AI agent integration
A developer often runs several AI agent sessions at once: one writing the step methods a new spec file needs, another looking into why a scenario started failing, a third opened for one question and never closed. Each session is busy with a different task, and only the developer's memory records which session is working on which spec file.
The plugin links each session to a spec file instead. A button on the spec file's toolbar opens a terminal session with an AI coding agent, and the agent's first instruction is to read that spec file. The plugin records the session against the spec file, so a later click on the button reopens that spec file's session instead of starting a second one.
Claude Code and GitHub Copilot CLI are both supported; choose one under Select Harness in the
button's dropdown. The agent's own command-line program has to be on the PATH. To show the button,
switch on Show a "Launch AI agent" button on the toolbar, under Settings → Tools → SpecBinder →
Agent integration. The button also needs the bundled Terminal plugin enabled.
Starting and resuming a session
A click with no session tab open always starts a new session; the button never resumes an older session on its own. To resume one, choose Resume a Past Session… in the dropdown: the list shows this spec file's previous sessions for the selected agent, newest first. New Parallel Session opens another session beside one already running.
The button's icon shows which of the two a click will do: grey while no session is running, and purple once a session is running and a click would reopen that session's tab.
The session starts with one instruction already sent: read this spec file, then wait. So the agent has read the spec file before the developer types anything, and the agent does nothing more until asked.
Read this SpecBinder spec file and wait for my instructions: src/test/java/…/ShoppingCart.specb
The path is relative to the directory the session runs in (described below), and the terminal tab is named after the spec file. Once the session exits, the terminal tab shows the whole command, including the session id the plugin assigns in order to find this session again.
The session's working directory is the root of the module the spec file belongs to, and not the folder that holds the spec file. So the agent starts with access to the whole module, whichever spec file the session was launched from.
The agent state on the spec file icon
With Badge spec file icons with the linked agent's working state switched on, under Settings → Tools → SpecBinder → Agent integration, a spec file with a linked session shows a small diamond over the spec file's icon in the Project view:
| Badge | Meaning | |
|---|---|---|
| Pulsing terracotta | The agent is working | |
| Green | The agent needs an answer or permission | |
| Slate grey | The agent has finished replying | |
| None | No live session |
The badge works with Claude Code only at present, because only Claude Code reports a working state back to the plugin. Copilot CLI sessions still work but show no badge.
The plugin sets up this reporting separately for each session, and the developer's own Claude Code configuration is never edited.
Following the spec's terminal tab
Auto-Select This Spec's Terminal Tab, in the button's dropdown, makes the Terminal tool window follow the editor: open another spec file, or click into that spec file's editor tab, and the Terminal tool window switches to that spec file's agent session tab.