diff --git a/docs/reference/session-summary.md b/docs/reference/session-summary.md
index e4bd30d59644..22f8c912e1fe 100644
--- a/docs/reference/session-summary.md
+++ b/docs/reference/session-summary.md
@@ -49,10 +49,29 @@ The summary is a **materialized view** over the session's own provider transcrip
 the agent CLI writes — not a re-reading of the terminal. `session-summary-transcript-source.ts`
 resolves that file per agent type; the fold turns it into a structured ledger.
 
+The fold reads a transcript only for agents it can name a file for, and each agent's file is
+found a different way:
+
+- **Native-chat agents** (Claude/OpenClaude, Codex, Grok, OMP) go through
+  `native-chat/session-file-resolver.ts`, which takes the hook's `transcript_path` when it is
+  host-readable and otherwise searches the agent's own roots by session id.
+- **Pi and Prime Agent** post `session_file` on every hook, so the reported path *is* the
+  transcript; `agent-session-resume.ts` only returns their provider session at all when that
+  path exists.
+- **Copilot** reports no path, but its `session_id` is the name of its session directory:
+  `$COPILOT_HOME/session-state/<id>/events.jsonl`.
+
+Every other agent type returns null and the pane says so. Adding one means naming its parser in
+`HOOK_RESOLVED_TRANSCRIPT_AGENTS` and reaching a readable file — the AI Vault parsers are
+agent-generic (`parseAgentSessionFile` dispatches on `candidate.agent`), so nothing else changes.
+
 - **Lazy by construction.** Opening the pane starts the fold; closing it cancels in-flight
   work. Nothing runs for sessions nobody opens, so there is no background LLM spend.
 - **Cheap facts are always available.** State, timestamps and event counts come from the
-  status entry, so the header hint ("Up to date" / "N events to catch up on") needs no fold.
+  status entry, so the header hint needs no fold. The hint still respects the fold's own state:
+  a fold in flight reads "Updating…" (set when the fold starts, not when its first chunk
+  returns), and a session with no readable transcript or a failed fold says that instead of
+  claiming to be up to date.
 - **Bounded, resumable chunks.** The transcript is split deterministically into slices of at
   most 24k chars or 40 events (`session-summary-fold.ts`), extracted in parallel and merged
   deterministically. Completed extracts are cached, so a fold that is cancelled or fails
@@ -74,7 +93,12 @@ configured it follows the default agent and that agent's default model and effor
 
 A **dynamic** agent ships only a stub model list — Pi advertises `Config default` — so the
 pane asks its CLI once (`<binary> --list-models`, the same probe Native Chat uses), keeps the
-answer and replaces the stub with it. Two rules matter there:
+answer and replaces the stub with it. **Copilot** is dynamic too, but its catalog is only
+documented, not listed: the CLI has no `models` subcommand, so the probe reads
+`copilot help config`, whose `model` setting enumerates the accepted ids. That list is the CLI's
+own and therefore tracks models released after the build, which a pinned array cannot; a machine
+without the binary keeps a pinned snapshot of it (`COPILOT_FALLBACK_MODELS`) that the probe
+replaces the moment it can run. Two rules matter there:
 
 - **The pane keys on the session, not just the pane.** A new agent session can start in the
   pane the user is already focused on, and the key includes the provider session id so the
diff --git a/src/main/ai-vault/session-scanner-agent-sources.ts b/src/main/ai-vault/session-scanner-agent-sources.ts
index 4a12303982cb..90848d3b2f50 100644
--- a/src/main/ai-vault/session-scanner-agent-sources.ts
+++ b/src/main/ai-vault/session-scanner-agent-sources.ts
@@ -30,7 +30,7 @@ const CODEX_SESSIONS_DIR = join(
   'sessions'
 )
 const GEMINI_SESSIONS_DIR = join(homedir(), '.gemini', 'tmp')
-const COPILOT_SESSIONS_DIR = join(
+export const COPILOT_SESSIONS_DIR = join(
   resolveAbsoluteDirOverride(process.env.COPILOT_HOME, join(homedir(), '.copilot')),
   'session-state'
 )
diff --git a/src/main/session-summary/session-summary-service.test.ts b/src/main/session-summary/session-summary-service.test.ts
index 89cabac6e067..bb7466f1f0b5 100644
--- a/src/main/session-summary/session-summary-service.test.ts
+++ b/src/main/session-summary/session-summary-service.test.ts
@@ -112,6 +112,19 @@ describe('SessionSummaryService', () => {
     expect(updates.some((u) => u.status === 'ready')).toBe(false)
   })
 
