Agent mode claimed to be making edits, did not make edits

[333fred]

Type: Bug

Agent mode repeatedly claimed to be making edits to a function in the file. It repeatedly failed to do so, recognized that it failed, and then failed again.

Extension version: 0.26.2025040204 VS Code version: Code 1.99.3 (17baf841131aa23349f217ca7c570c76ee87b957, 2025-04-15T23:18:46.076Z) OS version: Linux x64 6.8.0-58-generic Modes:

System Info

Item	Value
CPUs	AMD Ryzen 9 5950X 16-Core Processor (32 x 4674)
GPU Status	2d_canvas: enabled canvas_oop_rasterization: enabled_on direct_rendering_display_compositor: disabled_off_ok gpu_compositing: enabled multiple_raster_threads: enabled_on opengl: enabled_on rasterization: enabled raw_draw: disabled_off_ok skia_graphite: disabled_off video_decode: enabled video_encode: disabled_software vulkan: disabled_off webgl: enabled webgl2: enabled webgpu: disabled_off webnn: disabled_off
Load (avg)	1, 1, 3
Memory (System)	62.70GB (37.25GB free)
Process Argv	--crash-reporter-id c9f05a69-d019-464b-b1f2-b43e5fb75046
Screen Reader	no
VM	0%
DESKTOP_SESSION	regolith-x11
XDG_CURRENT_DESKTOP	Regolith:GNOME-Flashback:GNOME
XDG_SESSION_DESKTOP	regolith-x11
XDG_SESSION_TYPE	x11

A/B Experiments

vsliv368cf:30146710
vspor879:30202332
vspor708:30202333
vspor363:30204092
vscod805cf:30301675
binariesv615:30325510
14424c2-chatv2:31289885
c4g48928:30535728
azure-dev_surveyone:30548225
2i9eh265:30646982
962ge761:30959799
2e7ec940:31000449
pythontbext0:30879054
cppperfnew:31000557
dwnewjupyter:31046869
pythonrstrctxt:31112756
nativeloc1:31192215
5fd0e150:31155592
dwcopilot:31170013
6074i472:31201624
dwoutputs:31242946
customenabled:31248079
hdaa2157:31222309
copilot_t_ci:31222730
e5gg6876:31282496
pythoneinst12:31285622
bgtreat:31268568
4gafe986:31271826
31787653:31262186
3e8i5726:31271747
996jf627:31283433
useunpkgapi:31288273
7bj51361:31289155
747dc170:31275177
aj496949:31278748
aj953862:31281341

[333fred]

Here are the screenshots from the interaction. None of these edits that it talked about actually appeared. Additionally, rather than editing all of the tests in the file in the same edit, it persisted in doing a single method at a time, taking quite a while before I canceled it.

[p3nGu1nZz]

happens on 25% of my requests. at least put in a simple check to see if the file was actually edited so you do not charge us for the failed request.

[roblourens]

Which model is this? Can you check for errors in the Copilot Chat output channel?

[333fred]

This was with Sonnet 3.7. Far too much time has passed for me to be able to say anything about errors.

[roblourens]

Thanks. Sometimes this kind of loop happens due to an error in using the tool. We may have fixed an issue with tool calling for Claude in the meantime (server-side or in Insiders/this week's release). Sometimes it happens because the model is not smart. If you see it again, would be helpful if you can see whether it ran into an error.

[333fred]

It only did the first bullet of these edits. There was nothing that I could see in the chat log for errors:

2025-05-06 14:20:56.049 [info] Fetched model metadata in 624ms 63c19b26-ada9-40d8-913f-4b2b38407ea5
2025-05-06 14:20:56.663 [info] Fetched content exclusion rules in 341ms
2025-05-06 14:21:05.841 [info] message 0 returned. finish reason: [stop]
2025-05-06 14:21:05.841 [info] request done: requestId: [8025ce08-01f6-4634-861c-e9e802b4dc05] model deployment ID: []
2025-05-06 14:21:08.152 [info] message 0 returned. finish reason: [stop]
2025-05-06 14:21:08.152 [info] request done: requestId: [aee7b818-efcc-4a47-8f5c-277889d84eae] model deployment ID: []
2025-05-06 14:21:08.156 [info] srequest done: chatRequestId: [8025ce08-01f6-4634-861c-e9e802b4dc05], speculationRequestId: [8c2662c4-9805-4df5-9d34-6a10b8592bf2]

[roblourens]

Usually Claude is smarter than this ☹

[vs-code-engineering[bot]]

This issue has been closed automatically because it needs more information and has not had recent activity. See also our issue reporting guidelines.

Happy Coding!