Skip to content

Node-based terminal manager for AI coding agents — tmux-backed terminals and parallel agent sessions as draggable nodes on an infinite pan/zoom canvas. macOS, Linux, and a browser Server Edition.

fix(chat): keep the live tool run open while the agent waits on a prompt

reviewdade2efdb7a4
diff --git a/src/renderer/lib/chatToolGroups.test.ts b/src/renderer/lib/chatToolGroups.test.ts
index b5f392e8c..1081c1cf2 100644
--- a/src/renderer/lib/chatToolGroups.test.ts
+++ b/src/renderer/lib/chatToolGroups.test.ts
@@ -108,9 +108,19 @@ describe('groupToolCalls', () => {
const layout = groupToolCalls(messages, { foldTail: false })
expect(runs(messages)).toEqual([['1:0', '2:0'], ['4:0', '5:0']]) // default folds both
expect([...layout.groupAt.keys()]).toEqual(['1:0']) // tail run left alone
+ expect(layout.absorbed.has('2:0')).toBe(true) // the earlier run still folds
expect(layout.absorbed.has('5:0')).toBe(false)
})
+ it('foldTail:false still folds every run when the thread ends in something else', () => {
+ const messages = [user('go'), call('Bash'), call('Read'), say('mid'), call('Bash'), call('Edit'), say('done')]
+
+ const layout = groupToolCalls(messages, { foldTail: false })
+
+ expect([...layout.groupAt.keys()]).toEqual(['1:0', '4:0'])
+ expect([...layout.absorbed]).toEqual(['2:0', '5:0'])
+ })
+
it('memberKey uses the transcript line key, so a prepended page does not re-key a member', () => {
expect(memberKey([call('Bash', 120)], { message: 0, part: 0 })).toBe('k120:0')
expect(memberKey([call('Bash')], { message: 0, part: 0 })).toBe('i0:0')
diff --git a/src/renderer/nodes/ChatPanel.thread.test.tsx b/src/renderer/nodes/ChatPanel.thread.test.tsx
index 508606b23..89d2589e6 100644
--- a/src/renderer/nodes/ChatPanel.thread.test.tsx
+++ b/src/renderer/nodes/ChatPanel.thread.test.tsx
@@ -157,4 +157,17 @@ describe('⌘M thread look', () => {
expect(host.querySelector('.term-chat__tool-group')).toBeNull()
expect(host.querySelectorAll('.term-chat__tool')).toHaveLength(2)
})
+
+ it('while the agent waits on a prompt mid-turn, the trailing run of tool calls is not folded', async () => {
+ // `waiting` (a PermissionRequest) refuses with `dialog`: the turn is still live, so its calls stay
+ // chips rather than folding closed on every prompt. The afterEach restores `done`.
+ useAgentStatus.setState((s) => ({ byId: { ...s.byId, [NODE]: { state: 'waiting' } as (typeof s.byId)[string] } }))
+ await mount([
+ { role: 'user', parts: [{ kind: 'text', text: 'q' }], key: 0 },
+ { role: 'assistant', parts: [{ kind: 'tool', name: 'Bash', arg: 'a' }], key: 10 },
+ { role: 'assistant', parts: [{ kind: 'tool', name: 'Read', arg: 'b' }], key: 20 }
+ ])
+ expect(host.querySelector('.term-chat__tool-group')).toBeNull()
+ expect(host.querySelectorAll('.term-chat__tool')).toHaveLength(2)
+ })
})
diff --git a/src/renderer/nodes/ChatPanel.tsx b/src/renderer/nodes/ChatPanel.tsx
index e81bed9ae..79f716a9c 100644
--- a/src/renderer/nodes/ChatPanel.tsx
+++ b/src/renderer/nodes/ChatPanel.tsx
@@ -969,10 +969,11 @@ export function ChatPanel({
// (a per-row timer would be a timer per turn for a label that changes once a minute).
const turnEnds = useMemo(() => assistantTurnEnds(messages), [messages])
// Runs of consecutive tool calls, folded into one closed block each (lib/chatToolGroups.ts) —
- // except, while the agent works, the run at the end of the thread: that is the live turn, and its
- // calls stay open chips until something else ends the run.
- const foldTail = refusal !== 'working'
- const toolGroups = useMemo(() => groupToolCalls(messages, { foldTail }), [messages, foldTail])
