Agent defaults to partial scope execution on multi-file operations, producing incomplete migrations, regressions, and wasted compute time

[wendall911]

Type: Bug

Description

When asked to perform operations across multiple files (e.g., "migrate all components from $app/stores to $app/state"), the GitHub Copilot agent consistently executes on a random subset of the targets instead of all of them. This violates basic English language semantics — "migrate all" means ALL, not "some."

This is not a minor UX issue. It produces broken intermediate states, downstream regressions, and requires correction iterations that consume significant additional compute time and user effort.

---

Concrete Example: SvelteKit + Tailwind Migration (this session)

Request

User asked for a complete migration: upgrade to pnpm, latest SvelteKit/Svelte 5/Tailwind 4, and migrate all components.

What the agent did (Session 1 — committed as bbfe217)

The agent migrated some of the 4 affected Svelte components but not all:

File	What was done	What was skipped
mode-toggle.svelte	Updated $mode → mode.current	Did NOT update on:click → onclick
main-nav.svelte	Removed $: invert reactive; added CSS class	Did NOT migrate $app/stores → $app/state
nav.svelte	Fixed self-closing div syntax	Did NOT migrate $app/stores → $app/state
contacts.svelte	Fixed self-closing div syntax	Did NOT migrate $: → $derived

Resulting regression

The nav.svelte partial migration was the most harmful: the agent fixed the div syntax (/> → ></div>) so the green loading bar div now renders, but left the store import as $app/stores. When Session 2 completed the migration to $app/state, navigating became an object (always truthy) instead of a nullable value — making the green progress bar permanently visible at idle.

This regression required additional investigation and a specific fix: checking navigating.to instead of navigating for truthiness.

Session 1 chat history was also lost

When the Copilot Chat panel was closed, the entire session history was discarded. VS Code's chat session storage only persists single-request snapshots, not full multi-turn conversations. The user lost approximately one hour of context and had to reconstruct the migration plan from scratch by inspecting the git commit diff.

---

Root Cause Analysis

1. Partial scope execution: The agent defaults to executing on a subset of targets rather than all targets. No natural language interpretation of "migrate all components" should mean "migrate some components."

2. No scope discovery step: The agent should grep for all affected files first, display the complete target list, then execute. Instead, it appears to guess or assume a subset.

3. No completion verification: After partial execution, the agent did not verify that all targets had been migrated (e.g., by checking for remaining $app/stores imports).

4. Session history not persisted: Full multi-turn conversations are not saved anywhere recoverable. The existing chatSessions/*.json files only store one request per session entry. A long agentic session is unrecoverable if the panel is closed or the window crashes.

---

Expected Behavior

For any multi-target operation:

1. Discover all targets (e.g., grep -r "$app/stores" src/)
2. Display the complete list: "Found 4 files needing migration: nav.svelte, main-nav.svelte, contacts.svelte, mode-toggle.svelte"
3. Confirm with user if scope is large
4. Execute on ALL discovered targets
5. Verify completion (re-run the grep, confirm 0 matches)

The default assumption must be ALL unless the user explicitly limits scope.

---

Impact

• Incomplete migrations create broken intermediate states
• Regressions require additional debugging and correction
• Each correction iteration consumes additional compute tokens
• Lost session history means rediscovering context from scratch

In this case: ~2 hours of total session time was consumed correcting issues caused by the agent's partial execution and the resulting regression.

---

Session Transcript

Full readable conversation transcript (user + assistant messages):
https://dpaste.com/5FXBDBLCW

Raw JSONL transcript (local path):
/home/wendallc/.config/Code/User/workspaceStorage/856672e5688144dbee927bed322ce2a8/GitHub.copilot-chat/transcripts/10523a21-d72e-4680-892c-c63d6db3096e.jsonl

---

Related Issues / Workarounds

The user was forced to create a global guardrails instruction file at:
/home/wendallc/.config/Code/User/prompts/global-guardrails.instructions.md

This file explicitly instructs the agent to perform scope discovery before executing multi-target operations. The fact that such a workaround is necessary to get correct behavior underscores that this is a product defect, not a user configuration issue.

Extension version: 0.45.1
VS Code version: Code 1.116.0 (560a9db, 2026-04-15T00:28:13Z)
OS version: Linux x64 6.19.12-200.fc43.x86_64
Modes:

Logs

Trace: [messagesAPI]SSE: {"delta":{"partial_json":", \"timeou","type":"input_json_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"t\": 20000}","type":"input_json_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"index":0,"type":"content_block_stop"}
Trace: [messagesAPI]SSE: {"copilot_usage":{"token_details":null,"total_nano_aiu":0},"delta":{"stop_reason":"tool_use","stop_sequence":null},"type":"message_delta","usage":{"cache_creation_input_tokens":321,"cache_read_input_tokens":106371,"input_tokens":1,"output_tokens":216}}
Trace: [messagesAPI]SSE: {"amazon-bedrock-invocationMetrics":{"cacheReadInputTokenCount":106371,"cacheWriteInputTokenCount":321,"firstByteLatency":1431,"inputTokenCount":1,"invocationLatency":3542,"outputTokenCount":216},"type":"message_stop"}
Info: [messagesAPI] message 0 returned. finish reason: [stop]
Trace: [messagesAPI]SSE: [DONE]
Info: ccreq:1aced064.copilotmd | success | claude-sonnet-4.6 -> claude-sonnet-4-6 | 5665ms | [panel/editAgent]
Trace: Resolving chat model
Debug: [Agent] rendering with budget=105227 (baseBudget: 127997, toolTokens: 11078, totalTools: 49, toolSearchEnabled: true), summarizationEnabled=true
Trace: Resolving chat model
Trace: Resolving chat model
Trace: Resolving chat model
Trace: Built prompt
Trace: Sending prompt to model
Warning: Tool click_element failed validation: object has unsupported top-level schema keyword 'oneOf'
Warning: Tool drag_element failed validation: object has unsupported top-level schema keyword 'allOf'
Warning: Tool hover_element failed validation: object has unsupported top-level schema keyword 'oneOf'
Warning: Tool run_playwright_code failed validation: object has unsupported top-level schema keyword 'oneOf'
Warning: Tool type_in_page failed validation: object has unsupported top-level schema keyword 'oneOf'
Debug: [PowerService] Acquired power save blocker, active count: 1
Debug: modelMaxPromptTokens 127997
Debug: modelMaxResponseTokens 32000
Debug: chat model claude-sonnet-4.6
Debug: GH request id: 23E3:125C4D:138766:157866:6A038671
Debug: request.response: [{"type":"ChatMessages"}], took 4561 ms
Debug: [PowerService] Released power save blocker acquisition, active count: 0
Debug: [PowerService] Scheduling power save blocker release in 120000ms
Trace: [messagesAPI]SSE: {"message":{"content":[],"id":"msg_bdrk_01PY7tu4JuzNAk5MxzozfRom","model":"claude-sonnet-4-6","role":"assistant","stop_reason":null,"stop_sequence":null,"type":"message","usage":{"cache_creation":{"ephemeral_1h_input_tokens":0,"ephemeral_5m_input_tokens":244},"cache_creation_input_tokens":244,"cache_read_input_tokens":106692,"input_tokens":1,"output_tokens":1}},"type":"message_start"}
Trace: [messagesAPI]SSE: {"content_block":{"text":"","type":"text"},"index":0,"type":"content_block_start"}
Warning: [messagesAPI] Unknown content_block type 'text' at index 0 for model claude-sonnet-4-6
Trace: [messagesAPI]SSE: {"delta":{"text":"Got","type":"text_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"text":" it.","type":"text_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"text":" Now","type":"text_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"text":" creating","type":"text_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"text":" the bug report file","type":"text_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"text":":","type":"text_delta"},"index":0,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"index":0,"type":"content_block_stop"}
Trace: [messagesAPI]SSE: {"content_block":{"id":"toolu_bdrk_014fJcDCpdyH91qTVmZNrDnW","input":{},"name":"create_file","type":"tool_use"},"index":1,"type":"content_block_start"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"{\"filePa","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"th","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"\": \"","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"/home/","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"wendallc","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"/Rep","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"os/gi","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"t/github/m","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"ine","type":"input_json_delta"},"index":1,"type":"content_block_delta"}
Trace: [messagesAPI]SSE: {"delta":{"partial_json":"craft/wenda","type":"input_json_delta"},"index":1,"type":"content_block_delta"}