+  // The renderer's header reads `fold.running`, and open()'s return value is the
+  // first snapshot it sees. A fold that only reported progress after its first
+  // chunk read as "Up to date" for the whole first fold.
+  it('reports the fold as in flight before the first chunk comes back', async () => {
+    const { store } = testStore()
+    const { service, setBrainReply } = harness(store, events(5))
+    setBrainReply(() => new Promise<string>(() => {}))
+
+    const opened = service.open({ paneKey: 'tab:leaf' })
+
+    expect(opened.fold.running).toBe(true)
+  })
+
   // A pane whose status row was dropped keeps a provider-session remnant
   // (`server-cleanup.ts`): no state, prompt, or timestamps, but a resumable
   // session — which is all the fold needs to summarize a finished agent.
diff --git a/src/main/session-summary/session-summary-service.ts b/src/main/session-summary/session-summary-service.ts
index 9253a1216986..50a945105a93 100644
--- a/src/main/session-summary/session-summary-service.ts
+++ b/src/main/session-summary/session-summary-service.ts
@@ -180,6 +180,12 @@ export class SessionSummaryService {
     const controller = new AbortController()
     this.folds.set(paneKey, controller)
     const runtime = this.runtimes.get(paneKey)
+    // Why: the fold is in flight from here, before the transcript read resolves.
+    // Without this the header read "Up to date" for the whole first fold, because
+    // onProgress only fires once a chunk comes back.
+    if (runtime) {
+      runtime.fold = { running: true, completedChunks: 0, totalChunks: 0 }
+    }
     try {
       const entry = this.deps.getAgentStatusEntry(paneKey)
       const identity = identityOf(entry)
diff --git a/src/main/session-summary/session-summary-transcript-source.test.ts b/src/main/session-summary/session-summary-transcript-source.test.ts
new file mode 100644
index 000000000000..593993f01167
--- /dev/null
+++ b/src/main/session-summary/session-summary-transcript-source.test.ts
@@ -0,0 +1,177 @@
+import { mkdtemp, mkdir, rm, writeFile } from 'node:fs/promises'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+import { afterEach, describe, expect, it, vi } from 'vitest'
+
+/**
+ * Which provider transcripts the fold can read.
+ *
+ * Pi and Copilot are the agents this file exists for: neither is a native-chat
+ * transcript agent, so before this the fold resolved no path and the pane read
+ * "no readable transcript" for every one of their sessions.
+ */
+
+const tempDirs: string[] = []
+
+async function tempDir(): Promise<string> {
+  const dir = await mkdtemp(join(tmpdir(), 'session-summary-source-'))
+  tempDirs.push(dir)
+  return dir
+}
+
+afterEach(async () => {
+  vi.unstubAllEnvs()
+  await Promise.all(tempDirs.splice(0).map((dir) => rm(dir, { recursive: true, force: true })))
+})
+
+/** Pi writes a `session` record then `message` records whose content is blocks. */
+function piTranscript(prompt: string, reply: string): string {
+  return [
+    JSON.stringify({
+      type: 'session',
+      id: 'pi-session-1',
+      cwd: '/tmp/project',
+      timestamp: '2026-01-01T00:00:00.000Z'
+    }),
+    JSON.stringify({
+      type: 'message',
+      id: 'm1',
+      timestamp: '2026-01-01T00:00:01.000Z',
+      message: { role: 'user', content: [{ type: 'text', text: prompt }] }
+    }),
+    JSON.stringify({
+      type: 'message',
+      id: 'm2',
+      timestamp: '2026-01-01T00:00:02.000Z',
+      message: { role: 'assistant', content: [{ type: 'text', text: reply }] }
+    })
+  ].join('\n')
+}
+
+/** Copilot writes one event object per line, payload under `data`. */
+function copilotTranscript(prompt: string, reply: string): string {
+  return [
+    JSON.stringify({
+      type: 'session.start',
+      data: { sessionId: 'copilot-session-1', context: { cwd: '/tmp/project' } },
+      timestamp: '2026-01-01T00:00:00.000Z'
+    }),
+    JSON.stringify({
+      type: 'user.message',
+      data: { content: prompt },
+      timestamp: '2026-01-01T00:00:01.000Z'
+    }),
+    JSON.stringify({
+      type: 'assistant.message',
+      data: { content: reply },
+      timestamp: '2026-01-01T00:00:02.000Z'
+    })
+  ].join('\n')
+}
+
+/**
+ * Copilot's sessions root is a module-level constant read from the environment,
+ * so the stub has to land before the module graph is evaluated.
+ */
+async function loadSource() {
+  vi.resetModules()
+  return await import('./session-summary-transcript-source.js')
+}
+
+const signal = (): AbortSignal => new AbortController().signal
+
+describe('readSessionSummaryTranscriptEvents', () => {
