claude-code-sessions 0.16.1__tar.gz → 0.16.3__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/LICENSE +21 -21
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/PKG-INFO +58 -14
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/README.md +58 -14
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.egg-info/PKG-INFO +58 -14
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.egg-info/SOURCES.txt +1 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.py +288 -59
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions_gui.py +10 -9
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/pyproject.toml +50 -50
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_alignment.py +144 -144
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_anonymize.py +7 -7
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_cli.py +172 -172
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_containment.py +71 -71
- claude_code_sessions-0.16.3/tests/test_data_root.py +515 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_discovery.py +48 -48
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_encoding.py +41 -41
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_engine.py +274 -274
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_golden.py +18 -18
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_helpers.py +39 -39
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_journal.py +87 -87
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_plan_move.py +215 -35
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_recover.py +413 -400
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_rows.py +73 -73
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_scaffold.py +18 -18
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_transcripts.py +24 -24
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_undo.py +312 -298
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.egg-info/dependency_links.txt +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.egg-info/entry_points.txt +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.egg-info/top_level.txt +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/setup.cfg +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_commands.py +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_converge.py +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_duplicate_transcripts.py +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_gui_models.py +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_retitle.py +0 -0
- {claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/tests/test_sync.py +0 -0
|
@@ -1,21 +1,21 @@
|
|
|
1
|
-
MIT License
|
|
2
|
-
|
|
3
|
-
Copyright (c) 2026 Craig Stoller
|
|
4
|
-
|
|
5
|
-
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
-
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
-
in the Software without restriction, including without limitation the rights
|
|
8
|
-
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
-
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
-
furnished to do so, subject to the following conditions:
|
|
11
|
-
|
|
12
|
-
The above copyright notice and this permission notice shall be included in all
|
|
13
|
-
copies or substantial portions of the Software.
|
|
14
|
-
|
|
15
|
-
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
-
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
-
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
-
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
-
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
-
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
-
SOFTWARE.
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 Craig Stoller
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: claude-code-sessions
|
|
3
|
-
Version: 0.16.
|
|
3
|
+
Version: 0.16.3
|
|
4
4
|
Summary: Unofficial tool for Claude Code sessions on disk: make a session visible under your other Claude account on the same machine, or move one between project folders. Command line or GUI. Journaled, reversible, fails closed.
|
|
5
5
|
Author: Craig Stoller
|
|
6
6
|
License: MIT
|
|
@@ -124,13 +124,14 @@ of the standard library, but packaged separately on some Linux distributions
|
|
|
124
124
|
replaces the old "Claude session sync" shortcut with the new name; a taskbar pin to the
|
|
125
125
|
old name is yours to re-pin.
|
|
126
126
|
|
|
127
|
-
## Optional: let
|
|
127
|
+
## Optional: let the browser stay open
|
|
128
128
|
|
|
129
|
-
The desktop app ships a helper for its
|
|
130
|
-
guard counts it and you have to close
|
|
131
|
-
excluded - but **the helper auto-updates every few days**
|
|
132
|
-
measured), and each update correctly fails closed, so the
|
|
133
|
-
be closed again until someone re-runs the ~12-minute
|
|
129
|
+
The desktop app ships a helper for its browser extension that Chrome or Edge launches and
|
|
130
|
+
that can outlive the app, so the guard counts it and you have to close that browser too. A
|
|
131
|
+
measured build of that helper is excluded - but **the helper auto-updates every few days**
|
|
132
|
+
(three builds in eight days, as measured), and each update correctly fails closed, so the
|
|
133
|
+
exclusion lapses and the browser has to be closed again until someone re-runs the ~12-minute
|
|
134
|
+
measurement in
|
|
134
135
|
`tools/native-host-measurement/`.
|
|
135
136
|
|
|
136
137
|
If that trade isn't worth it to you, there is an opt-in: trust **any** helper at the app's
|
|
@@ -142,7 +143,7 @@ claude-code-sessions trust-signed-helper --on # enable it
|
|
|
142
143
|
claude-code-sessions trust-signed-helper --off # back to the default
|
|
143
144
|
```
|
|
144
145
|
|
|
145
|
-
In the window it is the **"Let
|
|
146
|
+
In the window it is the **"Let browsers stay open"** checkbox, which explains the trade before
|
|
146
147
|
enabling anything. Either way it is one marker file - `~/.claude-code-journal/trust-signed-helper` -
|
|
147
148
|
so deleting that file also revokes it.