Request IDs

9361de1e-3e3f-4408-8442-ba65582c6e88
5c61ffa3-b657-410f-8fa0-18de0481594c
3ca88782-bfe9-4d73-b792-6e89d0eeac7f
2caef609-5a48-4b17-8385-a5264d21495a
40f93a4d-218a-46d4-81ad-9958add6914a
b6dce181-602a-4022-ab46-bc8f196c663d
4cc0ae0b-037b-4623-ba07-4c68e4302c06
d81ccecd-4d16-4c39-839e-91c93c57d0aa
ad841241-bd3c-4203-ae4b-bf33928d8371
117a399c-0052-430c-9992-d3f08577e527
990324ca-a7e3-4bd1-ac73-7d7fffd35c5f
9c847227-c1e6-4c92-be88-2b2d04120c2a
44ed5fb5-6d27-48dd-b6b1-916ed331bdaa
00423002-79e2-4706-8ff2-79857e37a175
d3993c61-9171-42d2-bb1c-38ad6dd1b892
e74e2725-374f-406a-a2dc-de7460f84787
bb64ea6b-c3b1-4d58-be6f-7db60f2ddf27
0bbe8f37-a257-4025-84e3-4fb904da7c8c
3278e120-8434-46e7-b68a-32b30c218008
72584c85-f28a-4d15-a5a4-e38895f820a5
26d8473e-d89f-4ea5-b12b-110632f846e8
896ec0af-19f1-40a3-9fc8-df5f327abe49
dcaa4dd9-1148-4fc1-8c3f-c7a5964640dc
944e1040-58bc-42d4-9aa2-2cdfc53f3bd5
6ac5829e-e361-47d7-9d26-46cda9167698
1e298a36-8964-493f-b7fb-79a5f3e87b26
142c8fdf-d2b1-455a-9174-119946a7ff51
234739c9-6627-436d-9f8d-85d6200b1531
67b3f1b2-d256-4212-9def-85b34d0f8068
8a394527-74dc-4ee2-8e8f-2218fa4ca815
bb3affb6-2118-4ec9-b6d0-4342c155514f
4fa06b47-7b2b-4a42-a3b6-a2c101db0d67
ad41f2d2-e5b4-4975-aa30-32fe72afbaf4
e4312055-4ffb-4d70-a3a8-487c93d1660f
615dc6ad-dd57-4ba7-8f81-f9ab1f5bc306
34514a76-8866-4a12-82c8-7152686649d2
81de9c0c-d754-4dbb-b619-d2d27fd18929
f4443555-e838-441f-a8d3-d5f9d9421cc9
a419fed5-4e14-4fed-a5ca-394a24873142
88d39780-b154-4569-8f12-697fd08a9451
5dd0134d-8f34-420d-b3e1-9648b2748bbd
1b93396f-62a5-498b-a126-2543ae1a7827
9b3c7785-f737-4144-83e5-691758bca0b8
5c41cede-0a65-43ec-8352-dd0ebc7ce365
117ec59e-7e6b-4bff-b41a-dec053049ba0
bea3ab5f-ead8-4f0e-9774-4dfda0822f5c
5ac33635-14f8-4054-a326-ed5fe8b71895
89c50dac-b950-4aef-84ca-a570d5c1d610
33652003-2c7f-401e-a125-58bb1c7012d0

System Info

Item	Value
CPUs	AMD Ryzen 7 5700X 8-Core Processor (16 x 4645)
GPU Status	2d_canvas: enabled GPU0: VENDOR= 0x1002 [Google Inc. (AMD)], DEVICE=0x67df [ANGLE (AMD, AMD Radeon RX 580 Series (radeonsi polaris10 ACO), OpenGL ES 3.2 Mesa 25.3.6)], DRIVER_VENDOR=Mesa, DRIVER_VERSION=25.3.6 ACTIVE Machine model name: Machine model version: 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 trees_in_viz: 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, 1
Memory (System)	62.71GB (31.99GB free)
Process Argv	--crash-reporter-id 527f8027-1c60-4d80-a222-1e9037d52e13
Screen Reader	no
VM	0%
DESKTOP_SESSION	sway
XDG_CURRENT_DESKTOP	sway:wlroots
XDG_SESSION_DESKTOP	sway
XDG_SESSION_TYPE	wayland

A/B Experiments

vsliv368cf:30146710
pythonvspyt551:31249599
binariesv615:30325510
nativeloc1:31344060
dwcopilot:31170013
dwoutputs:31242946
copilot_t_ci:31333650
e5gg6876:31282496
pythonrdcb7:31342333
6518g693:31463988
aj953862:31281341
63221493:31336333
envsdeactivate2:31505458
cloudbuttont:31379625
3efgi100_wstrepl:31403338
ddidtcf:31399634
ec5jj548:31422691
cmp-cht-diffpatch-lysithea-0011:31509976
cp_cls_t_966_ss:31454198
inlinechat_v2_hd992725:31505030
4je02754:31466945
8hhj4413:31478653
ge8j1254_inline_auto_hint_haiku:31490510
38bie571_auto:31478678
cp_cls_c_1081:31454833
conptydll_true:31498968
ia-use-proxy-models-svc:31452481
e9c30283:31461165
test_treatment2:31471001
idci7584:31464702
e3e4d672:31494082
ei9d7968:31496641
nes-extended-on:31455476
chat:31457767
8hig5102:31480529
7e187181:31503455
i2gc6536:31499202
52612955:31513034
ghj88844:31499326
23c7c724:31491644
ddid_c:31478207
getcmakediagnosticsoff:31489825
pro_large_t:31499376
cp_cls_c_1082:31504161
logging_enabled_new:31498466
j0d79568:31499440
jb_cp_cls_c_632:31510883
56dj4588:31512888
ha629193:31508444
e6ac6c27:31509471

[vs-code-engineering]

Thanks for creating this issue! It looks like you may be using an old version of VS Code; the latest stable release is 1.119.1. Please try upgrading to the latest version and checking whether this issue remains.

Happy Coding!

[wendall911]

This is an agent issue. Not a VS Code version issue. The bot is incorrect, as these are not related. (To clarify I did update, thanks for the reminder)