Power Strip can run deployed agents and automations directly from the desktop surface. It can also attach agent hooks to supported actions and display execution feedback near the command that started the work.
Agent Commands
An Agent command slot is bound to an agent deployment. Power Strip supports multiple Agent slots, each with an independent binding.
The sidecar shows:
- Read-only derived input
- AI Reference Rules only when the Tool contains an AI skill
- Save New Files To only when the Tool can use a configurable destination
- Tool Configuration when the Tool has configurable skills
- The bound deployment and Run action
Selecting an agent through Tool Browser creates or associates a slot-specific deployment, then opens its configuration sidecar. This keeps one Agent button's settings independent from another.
An Agent can run in two ways:
- Run with the derived or saved input shown in the sidecar, from the footer
- Run with dropped context, by dropping files, folders, or text on the command button; the drop takes precedence
The output-folder control appears only for Tools that can create files in a configurable destination. Other Tools show a read-only explanation. AI reference choices remain visible and saved for AI-backed Tools, but the current executor does not yet read their contents.
Folder Tools
A Folder command resolves the shared Tool placements attached to its bound project folder. Each placement can run with that folder as input. Manage Folder Tools... opens the same manager used by Project Navigator and linked Mind Map nodes, including schedule and matching-file-created triggers, pause and resume, and confirmed removal.
Folder Tools are shared folder deployments, not Power Strip Agent slots or command hooks. Adding, reordering, disabling, triggering, or removing one is reflected everywhere that folder is shown.
Automation Commands
An Automation slot binds to a deployed automation. Multiple Automation slots are supported. Change Automation opens Tool Browser in Automations mode.
Selecting an automation creates a slot-owned placement and clones its node settings. Two Power Strip buttons can therefore start from the same library automation while keeping different script paths, instructions, and other node configuration. Replacing or removing the slot also removes its private placement.
The sidecar summarizes:
- Tool order
- Tool count
- Execution mode
- Deployment date
An automation can run without input from the sidecar, or with dropped files, folders, or text supplied to the command button.
Browser Agent and Browser Automation
Browser Agent requests the current page from the Chrome extension and passes that captured context to its bound deployment. Browser Automation follows the same pattern for a deployed automation.
These commands require the extension connection and an active page that can supply the requested content.
Known gap: the Browser Agent sidecar conditionally displays and saves AI reference and output-folder configuration, but those settings are not loaded during browser execution. They do not affect that run.
Agent Hooks
A supported command slot can have a before hook, an after hook, or both. The configuration attaches to that slot, not globally to every instance of the command.
Before hooks
A before hook runs before the main action and receives the context available at that point.
A before hook does not transform, rewrite, cancel, or replace the main action. It is sequencing, not an input-substitution pipeline.
After hooks
An after hook runs after the main action has produced its outcome. For example, a capture after hook can receive the saved document created by the import.
Currently wired triggers
Hooks are wired for these paths:
- Quick Note — after content is created
- Save Clipboard — click
- Drop File — drop
- Folder — drop, and content created by footer capture actions
- Agent — drop and agent run
- Automation — drop and automation run
- Capture Page — browser import completed
- Capture Selection — browser import completed
- Browser Agent — browser agent run
- Browser Automation — browser automation run
Create Doc From Clipboard advertises a Content Created hook but does not currently invoke it. Treat that combination as unsupported.
Hook Context
Hook inputs depend on the initiating action:
- Drop hooks receive the actual dropped files, folders, or text.
- Quick Note's after hook receives the saved note.
- Browser capture after hooks receive page context plus the saved imported document.
- Agent and automation drop hooks receive the dropped inputs.
- A Folder command supplies its bound folder to hook preview, sidecar Run, and click execution.
- Agent and automation runs without a drop use host-derived or saved deployment input when present.
- After hooks receive output paths when the main action returns saved files.
- Save Clipboard click hooks receive action metadata and result information, not the raw clipboard contents.
Before and after hooks on command-button and expanded Folder-panel drops receive the placement's resolved Output Folder. Stored AI reference documents are still not loaded into hook execution.
Ordering and Failures
When both hook phases exist, the order is:
- Before hook
- Main action
- After hook
A hook is a separate agent run. It is not a hardcoded workflow or a replacement for the main command.
Success for the main action and success for hook execution are separate. An after hook is only meaningful when the action reaches the point that supplies its completion context.
Execution Activity
Every Agent run reports its own run ID, launch origin, current step, progress, elapsed time, available output, and final success or failure to Power Strip's fixed Activity surface. Concurrent runs stay separate, even when they use the same Tool.
Selecting the Activity bell opens or closes the cards without stopping work. A running card updates in place; select it to expand step, output, or error details. Matching Agent buttons still show compact status, and active work keeps the strip glow.
Successful automatic presentation remains for five seconds and failures for ten seconds, pausing while the pointer is over the cards. Manually opened history shows the 20 most recent entries. The session retains at most 100 completed entries and clears them when ThinkForge restarts.
Saved Configuration
Power Strip stores, per slot:
- Agent and deployment bindings
- Automation bindings
- Before and after hook assignments
- Reference paths and output folders
- Slot relationships for multiple instances
Boundaries
Power Strip uses ThinkForge's normal agent and automation execution. It does not have an independent planner or a second agent runtime.
The distinctions are input source and launch surface:
- Desktop Agent and Automation use derived, saved, or dropped context.
- Folder Tools use the bound project folder and its shared placements.
- Browser Agent and Automation request current-page context.
- Hooks run around specifically wired Power Strip actions.
- Execution activity is projected back onto the floating desktop surface.