|
|
148
149
|
|
|
@@ -175,9 +176,10 @@ files agree - or, if you know which account the desktop app is signed into, asse
|
|
|
175
176
|
full). Signing into the desktop app does **not** refresh `~/.claude.json`. A refused
|
|
176
177
|
`sync --apply` (desktop app running) refuses **before anything is journaled** - no lock file,
|
|
177
178
|
no op directory - so there's nothing left behind for `doctor` to flag or `recover` to clean up;
|
|
178
|
-
just close the app and re-run. One measured wrinkle: the desktop ships a helper for its
|
|
179
|
-
extension (`chrome-native-host.exe`)
|
|
180
|
-
if a refusal names that process, fully exit
|
|
179
|
+
just close the app and re-run. One measured wrinkle: the desktop ships a helper for its browser
|
|
180
|
+
extension (`chrome-native-host.exe`), launched by Chrome or Edge, that can outlive the app
|
|
181
|
+
itself, and the guard counts it - if a refusal names that process, fully exit that browser too
|
|
182
|
+
(the refusal always names what it saw).
|
|
181
183
|
|
|
182
184
|
## Usage
|
|
183
185
|
|
|
@@ -192,7 +194,7 @@ alone cannot tell you which copy answered:
|
|
|
192
194
|
```
|
|
193
195
|
$ ccs --version
|
|
194
196
|
claude-code-sessions 0.10.0
|
|
195
|
-
running from: C:\Users
|
|
197
|
+
running from: C:\Users\<name>\AppData\Local\pipx\...\claude_code_sessions.py
|
|
196
198
|
```
|
|
197
199
|
|
|
198
200
|
**`list`** - inventory sessions, optionally filtered by a search term:
|
|
@@ -264,7 +266,7 @@ at. The app must be closed either way, and `undo` puts the old pointer back.
|
|
|
264
266
|
**`new-row`** - create a sidebar row for a conversation that has none:
|
|
265
267
|
|
|
266
268
|
```
|
|
267
|
-
ccs new-row --to
|
|
269
|
+
ccs new-row --to a1b2c3d4-e5f6-4a7b-8c9d-e0f1a2b3c4d5
|
|
268
270
|
```
|
|
269
271
|
|
|
270
272
|
`repoint` moves an existing row and `sync` copies one between accounts; neither can
|
|
@@ -618,11 +620,13 @@ window.
|
|
|
618
620
|
| Platform | Status | Mutations |
|
|
619
621
|
|---|---|---|
|
|
620
622
|
| Windows 11 + Claude Desktop | verified 2026-07-31; sync end-to-end incl. live continuation 2026-08-03; sync-undo drift refusal live 2026-08-04 | read-only + mutations |
|
|
623
|
+
| Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store | read-only + mutations, `--apply` not yet verified on this build |
|
|
621
624
|
| macOS / Linux desktop | unverified | **read-only only** - desktop-store mutations refuse, with no override |
|
|
622
625
|
| CLI-only sessions (any OS) | transcript layout verified | mutations allowed via `--transcript-only` |
|
|
623
626
|
|
|
624
627
|
The dates are measurements against specific builds - most recently Claude Desktop
|
|
625
|
-
2.42304.
|
|
628
|
+
2.42304.2.0 (Microsoft Store MSIX install) for the read-only paths, and 2.42304.0.0 with Claude
|
|
629
|
+
Code CLI 2.1.302 for the last `--apply` runs. On 2026-10-14 the
|
|
626
630
|
read-only paths were verified against a real Desktop 2.x store, build 2.37337.1.0 (`list`,
|
|
627
631
|
`doctor`, `alignment`, `recover` without `--apply`, and a `converge` dry run): the row format
|
|
628
632
|
had only gained keys, and every row this tool had written under 1.x was still present, some
|
|
@@ -631,6 +635,17 @@ already re-saved by the new app. Later on 2026-10-14 a `converge --apply` ran ag
|
|
|
631
635
|
present and `doctor` reported none vanished. On 2026-10-17 two more `converge --apply` runs
|
|
632
636
|
went through against 2.42304.0.0 with the app closed, with the same result. That build moved
|
|
633
637
|
the browser-extension helper, which the running-app guard follows from 0.16.1 (see Roadmap).