+  it('reads a Pi session from the file its hook reports', async () => {
+    const dir = await tempDir()
+    const transcriptPath = join(dir, '2026-01-01T00-00-00-000Z_pi-session-1.jsonl')
+    await writeFile(transcriptPath, piTranscript('Add a sticky-note icon', 'Icon replaced.'))
+    const { readSessionSummaryTranscriptEvents } = await loadSource()
+
+    const events = await readSessionSummaryTranscriptEvents({
+      identity: {
+        agentType: 'pi',
+        providerSession: { key: 'session_id', id: 'pi-session-1', transcriptPath }
+      },
+      signal: signal()
+    })
+
+    expect(events?.map((event) => event.text)).toEqual(['Add a sticky-note icon', 'Icon replaced.'])
+  })
+
+  it('reads a Copilot session from its session-state directory', async () => {
+    const home = await tempDir()
+    vi.stubEnv('COPILOT_HOME', home)
+    const sessionDir = join(home, 'session-state', 'copilot-session-1')
+    await mkdir(sessionDir, { recursive: true })
+    await writeFile(
+      join(sessionDir, 'events.jsonl'),
+      copilotTranscript('Why is the pane empty?', 'The transcript never resolved.')
+    )
+    const { readSessionSummaryTranscriptEvents } = await loadSource()
+
+    const events = await readSessionSummaryTranscriptEvents({
+      identity: {
+        agentType: 'copilot',
+        providerSession: { key: 'session_id', id: 'copilot-session-1' }
+      },
+      signal: signal()
+    })
+
+    expect(events?.map((event) => event.text)).toEqual([
+      'Why is the pane empty?',
+      'The transcript never resolved.'
+    ])
+  })
+
+  it('never joins a Copilot session id onto the sessions root', async () => {
+    const home = await tempDir()
+    vi.stubEnv('COPILOT_HOME', home)
+    // A transcript the traversal would reach if the id were joined unchecked.
+    await mkdir(join(home, 'session-state'), { recursive: true })
+    await writeFile(join(home, 'session-state', 'events.jsonl'), copilotTranscript('x', 'y'))
+    const { readSessionSummaryTranscriptEvents } = await loadSource()
+
+    const events = await readSessionSummaryTranscriptEvents({
+      identity: {
+        agentType: 'copilot',
+        providerSession: { key: 'session_id', id: '../' }
+      },
+      signal: signal()
+    })
+
+    expect(events).toBeNull()
+  })
+
+  it('returns null when the hook path is not a readable file', async () => {
+    const dir = await tempDir()
+    const { readSessionSummaryTranscriptEvents } = await loadSource()
+
+    const events = await readSessionSummaryTranscriptEvents({
+      identity: {
+        agentType: 'pi',
+        // A directory: resolvable, but not a transcript.
+        providerSession: { key: 'session_id', id: 'pi-session-1', transcriptPath: dir }
+      },
+      signal: signal()
+    })
+
+    expect(events).toBeNull()
+  })
+
+  it('returns null for an agent whose transcript the fold cannot parse', async () => {
+    const dir = await tempDir()
+    const transcriptPath = join(dir, 'gemini.json')
+    await writeFile(transcriptPath, '{}')
+    const { readSessionSummaryTranscriptEvents } = await loadSource()
+
+    const events = await readSessionSummaryTranscriptEvents({
+      identity: {
+        agentType: 'gemini',
+        providerSession: { key: 'session_id', id: 'gemini-1', transcriptPath }
+      },
+      signal: signal()
+    })
+
+    expect(events).toBeNull()
+  })
+})
diff --git a/src/main/session-summary/session-summary-transcript-source.ts b/src/main/session-summary/session-summary-transcript-source.ts
index e4ee5bd562e3..5727f82cb02e 100644
--- a/src/main/session-summary/session-summary-transcript-source.ts
+++ b/src/main/session-summary/session-summary-transcript-source.ts
@@ -2,6 +2,7 @@
 // ordered fold events. Preferred source per docs/reference/session-summary.md;
 // structured journal and PTY fallbacks are later adapters over the same events.
 import { stat } from 'node:fs/promises'
+import { join } from 'node:path'
 import type { AgentType } from '../../shared/agent-status-types'
 import type { AiVaultAgent } from '../../shared/ai-vault-types'
 import type { AgentProviderSessionMetadata } from '../../shared/agent-session-resume'
@@ -15,6 +16,7 @@ import {
   type TranscriptMessage
 } from '../ai-vault/session-transcript-consumers'
 import { resolveSessionFilePath } from '../native-chat/session-file-resolver'
+import { COPILOT_SESSIONS_DIR } from '../ai-vault/session-scanner-agent-sources'
 import type { SessionFileCandidate, FileWithMtime } from '../ai-vault/session-scanner-types'
 import type { SessionSummarySourceEvent } from './session-summary-fold'
 