+ // except, while the turn is live, the run at the end of the thread: its calls stay open chips until
+ // something else ends the run. Live = working, or waiting on a prompt (`dialog`: a permission
+ // request or question lands MID-turn, and folding on it would close the run on every prompt).
+ const liveTurn = refusal === 'working' || refusal === 'dialog'
+ const toolGroups = useMemo(() => groupToolCalls(messages, { foldTail: !liveTurn }), [messages, liveTurn])
// The last key IS the latest turn end: `assistantTurnEnds` inserts in thread order.
const latestTurnEnd = useMemo(() => {
let last = -1
diff --git a/src/renderer/styles.chat-view.test.ts b/src/renderer/styles.chat-view.test.ts
index 1f4f44114..6cd12f571 100644
--- a/src/renderer/styles.chat-view.test.ts
+++ b/src/renderer/styles.chat-view.test.ts
@@ -78,7 +78,9 @@ describe('⌘M thread: claude.ai look', () => {
})
it('a folded message collapses to its action row', () => {
- expect(ruleBody('.term-chat__msg--folded')).toMatch(/padding:\s*0;/)
+ const folded = ruleBody('.term-chat__msg--folded')
+ expect(folded).toMatch(/padding:\s*0;/)
+ expect(folded).toMatch(/min-height:\s*0;/)
})
})
diff --git a/src/shared/chat-fixtures/README.md b/src/shared/chat-fixtures/README.md
index 7c1a89508..e81281a89 100644
--- a/src/shared/chat-fixtures/README.md
+++ b/src/shared/chat-fixtures/README.md
@@ -72,7 +72,7 @@ the production paging, not a second implementation of it.
| fixture | pins |
|---|---|
| `plain-turns` | User/assistant text, markdown (headings, lists, a blockquote, code fences), string and array user content, two text blocks in one message, and non-ASCII text. Metadata-only records (`file-history-snapshot`, `system`) yield no message. |
-| `tools-cross-page` | A `tool_use` that lands in the second (older) page while its `tool_result` lands in the tail. The tail carries it in `unmatchedResults`, and the older page carries the tool part with its `id` so the port can attach it. One tool call is matched inside the tail. It also shows the `summarizeResult` rules: three lines joined, capped at 500. The desktop renderer folds a run of two or more consecutive agent tool calls into one block, and the parser output is unchanged, so a port that renders a thread should fold the same runs. |
+| `tools-cross-page` | A `tool_use` that lands in the second (older) page while its `tool_result` lands in the tail. The tail carries it in `unmatchedResults`, and the older page carries the tool part with its `id` so the port can attach it. One tool call is matched inside the tail. It also shows the `summarizeResult` rules: three lines joined, capped at 500. The desktop renderer folds a run of two or more consecutive agent tool calls into one block, except the run at the END of the thread while the agent is working or waiting on a prompt (the live turn), and the parser output is unchanged, so a port that renders a thread should fold the same runs. |
| `plan-mode` | An `ExitPlanMode` tool call with `input:{plan}`. The tool part carries the full plan as `body` and the approval text as `result`. |
| `ask-question` | `AskUserQuestion` with a single-select and a multi-select question: the rendered `body`, the parsed `questions` (the same reader the answer controls match on) and each result. |
| `utf8-edge` | Multi-byte characters on the window boundaries. The tail window (EOF − 262144) opens **inside** a 4-byte emoji of one line, which must be dropped as the partial line and never decoded torn. The second page (524288) starts **exactly** on the first byte of a multi-byte line, which is kept only because of the lookbehind `\n`. The straddled line reappears whole in that page. The test asserts both placements. |