|
|
638
|
+
On 2026-10-19, 2.42304.2.0 moved the app's whole user-data folder, session store included,
|
|
639
|
+
from the package's `LocalCache\Roaming\Claude` to `%LOCALAPPDATA%\Claude-Data\claude_<pkg>`,
|
|
640
|
+
and left the helper where it was. 0.16.2 and earlier report the store absent there. 0.16.3
|
|
641
|
+
reads the moved store and still finds the helper. Its read-only paths (`doctor`, `alignment`
|
|
642
|
+
and a `converge` dry run) were checked against the moved store the same day. One known gap
|
|
643
|
+
remains: rows this tool created before the move are journaled at their old paths, so `doctor`
|
|
644
|
+
reports them as "no longer on disk" although they moved with the store. Those lines are false,
|
|
645
|
+
and so is their advice to run `new-row` again. For the same reason `undo` cannot reverse an
|
|
646
|
+
operation from before the move, and changes nothing trying: it refuses a `converge`, refuses a
|
|
647
|
+
`new-row` with "that row is already gone", which is false too, and fails with an unhandled
|
|
648
|
+
error on a `move`.
|
|
634
649
|
The on-disk format has changed once already during this tool's own development; treat a
|
|
635
650
|
much newer build as unverified territory.
|
|
636
651
|
|
|
@@ -980,6 +995,35 @@ window into `journaled`/`completed`/`rolled_back`.
|
|
|
980
995
|
what a desktop session leaves behind when it enters a second worktree and is later
|
|
981
996
|
resumed in the one its sidebar row remembers: the copy in the second worktree stops
|
|
982
997
|
growing. It was seen twice on one machine in October.
|
|
998
|
+
- **0.16.2 puts the fake cast everywhere, and the helper note on every refusal.** Examples and
|
|
999
|
+
fixtures across the repository, from the README to the tests and the check scripts, now use
|
|
1000
|
+
the project's fake cast rather than values read off the maintainer's own machine: account and
|
|
1001
|
+
store identifiers, an address, a Windows user path, a sidebar title and a session id. Nothing
|
|
1002
|
+
about what the tool reads, writes or refuses changed, and every test still exercises what it
|
|
1003
|
+
did. Also, from the browser-helper work: the note that explains a counted helper now appears
|
|
1004
|
+
on `move`, `undo` and `recover` refusals, not only on converge and sync; when the signed-helper
|
|
1005
|
+
opt-in is on and a helper still counts, the note says why; and the opt-in's wording, in this
|
|
1006
|
+
README, the window's checkbox and the command's help, says Chrome or Edge, since either can
|
|
1007
|
+
launch the helper. Three constants left unused since the August anti-fabrication fix are gone.
|
|
1008
|
+
- **0.16.3 finds the session store where Desktop 2.42304.2 moved it.** That build moved the
|
|
1009
|
+
app's whole user-data folder out of the Microsoft Store package, from
|
|
1010
|
+
`%LOCALAPPDATA%\Packages\Claude_<pkg>\LocalCache\Roaming\Claude` to
|
|
1011
|
+
`%LOCALAPPDATA%\Claude-Data\claude_<pkg>`, and left a 0-byte file in the old folder naming
|
|
1012
|
+
the new one. 0.16.2 looked only in the old places, so `doctor` and `alignment` reported no
|
|
1013
|
+
store and `converge` planned nothing. This release reads the moved folder first. When a
|
|
1014
|
+
package's data has moved, its old folder is never read, and a store still sitting there is
|
|
1015
|
+
named in `doctor` rather than read beside the new one. The browser-extension helper did not
|
|
1016
|
+
move. The guard had found it two folders up from the store, which after the move is a
|
|
1017
|
+
folder where no helper lives, so the signed-helper opt-in would have stopped working again
|
|
1018
|
+
without a word. Helper locations are now listed on their own. The CLI the app spawns moved
|
|
1019
|
+
with the data, and the guard recognises it at its new path. When no store is found and the
|
|
1020
|
+
app's marker names a folder this version does not know, `doctor` says so instead of
|
|
1021
|
+
reporting a bare absence, and when the folder the data moved to cannot be listed, discovery
|
|
1022
|
+
reports an error rather than falling back to the old store. One gap remains: operations
|
|
1023
|
+
from before the move are journaled at their old paths, so `doctor` lists the rows this tool
|
|
1024
|
+
created then as no longer on disk although they moved with the store, and `undo` cannot
|
|
1025
|
+
reverse those operations (see the compatibility matrix). Until that is fixed, judge a
|
|
1026
|
+
`converge --apply` by `alignment`.
|
|
983
1027
|
- Platform rows above move from "unverified" to "verified" as contributors confirm the store
|
|
984
1028
|
paths and behavior on their own machines.