@@ -27,6 +29,22 @@ const CANDIDATE_AGENTS: Record<NativeChatTranscriptAgent, AiVaultAgent> = {
   opencode: 'opencode'
 }
 
+/**
+ * Transcript agents outside the native-chat set, which resolves its own paths.
+ *
+ * Pi and Prime Agent post `session_file` on every hook, so the reported path is
+ * the CLI's own artifact. Copilot reports no path at all, but names its session
+ * directory after the session id the hook carries.
+ */
+const HOOK_RESOLVED_TRANSCRIPT_AGENTS: Partial<Record<AgentType, AiVaultAgent>> = {
+  pi: 'pi',
+  'prime-agent': 'prime-agent',
+  copilot: 'copilot'
+}
+
+/** Copilot's own id shape; anything else must not be joined onto a root path. */
+const COPILOT_SESSION_ID_PATTERN = /^[A-Za-z0-9_-]{1,128}$/
+
 export type SessionSummarySessionIdentity = {
   agentType?: AgentType
   providerSession?: AgentProviderSessionMetadata
@@ -43,20 +61,15 @@ export async function readSessionSummaryTranscriptEvents(args: {
   if (!providerSession?.id || !agentType) {
     return null
   }
-  const transcriptAgent = resolveNativeChatTranscriptAgent(agentType)
-  if (!transcriptAgent) {
+  const source = await resolveTranscriptSource(agentType, providerSession, signal)
+  if (!source) {
     return null
   }
-  const filePath = await resolveSessionFilePath(
-    agentType,
-    providerSession.id,
-    providerSession.transcriptPath ? { transcriptPath: providerSession.transcriptPath } : {},
-    signal
-  )
-  if (!filePath) {
+  const filePath = source.path
+  const stats = await stat(filePath)
+  if (!stats.isFile()) {
     return null
   }
-  const stats = await stat(filePath)
   const file: FileWithMtime = {
     path: filePath,
     mtimeMs: stats.mtimeMs,
@@ -64,7 +77,7 @@ export async function readSessionSummaryTranscriptEvents(args: {
     sizeBytes: stats.size
   }
   const candidate: SessionFileCandidate = {
-    agent: CANDIDATE_AGENTS[transcriptAgent],
+    agent: source.parser,
     file,
     codexHome: null
   }
@@ -96,3 +109,56 @@ export async function readSessionSummaryTranscriptEvents(args: {
     unregister()
   }
 }
+
+/**
+ * The transcript file to fold and the AI Vault parser that decodes it.
+ *
+ * Native-chat agents keep the native-chat resolver, which owns the hook-path
+ * handoff, Claude/Codex profile roots and WSL host translation. Agents outside
+ * that set are only locatable by the path their own hook reports, or — Copilot —
+ * by the session id its hook names a directory with.
+ */
+async function resolveTranscriptSource(
+  agentType: AgentType,
+  providerSession: AgentProviderSessionMetadata,
+  signal: AbortSignal
+): Promise<{ path: string; parser: AiVaultAgent } | null> {
+  const transcriptAgent = resolveNativeChatTranscriptAgent(agentType)
+  if (transcriptAgent) {
+    const path = await resolveSessionFilePath(
+      agentType,
+      providerSession.id,
+      providerSession.transcriptPath ? { transcriptPath: providerSession.transcriptPath } : {},
+      signal
+    )
+    return path ? { path, parser: CANDIDATE_AGENTS[transcriptAgent] } : null
+  }
+
+  const parser = HOOK_RESOLVED_TRANSCRIPT_AGENTS[agentType]
+  if (!parser) {
+    return null
+  }
+  // A hook-named file is the exact artifact the CLI is writing. Copilot's hook
+  // carries no path at all, so its session id names the directory instead.
+  const hookPath = providerSession.transcriptPath?.trim()
+  if (hookPath && (await isReadableFile(hookPath))) {
+    return { path: hookPath, parser }
+  }
+  return agentType === 'copilot' ? copilotTranscriptSource(providerSession.id) : null
+}
+
+function copilotTranscriptSource(sessionId: string): { path: string; parser: AiVaultAgent } | null {
+  if (!COPILOT_SESSION_ID_PATTERN.test(sessionId)) {
+    return null
+  }
+  return { path: join(COPILOT_SESSIONS_DIR, sessionId, 'events.jsonl'), parser: 'copilot' }
+}
+
+/** A path the hook named is only usable when this host can actually open it. */
+async function isReadableFile(path: string): Promise<boolean> {
+  try {
+    return (await stat(path)).isFile()
+  } catch {
+    return false
+  }
+}
diff --git a/src/renderer/src/components/activity/session-summary-presentation.test.ts b/src/renderer/src/components/activity/session-summary-presentation.test.ts
index b37e59975446..21e8863fe2fc 100644
--- a/src/renderer/src/components/activity/session-summary-presentation.test.ts
+++ b/src/renderer/src/components/activity/session-summary-presentation.test.ts
@@ -43,6 +43,48 @@ describe('resolveSessionSummaryHeaderState', () => {
     expect(resolveSessionSummaryHeaderState(null)).toEqual({ hint: 'Up to date', folding: false })
   })
 
+  it('names the missing transcript instead of claiming to be up to date', () => {
+    const state = resolveSessionSummaryHeaderState(snapshot({ status: 'unavailable' }))
+
+    expect(state).toEqual({ hint: 'No readable transcript', folding: false })
+  })
+
+  it('names the failure instead of claiming to be up to date', () => {
+    const state = resolveSessionSummaryHeaderState(snapshot({ status: 'failed' }))
+
+    expect(state).toEqual({ hint: 'Could not summarize this session', folding: false })
+  })
+
+  // A failed fold over a session already folded once leaves the last good ledger
+  // on screen, so the header keeps counting what never got folded.
+  it('counts the backlog when a fold failed over an existing ledger', () => {
+    const state = resolveSessionSummaryHeaderState(
+      snapshot({
+        status: 'failed',
+        ledger: {
+          goal: null,
+          plan: [],
+          inProgress: null,
+          done: [],
+          blockers: [],
+          decisions: [],
+          timeline: [],
+          foldedThrough: 4,
+          foldedAt: 1
+        },
+        facts: {
+          paneKey: 'pane-1',
+          prompt: 'Refactor the parser',
+          messageCount: 9,
+          foldedThrough: 4,
+          backlogCount: 5
+        }
+      })
+    )
+
+    expect(state).toEqual({ hint: '5 events to catch up on', folding: false })
+  })
+
   it('reads as up to date when facts have no backlog', () => {
     const state = resolveSessionSummaryHeaderState(
       snapshot({
diff --git a/src/renderer/src/components/activity/session-summary-presentation.ts b/src/renderer/src/components/activity/session-summary-presentation.ts
index 593f6a5e4db9..595ac404c674 100644
--- a/src/renderer/src/components/activity/session-summary-presentation.ts
+++ b/src/renderer/src/components/activity/session-summary-presentation.ts
@@ -15,6 +15,23 @@ export function resolveSessionSummaryHeaderState(
   if (snapshot?.fold.running) {
     return { hint: translate('activity.sessionSummary.updating', 'Updating…'), folding: true }
   }
+  // Why: without a ledger there is nothing on screen, so the hint has to name
+  // why. Falling through to "Up to date" told the reader a session with no
+  // readable transcript was fully summarized.
+  if (!snapshot?.ledger) {
+    if (snapshot?.status === 'failed') {
+      return {
+        hint: translate('activity.sessionSummary.failedHint', 'Could not summarize this session'),
+        folding: false
+      }
+    }
+    if (snapshot?.status === 'unavailable') {
+      return {
+        hint: translate('activity.sessionSummary.noTranscriptHint', 'No readable transcript'),
+        folding: false
+      }
+    }
+  }
   const backlogCount = snapshot?.facts?.backlogCount ?? 0
   if (backlogCount > 0) {
     return {
diff --git a/src/renderer/src/components/right-sidebar/SessionSummaryPane.test.tsx b/src/renderer/src/components/right-sidebar/SessionSummaryPane.test.tsx
index 2c46c87bd9ec..abd4f0d66d36 100644
--- a/src/renderer/src/components/right-sidebar/SessionSummaryPane.test.tsx
+++ b/src/renderer/src/components/right-sidebar/SessionSummaryPane.test.tsx
@@ -89,7 +89,7 @@ describe('SessionSummaryPane', () => {
 
