diff --git a/.gitignore b/.gitignore
index 913dfc4a045..669ec4528f9 100644
--- a/.gitignore
+++ b/.gitignore
@@ -114,6 +114,7 @@ docs/**
 !docs/reference/windows-cmd-shim-resolution.md
 !docs/reference/windows-daemon-host-relocation.md
 !docs/reference/windows-edr-posture.md
+!docs/reference/windows-msys-job-breakaway.md
 !docs/reference/windows-process-enumeration.md
 !docs/reference/wsl-runner-verification.md
 !docs/reference/remote-wire-compatibility.md
diff --git a/AGENTS.md b/AGENTS.md
index 5ff66b95b0f..4d93a3ab738 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -55,6 +55,7 @@ Orca targets macOS, Linux, and Windows. Keep all platform-dependent behavior beh
 - **Windows setup scripts**: the setup/issue-command runner is a `.cmd` batch file unless the script starts with a `#!` line — never derive that from the user's terminal-shell preference, and never launch a `.cmd` runner with a bare `cmd.exe /c` from a Git Bash pane (MSYS rewrites the `/c`). See [`docs/reference/windows-setup-shell.md`](./docs/reference/windows-setup-shell.md).
 - **Windows child processes**: start them through `runProcess`/`spawnProcess` in `src/shared/child-process/` — never `child_process` directly. It pins `windowsHide`, refuses `shell: true`, and encodes `.cmd`/`.bat` arguments so neither `CommandLineToArgvW` nor `cmd.exe` mangles them. A ratchet test fails on any new direct import. Recognised npm/pnpm `.cmd` shims are resolved to their real target so the spawn skips `cmd.exe` entirely; see [`docs/reference/windows-cmd-shim-resolution.md`](./docs/reference/windows-cmd-shim-resolution.md) before adding a shim shape or debugging one.
 - **Windows process enumeration**: read the table through `src/main/windows/windows-process-table.ts`, never by forking `powershell.exe`. See [`docs/reference/windows-process-enumeration.md`](./docs/reference/windows-process-enumeration.md).
+- **Windows MSYS/Git Bash panes**: their children break away from the per-PTY job unless it is created without `JOB_OBJECT_LIMIT_BREAKAWAY_OK`, and a `conpty.node` built before that fix passes every existing gate. Before changing the per-PTY job or debugging `windows-msys-job.win32.test.ts`, read [`docs/reference/windows-msys-job-breakaway.md`](./docs/reference/windows-msys-job-breakaway.md).
 - **Windows daemon-host relocation**: the terminal daemon runs from a copy of the app runtime under `%LOCALAPPDATA%`, which is what survives an auto-update. Before touching that copy, its exe name, or the NSIS uninstall macro, read [`docs/reference/windows-daemon-host-relocation.md`](./docs/reference/windows-daemon-host-relocation.md).
 - **Windows EDR signal**: don't add `-ExecutionPolicy Bypass`, `-EncodedCommand`, `cmd.exe /c` with escaped free text, per-operation interpreter spawning, or runtime `Add-Type` compilation without reading [`docs/reference/windows-edr-posture.md`](./docs/reference/windows-edr-posture.md) first — behavioural EDR scores each of those, and being signed does not clear them.
 - **WSL commands**: build argv with `buildWslExecArgs` (always `--exec` — under `--`, `wsl.exe` expands `$name` in every argument and silently rewrites the script), and fence anything whose stdout you parse with `buildWslCapturedLoginShellCommand`, because the interactive login shell prints the distro banner to stdout. See [`docs/reference/wsl-command-execution.md`](./docs/reference/wsl-command-execution.md).
diff --git a/docs/reference/windows-msys-job-breakaway.md b/docs/reference/windows-msys-job-breakaway.md
new file mode 100644
index 00000000000..8b0261628c1
--- /dev/null
+++ b/docs/reference/windows-msys-job-breakaway.md
@@ -0,0 +1,100 @@
+# Why an MSYS pane's children escape the per-PTY job
+
+Every child started from a Git Bash / MSYS2 / Cygwin pane leaves the pane's job
+object unless the job is created **without** `JOB_OBJECT_LIMIT_BREAKAWAY_OK`.
+`terminatePtyJob` then reports `terminated` and leaves the child running — the
+orphan that holds a worktree directory open.
+
+The denial is already in `config/patches/node-pty@1.1.0.patch`
+(`usesCygwinRuntime`, added in #19068). This page records the measurement
+behind it, because the failure mode it prevents is indistinguishable from a
+stale native addon and the existing gates cannot tell the two apart.