|
|
985
1029
|
|
|
@@ -94,13 +94,14 @@ of the standard library, but packaged separately on some Linux distributions
|
|
|
94
94
|
replaces the old "Claude session sync" shortcut with the new name; a taskbar pin to the
|
|
95
95
|
old name is yours to re-pin.
|
|
96
96
|
|
|
97
|
-
## Optional: let
|
|
98
|
-
|
|
99
|
-
The desktop app ships a helper for its
|
|
100
|
-
guard counts it and you have to close
|
|
101
|
-
excluded - but **the helper auto-updates every few days**
|
|
102
|
-
measured), and each update correctly fails closed, so the
|
|
103
|
-
be closed again until someone re-runs the ~12-minute
|
|
97
|
+
## Optional: let the browser stay open
|
|
98
|
+
|
|
99
|
+
The desktop app ships a helper for its browser extension that Chrome or Edge launches and
|
|
100
|
+
that can outlive the app, so the guard counts it and you have to close that browser too. A
|
|
101
|
+
measured build of that helper is excluded - but **the helper auto-updates every few days**
|
|
102
|
+
(three builds in eight days, as measured), and each update correctly fails closed, so the
|
|
103
|
+
exclusion lapses and the browser has to be closed again until someone re-runs the ~12-minute
|
|
104
|
+
measurement in
|
|
104
105
|
`tools/native-host-measurement/`.
|
|
105
106
|
|
|
106
107
|
If that trade isn't worth it to you, there is an opt-in: trust **any** helper at the app's
|
|
@@ -112,7 +113,7 @@ claude-code-sessions trust-signed-helper --on # enable it
|
|
|
112
113
|
claude-code-sessions trust-signed-helper --off # back to the default
|
|
113
114
|
```
|
|
114
115
|
|
|
115
|
-
In the window it is the **"Let
|
|
116
|
+
In the window it is the **"Let browsers stay open"** checkbox, which explains the trade before
|
|
116
117
|
enabling anything. Either way it is one marker file - `~/.claude-code-journal/trust-signed-helper` -
|
|
117
118
|
so deleting that file also revokes it.
|
|
118
119
|
|
|
@@ -145,9 +146,10 @@ files agree - or, if you know which account the desktop app is signed into, asse
|
|
|
145
146
|
full). Signing into the desktop app does **not** refresh `~/.claude.json`. A refused
|
|
146
147
|
`sync --apply` (desktop app running) refuses **before anything is journaled** - no lock file,
|
|
147
148
|
no op directory - so there's nothing left behind for `doctor` to flag or `recover` to clean up;
|
|
148
|
-
just close the app and re-run. One measured wrinkle: the desktop ships a helper for its
|
|
149
|
-
extension (`chrome-native-host.exe`)
|
|
150
|
-
if a refusal names that process, fully exit
|
|
149
|
+
just close the app and re-run. One measured wrinkle: the desktop ships a helper for its browser
|
|
150
|
+
extension (`chrome-native-host.exe`), launched by Chrome or Edge, that can outlive the app
|
|
151
|
+
itself, and the guard counts it - if a refusal names that process, fully exit that browser too
|
|
152
|
+
(the refusal always names what it saw).
|
|
151
153
|
|
|
152
154
|
## Usage
|
|
153
155
|
|
|
@@ -162,7 +164,7 @@ alone cannot tell you which copy answered:
|
|
|
162
164
|
```
|
|
163
165
|
$ ccs --version
|
|
164
166
|
claude-code-sessions 0.10.0
|
|
165
|
-
running from: C:\Users
|
|
167
|
+
running from: C:\Users\<name>\AppData\Local\pipx\...\claude_code_sessions.py
|
|
166
168
|
```
|
|
167
169
|
|
|
168
170
|
**`list`** - inventory sessions, optionally filtered by a search term:
|
|
@@ -234,7 +236,7 @@ at. The app must be closed either way, and `undo` puts the old pointer back.
|
|
|
234
236
|
**`new-row`** - create a sidebar row for a conversation that has none:
|
|
235
237
|
|
|
236
238
|
```
|
|
237
|
-
ccs new-row --to
|
|
239
|
+
ccs new-row --to a1b2c3d4-e5f6-4a7b-8c9d-e0f1a2b3c4d5
|
|
238
240
|
```
|
|
239
241
|
|
|
240
242
|
`repoint` moves an existing row and `sync` copies one between accounts; neither can
|
|
@@ -588,11 +590,13 @@ window.
