status: Run status indicator (running / completed / failed / canceled, etc.)
• Backend: The ACP backend executing the run
• Action Buttons: Connect/Disconnect, Cancel Task

Run Drawer (Left)

The left drawer organizes all ACP Skill Runs in a tree structure:

Grouping

Group	Description
Running	Currently running tasks, grouped by backend
Completed	Finished tasks, grouped by backend

Each task entry displays summary information (skill ID, status, time) and has an attention indicator (LED) to mark status changes. Click any task entry to switch to the detail view of that run.

Archiving

Completed tasks can be removed from the list via the archive button (archiving only hides them in the current session and does not affect run records).

Main Content Area

Transcript View

After selecting a run, the main content area displays the full transcript of that run, including:

• Messages: Assistant and user dialogue content
• Tool Calls: Tools invoked by the AI and their results, showing tool name, input summary, and status LED
• Thinking Process: The AI's reasoning process (if available)
• Status Events: State changes during the run

The transcript supports Plain mode (messages colored by role on the left border) and Bubble mode (messages in bubble style, consecutive tool calls automatically collapsed into groups), toggled via the button in the top-right corner.

Plan Component

When a run includes a multi-step plan, the plan component displays current progress, completed steps, and pending steps, with each step having a status icon (in-progress/completed/failed).

Prompt Component

The prompt component displays different interactive prompts based on run status:

Status	Display Content
waiting_user	Prompt awaiting user reply, with context description and quick reply options
permission	Permission request prompt, with command preview and approve/reject buttons
disconnected	Reconnection prompt; click to connect
running	In-progress indicator
completed	Completion status confirmation
error	Error information and troubleshooting suggestions

Reply Area

The reply area at the bottom contains:

• Text Input Box: Enter reply content
• Mode Selection (optional): Run mode toggle
• Model Selection (optional): AI model toggle
• Reasoning Effort (optional): Reasoning effort level
• Send/Cancel Button
• Usage Meter: Circular chart showing token usage (used/limit)
• Keyboard Shortcut Hint: Keyboard shortcut for sending replies

Reply drafts are saved per request — switching runs and switching back preserves unsent content.

Details Drawer (Right)

The right drawer displays detailed information about the selected run, with the following collapsible areas:

Area	Content
Run Path	Workspace directory, result file paths
Runner Info	backends, agent, mode, model, reasoning, skill, session
Validation Info	Validation status, fix count, error details
Runtime Dependencies	List of runtime environment dependencies
Output Revision	Output revision history
Runtime Log	Log entries during the run
Result JSON	Final structured output (expandable)

Permission Handling

When a run requires Zotero write permissions or ACP tool call permissions, the prompt component displays a permission request:

• Command Preview: Shows the operation being requested
• Source Info: Who initiated the request
• Action Buttons: Approve / Reject
• Expand to view full request details

Related Configuration

The ACP Skills panel requires an ACP backend to be configured before use.