     expect(markup).toContain('Session summary')
     expect(markup).toContain('Up to date')
-    expect(markup).toContain('lucide-info')
+    expect(markup).toContain('lucide-sticky-note')
     expect(markup).toContain('Wiring the right-sidebar tab')
     expect(markup).toContain('Moved the ledger view out of the activity strip')
   })
diff --git a/src/renderer/src/components/right-sidebar/SessionSummaryPane.tsx b/src/renderer/src/components/right-sidebar/SessionSummaryPane.tsx
index 9ba0705a767a..638beaf4ce8a 100644
--- a/src/renderer/src/components/right-sidebar/SessionSummaryPane.tsx
+++ b/src/renderer/src/components/right-sidebar/SessionSummaryPane.tsx
@@ -1,5 +1,5 @@
 import React from 'react'
-import { Info } from 'lucide-react'
+import { Info, StickyNote } from 'lucide-react'
 import { cn } from '@/lib/utils'
 import { translate } from '@/i18n/i18n'
 import { SessionSummaryLedgerView } from '@/components/activity/session-summary-ledger-view'
@@ -21,7 +21,7 @@ function SessionSummarySubjectBody({
   return (
     <>
       <div className="flex shrink-0 items-center gap-1.5 border-b border-border px-4 py-1.5">
-        <Info className="size-3.5 shrink-0 text-muted-foreground" aria-hidden="true" />
+        <StickyNote className="size-3.5 shrink-0 text-muted-foreground" aria-hidden="true" />
         <span className="text-[11px] font-semibold uppercase tracking-wider text-muted-foreground">
           {translate('activity.sessionSummary.label', 'Session summary')}
         </span>
diff --git a/src/renderer/src/components/right-sidebar/use-right-sidebar-activity-items.ts b/src/renderer/src/components/right-sidebar/use-right-sidebar-activity-items.ts
index 45e64b094462..a756ee04740c 100644
--- a/src/renderer/src/components/right-sidebar/use-right-sidebar-activity-items.ts
+++ b/src/renderer/src/components/right-sidebar/use-right-sidebar-activity-items.ts
@@ -1,5 +1,5 @@
 import { useMemo } from 'react'
-import { CircleDot, Info, Plug, Files, GitBranch, ListChecks, Workflow } from 'lucide-react'
+import { CircleDot, Plug, Files, GitBranch, ListChecks, StickyNote, Workflow } from 'lucide-react'
 import { useAppStore } from '@/store'
 import { useRepoById } from '@/store/selectors'
 import { isFolderRepo } from '../../../../shared/repo-kind'
@@ -89,7 +89,7 @@ export function useRightSidebarActivityItems({
               // Why: the pane summarizes an agent session in this workspace, so it sits
               // beside Agents rather than with the source-control group.
               id: 'summary' as const,
-              icon: Info,
+              icon: StickyNote,
               title: translate('activity.sessionSummary.label', 'Session summary'),
               shortcut: ''
             }
diff --git a/src/renderer/src/i18n/locales/en.json b/src/renderer/src/i18n/locales/en.json
index f4645e3d761d..db68316a14d9 100644
--- a/src/renderer/src/i18n/locales/en.json
+++ b/src/renderer/src/i18n/locales/en.json
@@ -19970,7 +19970,9 @@
       "upToDate": "Up to date",
       "label": "Session summary",
       "failed": "Could not summarize this session.",
+      "failedHint": "Could not summarize this session",
       "unavailable": "No readable transcript for this session yet.",
+      "noTranscriptHint": "No readable transcript",
       "unavailableInWindow": "Session summaries are not available in this window.",
       "notFocused": "Focus an agent session to see its summary."
     }
diff --git a/src/renderer/src/lib/ai-generation-settings.test.ts b/src/renderer/src/lib/ai-generation-settings.test.ts
index 78bb4d3013f5..fe6a79ce7edb 100644
--- a/src/renderer/src/lib/ai-generation-settings.test.ts
+++ b/src/renderer/src/lib/ai-generation-settings.test.ts
@@ -110,8 +110,9 @@ describe('resolveAiGenerationSelection', () => {
 })
 
 describe('getCommitMessageAgentCapability', () => {
-  it('reports Pi as dynamic so the pane knows to probe', () => {
+  it('reports the agents whose model list is probed so the pane knows to probe', () => {
     expect(getCommitMessageAgentCapability('pi')?.modelSource).toBe('dynamic')
-    expect(getCommitMessageAgentCapability('copilot')?.modelSource).toBe('static')
+    expect(getCommitMessageAgentCapability('copilot')?.modelSource).toBe('dynamic')
+    expect(getCommitMessageAgentCapability('amp')?.modelSource).toBe('static')
   })
 })
diff --git a/src/shared/commit-message-agent-spec.test.ts b/src/shared/commit-message-agent-spec.test.ts
index 18635420b4cc..54aa4b9539a2 100644
--- a/src/shared/commit-message-agent-spec.test.ts
+++ b/src/shared/commit-message-agent-spec.test.ts
@@ -12,6 +12,7 @@ import {
 } from './commit-message-agent-spec'
 import {
   COMMIT_MESSAGE_MODEL_JSON_STRUCTURE_LIMITS,
+  OPENAI_THINKING_LEVELS,
   parseAntigravityModels,
   parseClaudeModels,
   parseCodexModels,
@@ -19,6 +20,7 @@ import {
   parseLineModels,
   parsePiModels
 } from './commit-message-model-parsers'
+import { COPILOT_MODEL_DISCOVERY, parseCopilotModels } from './copilot-model-list-probe'
 