|
|
|
588
590
|
| Platform | Status | Mutations |
|
|
589
591
|
|---|---|---|
|
|
590
592
|
| Windows 11 + Claude Desktop | verified 2026-07-31; sync end-to-end incl. live continuation 2026-08-03; sync-undo drift refusal live 2026-08-04 | read-only + mutations |
|
|
593
|
+
| Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store | read-only + mutations, `--apply` not yet verified on this build |
|
|
591
594
|
| macOS / Linux desktop | unverified | **read-only only** - desktop-store mutations refuse, with no override |
|
|
592
595
|
| CLI-only sessions (any OS) | transcript layout verified | mutations allowed via `--transcript-only` |
|
|
593
596
|
|
|
594
597
|
The dates are measurements against specific builds - most recently Claude Desktop
|
|
595
|
-
2.42304.
|
|
598
|
+
2.42304.2.0 (Microsoft Store MSIX install) for the read-only paths, and 2.42304.0.0 with Claude
|
|
599
|
+
Code CLI 2.1.302 for the last `--apply` runs. On 2026-10-14 the
|
|
596
600
|
read-only paths were verified against a real Desktop 2.x store, build 2.37337.1.0 (`list`,
|
|
597
601
|
`doctor`, `alignment`, `recover` without `--apply`, and a `converge` dry run): the row format
|
|
598
602
|
had only gained keys, and every row this tool had written under 1.x was still present, some
|
|
@@ -601,6 +605,17 @@ already re-saved by the new app. Later on 2026-10-14 a `converge --apply` ran ag
|
|
|
601
605
|
present and `doctor` reported none vanished. On 2026-10-17 two more `converge --apply` runs
|
|
602
606
|
went through against 2.42304.0.0 with the app closed, with the same result. That build moved
|
|
603
607
|
the browser-extension helper, which the running-app guard follows from 0.16.1 (see Roadmap).
|
|
608
|
+
On 2026-10-19, 2.42304.2.0 moved the app's whole user-data folder, session store included,
|
|
609
|
+
from the package's `LocalCache\Roaming\Claude` to `%LOCALAPPDATA%\Claude-Data\claude_<pkg>`,
|
|
610
|
+
and left the helper where it was. 0.16.2 and earlier report the store absent there. 0.16.3
|
|
611
|
+
reads the moved store and still finds the helper. Its read-only paths (`doctor`, `alignment`
|
|
612
|
+
and a `converge` dry run) were checked against the moved store the same day. One known gap
|
|
613
|
+
remains: rows this tool created before the move are journaled at their old paths, so `doctor`
|
|
614
|
+
reports them as "no longer on disk" although they moved with the store. Those lines are false,
|
|
615
|
+
and so is their advice to run `new-row` again. For the same reason `undo` cannot reverse an
|
|
616
|
+
operation from before the move, and changes nothing trying: it refuses a `converge`, refuses a
|
|
617
|
+
`new-row` with "that row is already gone", which is false too, and fails with an unhandled
|
|
618
|
+
error on a `move`.
|
|
604
619
|
The on-disk format has changed once already during this tool's own development; treat a
|
|
605
620
|
much newer build as unverified territory.
|
|
606
621
|
|
|
@@ -950,6 +965,35 @@ window into `journaled`/`completed`/`rolled_back`.
|
|
|
950
965
|
what a desktop session leaves behind when it enters a second worktree and is later
|
|
951
966
|
resumed in the one its sidebar row remembers: the copy in the second worktree stops
|
|
952
967
|
growing. It was seen twice on one machine in October.
|
|
968
|
+
- **0.16.2 puts the fake cast everywhere, and the helper note on every refusal.** Examples and
|
|
969
|
+
fixtures across the repository, from the README to the tests and the check scripts, now use
|
|
970
|
+
the project's fake cast rather than values read off the maintainer's own machine: account and
|
|
971
|
+
store identifiers, an address, a Windows user path, a sidebar title and a session id. Nothing
|
|
972
|
+
about what the tool reads, writes or refuses changed, and every test still exercises what it
|
|
973
|
+
did. Also, from the browser-helper work: the note that explains a counted helper now appears
|
|
974
|
+
on `move`, `undo` and `recover` refusals, not only on converge and sync; when the signed-helper
|
|
975
|
+
opt-in is on and a helper still counts, the note says why; and the opt-in's wording, in this
|
|
976
|
+
README, the window's checkbox and the command's help, says Chrome or Edge, since either can
|
|
977
|
+
launch the helper. Three constants left unused since the August anti-fabrication fix are gone.
