Update view replay captures after line numbers were removed

[stephentoub]

Overview

What

Update the view replay captures to the raw-content form that current runtimes emit, and add a view-scoped tool-result normalizer so captures replay against runtimes on either side of the behavior change.

• 30 view tool results across 27 snapshots in test/snapshots/ lose their N. line-number prefixes.
• normalizeViewLineNumbers strips a leading run of consecutively numbered lines from view results only.

Why

github/copilot-agent-runtime#13802 (CLI: Remove line numbers from view output, merged 2026-08-26, merge commit 36db7e12b91d2264060eb3f8892c25485fcefcef) graduated the winning experiment and made raw file content the unconditional view behavior. That PR updated the runtime's own prompt snapshots and replay captures, but the external captures living in this repo were not updated, so they still hold the numbered form:

- role: tool
  tool_call_id: toolcall_2
  content: 1. Hello from multi-turn test

Under the current runtime the CLI sends Hello from multi-turn test. Strict replay matching rejects it, the replaying CAPI proxy answers 500 Internal Server Error with Request-ID proxy-error, and the CLI retries five times before the session fails:

System.InvalidOperationException : Session error: Execution failed: Failed to get response
from the AI model; retried 5 times (Request-ID proxy-error) Last error: 500 Internal Server Error

Because the drift is in a tool result rather than in any provider protocol, it cascades to every backend leg of agent-runtime's Copilot SDK C# tests workflow — capi, anthropic-messages, openai-responses, and openai-completions — for every test whose conversation contains a view call. See agent-runtime workflow runs 33072804346 (main) and 33046860889. This is upstream fixture drift in this repo; it is not caused by agent-runtime #17389.

Why a normalizer as well as updated fixtures

nodejs/package-lock.json pins @github/copilot to 1.0.81-11, published before #13802 merged, so this repo's own CI still runs a CLI that emits the prefixes. Updating the fixtures alone would fix the agent-runtime canary and break every SDK's E2E suite here. The normalizer bridges both runtimes, and it also keeps locally re-recorded captures in the shipped raw-content form.

This follows the precedent set by #2013 (Collapse built-in tool list in replay-proxy snapshot matching), which resolved the same class of runtime-version drift for the same C# canary.

Matching is not loosened beyond the affected shape:

• The normalizer is registered for toolName: "view", so tools that legitimately return numbered lines keep matching strictly. test/snapshots/session_todos_changed/fires_session_todos_changed_and_exposes_rows_and_dependencies.yaml has a sql result of 1. INSERT ... and is deliberately untouched.
• Only a leading run of consecutive numbers is stripped; it stops at the first gap, so trailing content such as the truncation notice is preserved.
• Nothing in the test assertions changed.

Validation

Local, Windows, net8.0, snapshots forced read-only (GITHUB_ACTIONS=true).

Reproduced first: with the fix stashed and COPILOT_CLI_PATH pointed at a post-#13802 CLI (1.0.81-14), MultiTurnE2ETests fails with exactly the CI signature — proxy-error, 500, retried 5 times.

With the fix applied:

Suite	Runtime	Result
test/harness vitest (incl. 3 new tests)	n/a	67 passed
.NET MultiTurn, BuiltinTools, Tools, Hooks, SubagentHooks, EventFidelity, SystemMessageTransform	1.0.81-14 (post-change)	37 passed, 1 skipped
.NET Permission, SessionConfig, Session, ClientOptions, SessionTodosChanged	1.0.81-14 (post-change)	106 passed, 2 failed
.NET MultiTurn	1.0.81-11 (pinned, pre-change)	2 passed
Node.js multi_turn, builtin_tools E2E	1.0.81-14 (post-change)	9 passed, 1 skipped

The 2 failures are SessionConfigE2ETests.Vision_Enabled_Then_Disabled_Via_SetModel and Vision_Disabled_Then_Enabled_Via_SetModel, which time out identically on a clean tree at main with the pinned CLI. They are pre-existing and unrelated: their view results are Viewed image file successfully. and were never numbered.

New harness tests cover capture-time stripping, the non-consecutive stop condition, sql results being left alone, CRLF content, and replaying both the numbered and unnumbered request forms against the same snapshot.