 afterEach(() => {
   vi.restoreAllMocks()
@@ -119,28 +121,64 @@ describe('COMMIT_MESSAGE_AGENT_SPECS', () => {
     }
   })
 
-  it('lists Copilot hosted CLI models even when account policy filters the picker', () => {
-    expect(COMMIT_MESSAGE_AGENT_SPECS.copilot?.defaultModelId).toBe('gpt-5.4')
-    expect(COMMIT_MESSAGE_AGENT_SPECS.copilot?.models.map((m) => m.id)).toEqual([
+  it('probes Copilot for its hosted CLI catalog instead of pinning a snapshot', () => {
+    const copilot = COMMIT_MESSAGE_AGENT_SPECS.copilot!
+
+    // The CLI ships no `models` subcommand; `help config` is the catalog it
+    // publishes, and it is what keeps this list current.
+    expect(copilot.modelSource).toBe('dynamic')
+    expect(copilot.modelDiscovery?.binary).toBe('copilot')
+    expect(copilot.modelDiscovery?.args).toEqual(['help', 'config'])
+    expect(copilot.defaultModelId).toBe('gpt-5.4')
+    // Models released after this fallback was frozen still resolve, because the
+    // probe reports them rather than the spec's own list.
+    expect(copilot.modelDiscovery).toEqual(COPILOT_MODEL_DISCOVERY)
+    expect(copilot.modelDiscovery?.parse('`model`: AI model to use.\n  - "gpt-9"\n')).toEqual([
+      { id: 'auto', label: 'Auto' },
+      {
+        id: 'gpt-9',
+        label: 'GPT 9',
+        thinkingLevels: OPENAI_THINKING_LEVELS,
+        defaultThinkingLevel: 'low'
+      }
+    ])
+  })
+
+  it('lists Copilot models from its static fallback when the CLI cannot be probed', () => {
+    const ids = COMMIT_MESSAGE_AGENT_SPECS.copilot!.models.map((model) => model.id)
+
+    expect(ids[0]).toBe('auto')
+    expect(ids).toContain('gpt-5.4')
+    expect(ids.length).toBeGreaterThan(20)
+    // Never re-offer a model the CLI has dropped.
+    expect(ids).not.toContain('gpt-4.1')
+  })
+
+  it('reads Copilot model ids only from the `model` setting block', () => {
+    const stdout = [
+      '  `defaultMode`: mode for new sessions.',
+      '    - "interactive"',
+      '  `model`: AI model to use for Copilot CLI.',
+      '    - "claude-opus-5.5"',
+      '    - "gpt-6-luna"',
+      '  `contextTier`: context window tier.',
+      '    - "default"',
+      ''
+    ].join('\n')
+
+    expect(parseCopilotModels(stdout).map((model) => model.id)).toEqual([
       'auto',
-      'claude-haiku-4.5',
-      'claude-sonnet-4.5',
-      'claude-sonnet-4.6',
-      'claude-opus-4.5',
-      'claude-opus-4.6',
-      'claude-opus-4.6-fast',
-      'claude-opus-4.7',
-      'gpt-4.1',
-      'gpt-5-mini',
-      'gpt-5.2',
-      'gpt-5.2-codex',
-      'gpt-5.3-codex',
-      'gpt-5.4',
-      'gpt-5.4-mini',
-      'gpt-5.5'
+      'claude-opus-5.5',
+      'gpt-6-luna'
     ])
   })
 
+  it('parses nothing when the Copilot help format changes', () => {
+    // An empty parse is the safe failure: the spec's fallback list is used and
+    // the result is never cached, so the next probe tries again.
+    expect(parseCopilotModels('Usage: copilot config [OPTIONS]')).toEqual([])
+  })
+
   it('treats disabled default agents as unavailable for implicit Source Control AI choices', () => {
     expect(resolveCommitMessageAgentChoice(null, 'codex', ['codex'])).toBe('claude')
     expect(resolveCommitMessageAgentChoice(null, null, ['claude'])).toBeNull()
diff --git a/src/shared/commit-message-agent-spec.ts b/src/shared/commit-message-agent-spec.ts
index bd47acc5bac9..684771decf78 100644
--- a/src/shared/commit-message-agent-spec.ts
+++ b/src/shared/commit-message-agent-spec.ts
@@ -16,6 +16,7 @@ import {
   parsePiModels,
   withOpenAiThinking
 } from './commit-message-model-parsers'
+import { COPILOT_FALLBACK_MODELS, COPILOT_MODEL_DISCOVERY } from './copilot-model-list-probe'
 