|
|
978
|
+
- **0.16.3 finds the session store where Desktop 2.42304.2 moved it.** That build moved the
|
|
979
|
+
app's whole user-data folder out of the Microsoft Store package, from
|
|
980
|
+
`%LOCALAPPDATA%\Packages\Claude_<pkg>\LocalCache\Roaming\Claude` to
|
|
981
|
+
`%LOCALAPPDATA%\Claude-Data\claude_<pkg>`, and left a 0-byte file in the old folder naming
|
|
982
|
+
the new one. 0.16.2 looked only in the old places, so `doctor` and `alignment` reported no
|
|
983
|
+
store and `converge` planned nothing. This release reads the moved folder first. When a
|
|
984
|
+
package's data has moved, its old folder is never read, and a store still sitting there is
|
|
985
|
+
named in `doctor` rather than read beside the new one. The browser-extension helper did not
|
|
986
|
+
move. The guard had found it two folders up from the store, which after the move is a
|
|
987
|
+
folder where no helper lives, so the signed-helper opt-in would have stopped working again
|
|
988
|
+
without a word. Helper locations are now listed on their own. The CLI the app spawns moved
|
|
989
|
+
with the data, and the guard recognises it at its new path. When no store is found and the
|
|
990
|
+
app's marker names a folder this version does not know, `doctor` says so instead of
|
|
991
|
+
reporting a bare absence, and when the folder the data moved to cannot be listed, discovery
|
|
992
|
+
reports an error rather than falling back to the old store. One gap remains: operations
|
|
993
|
+
from before the move are journaled at their old paths, so `doctor` lists the rows this tool
|
|
994
|
+
created then as no longer on disk although they moved with the store, and `undo` cannot
|
|
995
|
+
reverse those operations (see the compatibility matrix). Until that is fixed, judge a
|
|
996
|
+
`converge --apply` by `alignment`.
|
|
953
997
|
- Platform rows above move from "unverified" to "verified" as contributors confirm the store
|
|
954
998
|
paths and behavior on their own machines.
|
|
955
999
|
|
{claude_code_sessions-0.16.1 → claude_code_sessions-0.16.3}/claude_code_sessions.egg-info/PKG-INFO
RENAMED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: claude-code-sessions
|
|
3
|
-
Version: 0.16.
|
|
3
|
+
Version: 0.16.3
|
|
4
4
|
Summary: Unofficial tool for Claude Code sessions on disk: make a session visible under your other Claude account on the same machine, or move one between project folders. Command line or GUI. Journaled, reversible, fails closed.
|
|
5
5
|
Author: Craig Stoller
|
|
6
6
|
License: MIT
|
|
@@ -124,13 +124,14 @@ of the standard library, but packaged separately on some Linux distributions
|
|
|
124
124
|
replaces the old "Claude session sync" shortcut with the new name; a taskbar pin to the
|
|
125
125
|
old name is yours to re-pin.
|
|
126
126
|
|
|
127
|
-
## Optional: let
|
|
127
|
+
## Optional: let the browser stay open
|
|
128
128
|
|
|
129
|
-
The desktop app ships a helper for its
|
|
130
|
-
guard counts it and you have to close
|
|
131
|
-
excluded - but **the helper auto-updates every few days**
|
|
132
|
-
measured), and each update correctly fails closed, so the
|
|
133
|
-
be closed again until someone re-runs the ~12-minute
|
|
129
|
+
The desktop app ships a helper for its browser extension that Chrome or Edge launches and
|
|
130
|
+
that can outlive the app, so the guard counts it and you have to close that browser too. A
|
|
131
|
+
measured build of that helper is excluded - but **the helper auto-updates every few days**
|
|
132
|
+
(three builds in eight days, as measured), and each update correctly fails closed, so the
|
|
133
|
+
exclusion lapses and the browser has to be closed again until someone re-runs the ~12-minute
|
|
134
|
+
measurement in
|
|
134
135
|
`tools/native-host-measurement/`.
|
|
135
136
|
|
|
136
137
|
If that trade isn't worth it to you, there is an opt-in: trust **any** helper at the app's
|
|
@@ -142,7 +143,7 @@ claude-code-sessions trust-signed-helper --on # enable it
|
|
|
142
143
|
claude-code-sessions trust-signed-helper --off # back to the default
|
|
143
144
|
```
|
|
144
145
|
|
|
145
|
-
In the window it is the **"Let
|
|
146
|
+
In the window it is the **"Let browsers stay open"** checkbox, which explains the trade before
|
|
146
147
|
enabling anything. Either way it is one marker file - `~/.claude-code-journal/trust-signed-helper` -
|
|
147
148
|
so deleting that file also revokes it.