Checklist

• The Overview section above is filled with a description of the PR
• Fixtures reflect shipped runtime behavior, with no unrelated or machine-specific capture churn
• Strict replay matching preserved outside the view line-number shape

[Copilot]

Pull request overview

Updates replay fixtures for raw view output while supporting older runtimes that emit line-number prefixes.

Changes:

• Removes line-number prefixes from 27 snapshots.
• Adds view-scoped normalization and replay/capture tests.
• Preserves strict normalization for other tools.

Show a summary per file

File	Description
test/harness/replayingCapiProxy.ts	Adds view result normalization.
test/harness/replayingCapiProxy.test.ts	Tests capture and replay normalization.
test/snapshots/tools/invokes_built_in_tools.yaml	Updates raw view result.
test/snapshots/system_message_transform/should_work_with_static_overrides_and_transforms_together.yaml	Updates transformed-message fixture.
test/snapshots/system_message_transform/should_invoke_transform_callbacks_with_section_content.yaml	Updates transform-callback fixture.
test/snapshots/system_message_transform/should_apply_transform_modifications_to_section_content.yaml	Updates transform-modification fixture.
test/snapshots/subagent_hooks/should_invoke_pretooluse_and_posttooluse_hooks_for_sub_agent_tool_calls.yaml	Updates subagent view result.
test/snapshots/session/should_send_with_file_attachment.yaml	Updates attachment fixture.
test/snapshots/session_config/should_use_workingdirectory_for_tool_execution.yaml	Updates working-directory fixture.
test/snapshots/session_config/should_apply_workingdirectory_on_session_resume.yaml	Updates resumed-session fixture.
test/snapshots/session_config/should_accept_message_attachments.yaml	Updates message-attachment fixture.
test/snapshots/permissions/should_invoke_permission_handler_for_write_operations.yaml	Updates permission conversations.
test/snapshots/permissions/permission_handler_for_write_operations.yaml	Updates equivalent permission fixture.
test/snapshots/multi_turn/should_use_tool_results_from_previous_turns.yaml	Updates result and removes redundant conversation.
test/snapshots/multi_turn/should_handle_file_creation_then_reading_across_turns.yaml	Updates result and removes redundant conversation.
test/snapshots/hooks/should_invoke_pretooluse_hook_when_model_runs_a_tool.yaml	Updates pre-tool hook fixture.
test/snapshots/hooks/should_invoke_posttooluse_hook_after_model_runs_a_tool.yaml	Updates post-tool hook fixture.
test/snapshots/hooks/should_invoke_both_pretooluse_and_posttooluse_hooks_for_single_tool_call.yaml	Updates combined-hook fixture.
test/snapshots/hooks/should_invoke_both_pretooluse_and_posttooluse_hooks_for_a_single_tool_call.yaml	Updates alternate combined-hook fixture.
test/snapshots/hooks/invoke_pre_tool_use_hook_when_model_runs_a_tool.yaml	Updates legacy pre-hook fixture.
test/snapshots/hooks/invoke_post_tool_use_hook_after_model_runs_a_tool.yaml	Updates legacy post-hook fixture.
test/snapshots/hooks/invoke_both_hooks_for_single_tool_call.yaml	Updates legacy combined-hook fixture.
test/snapshots/event_fidelity/should_preserve_message_order_in_getmessages_after_tool_use.yaml	Updates message-order fixture.
test/snapshots/event_fidelity/should_emit_tool_execution_events_with_correct_fields.yaml	Updates tool-event fixture.
test/snapshots/event_fidelity/should_emit_events_in_correct_order_for_tool_using_conversation.yaml	Updates event-order fixture.
test/snapshots/client_options/should_use_client_cwd_for_default_workingdirectory.yaml	Updates client-CWD fixture.
test/snapshots/builtin_tools/should_read_file_with_line_range.yaml	Updates ranged view output.
test/snapshots/builtin_tools/should_edit_a_file_successfully.yaml	Updates edited-file output.
test/snapshots/builtin_tools/should_create_a_new_file.yaml	Updates created-file output.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 29/29 changed files
• Comments generated: 1
• Review effort level: Balanced