 // Why: this file is the source of truth for non-interactive agent invocation
 // (commit-message generation). It is intentionally separate from
@@ -127,10 +128,13 @@ export const COMMIT_MESSAGE_AGENT_SPECS: Partial<Record<TuiAgent, CommitMessageA
   }),
   ...buildSecondaryCommitMessageAgentSpecs({
     BASIC_THINKING_LEVELS,
-    OPENAI_THINKING_LEVELS,
     parseCursorModels,
     parseAntigravityModels,
-    parseLineModels
+    parseLineModels,
+    copilotCatalog: {
+      modelDiscovery: COPILOT_MODEL_DISCOVERY,
+      fallbackModels: COPILOT_FALLBACK_MODELS
+    }
   })
 }
 
diff --git a/src/shared/commit-message-agent-specs-secondary.ts b/src/shared/commit-message-agent-specs-secondary.ts
index a287d91829a5..58c60d67d0b4 100644
--- a/src/shared/commit-message-agent-specs-secondary.ts
+++ b/src/shared/commit-message-agent-specs-secondary.ts
@@ -7,18 +7,22 @@ import type {
 
 type SecondaryAgentSpecDeps = {
   BASIC_THINKING_LEVELS: ThinkingLevel[]
-  OPENAI_THINKING_LEVELS: ThinkingLevel[]
   parseCursorModels: (stdout: string) => CommitMessageModel[]
   parseAntigravityModels: (stdout: string) => CommitMessageModel[]
   parseLineModels: (stdout: string) => CommitMessageModel[]
+  /** Copilot's probe and the catalog a machine without the binary falls back to. */
+  copilotCatalog: {
+    modelDiscovery: NonNullable<CommitMessageAgentSpec['modelDiscovery']>
+    fallbackModels: CommitMessageModel[]
+  }
 }
 