+
+## The mechanism
+
+The MSYS/Cygwin runtime asks for `CREATE_BREAKAWAY_FROM_JOB` on the
+`CreateProcessW` inside its `spawn`/`exec` path. A job that carries
+`JOB_OBJECT_LIMIT_BREAKAWAY_OK` grants it, so the child is created outside the
+job; a job without that limit denies it with `ERROR_ACCESS_DENIED`, and the
+runtime retries without the flag rather than failing the spawn. `fork` is not
+affected — forked Cygwin processes stay in the job either way.
+
+Measured on Windows 11 `10.0.26200.9168`, Git `2.55.0.windows.3`,
+bash `5.3.15(1)-release`, node `v24.18.0`, `useConptyDll: true`, for
+`node-pty.spawn('C:\Program Files\Git\bin\bash.exe', ['--noprofile','--norc','-i'])`
+— `+J` / `-J` is membership of the per-PTY job, read with
+`QueryInformationJobObject(JobObjectBasicProcessIdList)`:
+
+```
+bin\bash.exe            +J   ConPTY shell (assigned by node-pty)
+ └ ..\usr\bin\bash.exe  +J   launcher hand-off, plain CreateProcess
+    └ usr\bin\bash.exe  +J   Cygwin fork for the typed command
+       └ node.exe       -J   Cygwin exec -- ESCAPES HERE
+```
+
+`bin\bash.exe` is a 47 KB launcher, not an MSYS binary: `C:\Program Files\Git\bin`
+holds only `bash.exe`, `git.exe` and `sh.exe`, with no `msys-2.0.dll`. Its
+hand-off to `bin\..\usr\bin\bash.exe` is an ordinary `CreateProcess` and keeps
+job membership. Only the MSYS runtime's own spawn breaks away.
+
+The shell-replacement shape (`bash -c 'exec "$BASH" --noprofile --norc -i'`)
+loses membership one step earlier, at the `exec`, and everything below inherits
+the loss:
+
+```
+bin\bash.exe            +J
+ └ ..\usr\bin\bash.exe  +J
+    └ usr\bin\bash.exe  -J   Cygwin exec -- ESCAPES HERE
+       └ usr\bin\bash   -J
+          └ node.exe    -J
+```
+
+Both shapes leak. The `exec` is not the cause; it only moves the escape earlier.
+
+## The A/B that pins it
+
+One source tree, one toolchain, one variable — `usesCygwinRuntime` forced to
+`false` so the per-PTY job keeps `JOB_OBJECT_LIMIT_BREAKAWAY_OK`:
+
+| per-PTY job limit      | `listPtyJobProcessIds` | child reaped by `terminatePtyJob` | runs |
+| ---------------------- | ---------------------- | --------------------------------- | ---- |
+| `BREAKAWAY_OK` set     | 2 pids, child absent   | no                                | 0/2  |
+| `BREAKAWAY_OK` cleared | 5 pids, child present  | yes                               | 4/4  |
+
+The job **is** the right boundary. With breakaway denied it holds the whole MSYS
+tree, including the child that detached from the console, and one
+`terminateJob` reaps all of it. No alternative tracking mechanism is needed.
+
+Denying breakaway did not break ordinary launches from the pane: `git`,
+`cmd //c`, an absolute-path `node`, a `&`-backgrounded job with `disown`, and
+`where.exe` all returned 0 with no `Access is denied`, identically to the
+breakaway-allowed control. Untested: a **non-Cygwin** program that itself passes
+`CREATE_BREAKAWAY_FROM_JOB` (installers, updaters) and therefore has no runtime
+to retry for it. That needs a helper that calls `CreateProcess` with the flag;
+`start /b` does not exercise it (it uses `CREATE_NEW_CONSOLE`).
+
+## A stale addon looks exactly like the bug
+
+`config/scripts/node-pty-job-ownership.cjs` asserts only that `terminateJob`,
+`listJobProcessIds` and `assignCurrentProcessToJob` are exported. All three
+predate #19068, so a `conpty.node` built before it passes every gate,
+`isPtyJobOwnershipAvailable()` returns true, `windows-pty-job.win32.test.ts`
+passes 6/6 — and `windows-msys-job.win32.test.ts` fails with a two-pid job list
+that reads as a source defect rather than a build-freshness one.
+
+When that test fails, check the binary before the code:
+
+```js
+// UTF-16LE, because usesCygwinRuntime holds the literals
+readFileSync(conptyNodePath).includes(Buffer.from('msys-2.0.dll', 'utf16le'))
+```
+
+False means the addon predates the fix; rebuild node-pty from patched source.
+Note that a git worktree sharing `node_modules` with its main checkout shares
+that checkout's `build/Release/conpty.node`, so pinning the _source_ to a commit
+does not pin the _addon_.
+
+The gate should assert the same marker, the way `stagedRelayAddonIsUnpatched()`
+in `src/main/windows/windows-process-table.ts` already sniffs a patched addon by
+a binary import name. Symbol presence cannot distinguish patch revisions; a
+marker or an exported revision number can.