|
|
148
149
|
|
|
@@ -175,9 +176,10 @@ files agree - or, if you know which account the desktop app is signed into, asse
|
|
|
175
176
|
full). Signing into the desktop app does **not** refresh `~/.claude.json`. A refused
|
|
176
177
|
`sync --apply` (desktop app running) refuses **before anything is journaled** - no lock file,
|
|
177
178
|
no op directory - so there's nothing left behind for `doctor` to flag or `recover` to clean up;
|
|
178
|
-
just close the app and re-run. One measured wrinkle: the desktop ships a helper for its
|
|
179
|
-
extension (`chrome-native-host.exe`)
|
|
180
|
-
if a refusal names that process, fully exit
|
|
179
|
+
just close the app and re-run. One measured wrinkle: the desktop ships a helper for its browser
|
|
180
|
+
extension (`chrome-native-host.exe`), launched by Chrome or Edge, that can outlive the app
|
|
181
|
+
itself, and the guard counts it - if a refusal names that process, fully exit that browser too
|
|
182
|
+
(the refusal always names what it saw).
|
|
181
183
|
|
|
182
184
|
## Usage
|
|
183
185
|
|
|
@@ -192,7 +194,7 @@ alone cannot tell you which copy answered:
|
|
|
192
194
|
```
|
|
193
195
|
$ ccs --version
|
|
194
196
|
claude-code-sessions 0.10.0
|
|
195
|
-
running from: C:\Users
|
|
197
|
+
running from: C:\Users\<name>\AppData\Local\pipx\...\claude_code_sessions.py
|
|
196
198
|
```
|
|
197
199
|
|
|
198
200
|
**`list`** - inventory sessions, optionally filtered by a search term:
|
|
@@ -264,7 +266,7 @@ at. The app must be closed either way, and `undo` puts the old pointer back.
|
|
|
264
266
|
**`new-row`** - create a sidebar row for a conversation that has none:
|
|
265
267
|
|
|
266
268
|
```
|
|
267
|
-
ccs new-row --to
|
|
269
|
+
ccs new-row --to a1b2c3d4-e5f6-4a7b-8c9d-e0f1a2b3c4d5
|
|
268
270
|
```
|
|
269
271
|
|
|
270
272
|
`repoint` moves an existing row and `sync` copies one between accounts; neither can
|
|
@@ -618,11 +620,13 @@ window.
|
|
|
618
620
|
| Platform | Status | Mutations |
|
|
619
621
|
|---|---|---|
|
|
620
622
|
| Windows 11 + Claude Desktop | verified 2026-07-31; sync end-to-end incl. live continuation 2026-08-03; sync-undo drift refusal live 2026-08-04 | read-only + mutations |
|
|
623
|
+
| Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store | read-only + mutations, `--apply` not yet verified on this build |
|
|
621
624
|
| macOS / Linux desktop | unverified | **read-only only** - desktop-store mutations refuse, with no override |
|
|
622
625
|
| CLI-only sessions (any OS) | transcript layout verified | mutations allowed via `--transcript-only` |
|
|
623
626
|
|
|
624
627
|
The dates are measurements against specific builds - most recently Claude Desktop
|
|
625
|
-
2.42304.
|
|
628
|
+
2.42304.2.0 (Microsoft Store MSIX install) for the read-only paths, and 2.42304.0.0 with Claude
|
|
629
|
+
Code CLI 2.1.302 for the last `--apply` runs. On 2026-10-14 the
|
|
626
630
|
read-only paths were verified against a real Desktop 2.x store, build 2.37337.1.0 (`list`,
|
|
627
631
|
`doctor`, `alignment`, `recover` without `--apply`, and a `converge` dry run): the row format
|
|
628
632
|
had only gained keys, and every row this tool had written under 1.x was still present, some
|
|
@@ -631,6 +635,17 @@ already re-saved by the new app. Later on 2026-10-14 a `converge --apply` ran ag
|
|
|
631
635
|
present and `doctor` reported none vanished. On 2026-10-17 two more `converge --apply` runs
|
|
632
636
|
went through against 2.42304.0.0 with the app closed, with the same result. That build moved
|
|
633
637
|
the browser-extension helper, which the running-app guard follows from 0.16.1 (see Roadmap).