 export function buildSecondaryCommitMessageAgentSpecs({
   BASIC_THINKING_LEVELS,
-  OPENAI_THINKING_LEVELS,
   parseCursorModels,
   parseAntigravityModels,
-  parseLineModels
+  parseLineModels,
+  copilotCatalog
 }: SecondaryAgentSpecDeps): Partial<Record<TuiAgent, CommitMessageAgentSpec>> {
   return {
     amp: {
@@ -174,102 +178,17 @@ export function buildSecondaryCommitMessageAgentSpecs({
         '--no-custom-instructions',
         '--model',
         model,
-        ...(thinkingLevel ? ['--effort', thinkingLevel] : [])
-      ],
-      modelSource: 'static',
-      // Why: Copilot CLI's picker is policy-filtered per account/org. Keep the
-      // full hosted CLI catalog here so users can select models enabled for them.
-      models: [
-        { id: 'auto', label: 'Auto' },
-        {
-          id: 'claude-haiku-4.5',
-          label: 'Claude Haiku 4.5',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'claude-sonnet-4.5',
-          label: 'Claude Sonnet 4.5',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'claude-sonnet-4.6',
-          label: 'Claude Sonnet 4.6',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'claude-opus-4.5',
-          label: 'Claude Opus 4.5',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'claude-opus-4.6',
-          label: 'Claude Opus 4.6',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'claude-opus-4.6-fast',
-          label: 'Claude Opus 4.6 Fast',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'claude-opus-4.7',
-          label: 'Claude Opus 4.7',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-4.1',
-          label: 'GPT-4.1'
-        },
-        {
-          id: 'gpt-5-mini',
-          label: 'GPT-5 Mini',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-5.2',
-          label: 'GPT-5.2',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-5.2-codex',
-          label: 'GPT-5.2 Codex',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-5.3-codex',
-          label: 'GPT-5.3 Codex',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-5.4',
-          label: 'GPT-5.4',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-5.4-mini',
-          label: 'GPT-5.4 Mini',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        },
-        {
-          id: 'gpt-5.5',
-          label: 'GPT-5.5',
-          thinkingLevels: OPENAI_THINKING_LEVELS,
-          defaultThinkingLevel: 'low'
-        }
+        ...(thinkingLevel ? ['--reasoning-effort', thinkingLevel] : [])
       ],
+      modelSource: 'dynamic',
+      // Why: the CLI has no `models` subcommand, but `help config` documents the
+      // `model` setting's accepted ids — the CLI's own catalog, so it keeps up
+      // with models released after this file was written. (The interactive picker
+      // is policy-filtered per account/org; this is the catalog behind it.)
+      modelDiscovery: copilotCatalog.modelDiscovery,
+      // Used verbatim only when the probe cannot run — no `copilot` on PATH — and
+      // kept out of the cache, so a later probe replaces it.
+      models: copilotCatalog.fallbackModels,
       defaultModelId: 'gpt-5.4'
     },
     antigravity: {
diff --git a/src/shared/copilot-model-list-probe.ts b/src/shared/copilot-model-list-probe.ts
new file mode 100644
index 000000000000..cdb6e06f631e
--- /dev/null
+++ b/src/shared/copilot-model-list-probe.ts
@@ -0,0 +1,119 @@
+import type { CommitMessageAgentSpec, CommitMessageModel } from './commit-message-agent-spec'
+import { OPENAI_THINKING_LEVELS } from './commit-message-model-parsers'
+import { labelFromModelId } from './model-id-label'
+
+// Why: the Copilot CLI has no `models` subcommand, but `help config` documents the
+// `model` setting's accepted ids — the CLI's own catalog, so the list keeps up with
+// models released after this file was written. Its interactive picker is
+// policy-filtered per account/org; this is the catalog behind it.
+const COPILOT_MODEL_LIST_ARGS = ['help', 'config']
+
+const COPILOT_AUTO_MODEL_ID = 'auto'
+
+/**
+ * Parses `copilot help config`: the `model` setting's heading, then its accepted
+ * ids as quoted bullets, ended by the next documented setting.
+ *
+ * A format change parses to nothing, which falls back to `COPILOT_FALLBACK_MODELS`
+ * and is never cached, so the next probe tries again.
+ */
+export function parseCopilotModels(stdout: string): CommitMessageModel[] {
+  const ids: string[] = []
+  let inModelSetting = false
+  for (const rawLine of stdout.split(/\r\n|\r|\n/)) {
+    if (/^\s*`model`\s*:/.test(rawLine)) {
+      inModelSetting = true
+      continue
+    }
+    if (!inModelSetting) {
+      continue
+    }
+    // The next documented setting ends this one's value list.
+    if (/^\s*`[^`]+`\s*:/.test(rawLine)) {
+      break
+    }
+    const quoted = /^\s+-\s+"([^"]+)"\s*$/.exec(rawLine)
+    const id = quoted?.[1]?.trim()
+    if (id) {
+      ids.push(id)
+    }
+  }
+  return copilotModelsFromIds(ids)
+}
+
+/**
+ * Copilot publishes no display names, so a label is derived from the id like every
+ * other probed model. Every model but `auto` takes a reasoning effort:
+ * `--reasoning-effort` is a CLI-level flag, so the picker offers the documented
+ * subset for each one rather than guessing per model.
+ */
+function copilotModelsFromIds(ids: readonly string[]): CommitMessageModel[] {
+  // Why: `--model` documents `auto` as "let Copilot pick", but the setting's value
+  // list omits it, so the CLI's own routing choice is added back ahead of the ids.
+  const withAuto =
+    ids.length > 0 && !ids.includes(COPILOT_AUTO_MODEL_ID) ? [COPILOT_AUTO_MODEL_ID, ...ids] : ids
+  const seen = new Set<string>()
+  return withAuto
+    .filter((id) => {
+      if (!id || seen.has(id)) {
+        return false
+      }
+      seen.add(id)
+      return true
+    })
+    .map((id) =>
+      id === COPILOT_AUTO_MODEL_ID
+        ? { id, label: 'Auto' }
+        : {
+            id,
+            label: labelFromModelId(id),
+            thinkingLevels: OPENAI_THINKING_LEVELS,
+            defaultThinkingLevel: 'low'
+          }
+    )
+}
+
+/** Probe descriptor for the copilot spec; owns the argv it spawns. */
+export const COPILOT_MODEL_DISCOVERY: NonNullable<CommitMessageAgentSpec['modelDiscovery']> = {
+  binary: 'copilot',
+  args: COPILOT_MODEL_LIST_ARGS,
+  parse: parseCopilotModels
+}
+
+/**
+ * What a machine with no `copilot` binary falls back to, frozen at the catalog the
+ * CLI shipped when this list was last read. The live catalog comes from the probe,
+ * so this only has to be a usable starting point — never authoritative.
+ */
+export const COPILOT_FALLBACK_MODELS: CommitMessageModel[] = copilotModelsFromIds([
+  'claude-sonnet-5.5',
+  'claude-sonnet-5',
+  'claude-fable-5.1',
+  'claude-fable-5',
+  'claude-opus-5.5',
+  'claude-opus-5',
+  'claude-opus-4.8',
+  'claude-opus-4.8-fast',
+  'claude-sonnet-4.6',
+  'claude-haiku-5.5',
+  'claude-haiku-4.5',
+  'gpt-6.1-sol',
+  'gpt-6-sol',
+  'gpt-6-luna',
+  'gpt-6-astra',
+  'gpt-5.6-sol',
+  'gpt-5.6-terra',
+  'gpt-5.6-luna',
+  'gpt-5.5',
+  'gpt-5.4',
+  'gpt-5.4-mini',
+  'gpt-5.3-codex',
+  'gpt-5-mini',
+  'mai-code-1.1-flash',
+  'gemini-3.8-flash',
+  'gemini-3.7-flash',
+  'grok-4.7',
+  'grok-4.6',
+  'grok-4.5',
+  'kimi-k3'
+])
diff --git a/src/shared/session-summary-ai.test.ts b/src/shared/session-summary-ai.test.ts
index e3ae756f152a..845459275fee 100644
--- a/src/shared/session-summary-ai.test.ts
+++ b/src/shared/session-summary-ai.test.ts
@@ -63,17 +63,30 @@ describe('resolveSessionSummaryAiParams', () => {
   })
 
   it('falls back to the default model when a static agent no longer offers the stored one', () => {
-    const copilot = getCommitMessageAgentSpec('copilot')
+    const amp = getCommitMessageAgentSpec('amp')
+    const resolved = resolveSessionSummaryAiParams(
+      input({
+        sessionSummaryAi: config({
+          agentId: 'amp',
+          selectedModelByAgent: { amp: 'amp-9000' }
+        })
+      })
+    )
+
+    expect(resolved?.model).toBe(amp?.defaultModelId)
+  })
+
+  it('passes an unknown model through for an agent whose list is probed', () => {
     const resolved = resolveSessionSummaryAiParams(
       input({
         sessionSummaryAi: config({
           agentId: 'copilot',
-          selectedModelByAgent: { copilot: 'copilot-9000' }
+          selectedModelByAgent: { copilot: 'gpt-9' }
         })
       })
     )
 
-    expect(resolved?.model).toBe(copilot?.defaultModelId)
+    expect(resolved?.model).toBe('gpt-9')
   })
 
   it('falls back to the model default when the stored effort is not offered', () => {