|
|
638
|
+
On 2026-10-19, 2.42304.2.0 moved the app's whole user-data folder, session store included,
|
|
639
|
+
from the package's `LocalCache\Roaming\Claude` to `%LOCALAPPDATA%\Claude-Data\claude_<pkg>`,
|
|
640
|
+
and left the helper where it was. 0.16.2 and earlier report the store absent there. 0.16.3
|
|
641
|
+
reads the moved store and still finds the helper. Its read-only paths (`doctor`, `alignment`
|
|
642
|
+
and a `converge` dry run) were checked against the moved store the same day. One known gap
|
|
643
|
+
remains: rows this tool created before the move are journaled at their old paths, so `doctor`
|
|
644
|
+
reports them as "no longer on disk" although they moved with the store. Those lines are false,
|
|
645
|
+
and so is their advice to run `new-row` again. For the same reason `undo` cannot reverse an
|
|
646
|
+
operation from before the move, and changes nothing trying: it refuses a `converge`, refuses a
|
|
647
|
+
`new-row` with "that row is already gone", which is false too, and fails with an unhandled
|
|
648
|
+
error on a `move`.
|
|
634
649
|
The on-disk format has changed once already during this tool's own development; treat a
|
|
635
650
|
much newer build as unverified territory.
|
|
636
651
|
|
|
@@ -980,6 +995,35 @@ window into `journaled`/`completed`/`rolled_back`.
|
|
|
980
995
|
what a desktop session leaves behind when it enters a second worktree and is later
|
|
981
996
|
resumed in the one its sidebar row remembers: the copy in the second worktree stops
|
|
982
997
|
growing. It was seen twice on one machine in October.
|
|
998
|
+
- **0.16.2 puts the fake cast everywhere, and the helper note on every refusal.** Examples and
|
|
999
|
+
fixtures across the repository, from the README to the tests and the check scripts, now use
|
|
1000
|
+
the project's fake cast rather than values read off the maintainer's own machine: account and
|
|
1001
|
+
store identifiers, an address, a Windows user path, a sidebar title and a session id. Nothing
|
|
1002
|
+
about what the tool reads, writes or refuses changed, and every test still exercises what it
|
|
1003
|
+
did. Also, from the browser-helper work: the note that explains a counted helper now appears
|
|
1004
|
+
on `move`, `undo` and `recover` refusals, not only on converge and sync; when the signed-helper
|
|
1005
|
+
opt-in is on and a helper still counts, the note says why; and the opt-in's wording, in this
|
|
1006
|
+
README, the window's checkbox and the command's help, says Chrome or Edge, since either can
|
|
1007
|
+
launch the helper. Three constants left unused since the August anti-fabrication fix are gone.
|
|
1008
|
+
- **0.16.3 finds the session store where Desktop 2.42304.2 moved it.** That build moved the
|
|
1009
|
+
app's whole user-data folder out of the Microsoft Store package, from
|
|
1010
|
+
`%LOCALAPPDATA%\Packages\Claude_<pkg>\LocalCache\Roaming\Claude` to
|
|
1011
|
+
`%LOCALAPPDATA%\Claude-Data\claude_<pkg>`, and left a 0-byte file in the old folder naming
|
|
1012
|
+
the new one. 0.16.2 looked only in the old places, so `doctor` and `alignment` reported no
|
|
1013
|
+
store and `converge` planned nothing. This release reads the moved folder first. When a
|
|
1014
|
+
package's data has moved, its old folder is never read, and a store still sitting there is
|
|
1015
|
+
named in `doctor` rather than read beside the new one. The browser-extension helper did not
|
|
1016
|
+
move. The guard had found it two folders up from the store, which after the move is a
|
|
1017
|
+
folder where no helper lives, so the signed-helper opt-in would have stopped working again
|
|
1018
|
+
without a word. Helper locations are now listed on their own. The CLI the app spawns moved
|
|
1019
|
+
with the data, and the guard recognises it at its new path. When no store is found and the
|
|
1020
|
+
app's marker names a folder this version does not know, `doctor` says so instead of
|
|
1021
|
+
reporting a bare absence, and when the folder the data moved to cannot be listed, discovery
|
|
1022
|
+
reports an error rather than falling back to the old store. One gap remains: operations
|
|
1023
|
+
from before the move are journaled at their old paths, so `doctor` lists the rows this tool
|
|
1024
|
+
created then as no longer on disk although they moved with the store, and `undo` cannot
|
|
1025
|
+
reverse those operations (see the compatibility matrix). Until that is fixed, judge a
|
|
1026
|
+
`converge --apply` by `alignment`.
|
|
983
1027
|
- Platform rows above move from "unverified" to "verified" as contributors confirm the store
|
|
984
1028
|
paths and behavior on their own machines.
|
|
985
1029
|
|