Add experimental alt prompt for Claude Opus 4.7

[bhavyaus]

Summary

• Adds a new Claude47OpusPrompt (standalone copy of Claude46OpusPrompt) gated behind the experiment-based setting github.copilot.chat.claude47OpusPrompt.enabled (default false). When the setting is on, claude-opus-4-7* / claude-opus-4.7* traffic routes to the new prompt; otherwise it falls back to Claude46OpusPrompt.
• Behavioral additions vs Claude 4.6 Opus reflect the Opus 4.7 prompting guide (tool triggering, subagent fan-out, response shape, exact verbosity wording, full operationalSafety with three categories) plus targeted lessons - no internal narration between tool calls, end-of-turn summary cap, comment discipline, "don't duplicate in-flight subagent work", "verify subagent output before reporting done", <skillUsage> primer.

[github-actions]

Base: a416899e Current: 89c70cbb

No screenshot changes.

[Copilot]

Pull request overview

Adds an experiment-gated Claude Opus 4.7 agent prompt to the Copilot extension, so Anthropic Opus 4.7 traffic can use a separate system prompt without changing the existing Opus 4.6 path.

Changes:

• Adds a new experiment-based config key for github.copilot.chat.claude47OpusPrompt.enabled, defaulting to false.
• Introduces Claude47OpusPrompt as a standalone Anthropic agent prompt variant with updated instructions for Opus 4.7.
• Updates the Anthropic prompt resolver to route claude-opus-4-7* / claude-opus-4.7* models to the new prompt when the experiment is enabled.

Show a summary per file

File	Description
extensions/copilot/src/platform/configuration/common/configurationService.ts	Adds the experiment-backed config key used to gate the new prompt.
extensions/copilot/src/extension/prompts/node/agent/anthropicPrompts.tsx	Adds the new Opus 4.7 prompt class and resolver logic that switches prompt selection based on model + experiment flag.
extensions/copilot/package.nls.json	Adds the localized description string for the new setting.
extensions/copilot/package.json	Exposes the new experimental setting in the extension manifest.

Copilot's findings

Comments suppressed due to low confidence (1)

extensions/copilot/src/extension/prompts/node/agent/anthropicPrompts.tsx:651

• This adds new model-selection logic for claude-opus-4-7* / claude-opus-4.7*, but there is no regression test covering the on/off behavior of github.copilot.chat.claude47OpusPrompt.enabled. The existing prompt snapshot suite already exercises multiple Anthropic families, so without a focused test here a typo in the model match or flag check would silently route 4.7 traffic to the wrong prompt.

		if (this.isOpus47(endpoint) && this.configurationService.getExperimentBasedConfig(ConfigKey.Claude47OpusPromptEnabled, this.experimentationService)) {
			return Claude47OpusPrompt;

• Files reviewed: 4/4 changed files
• Comments generated: 1

[bhavyaus]

Tracking issue: microsoft/vscode-internalbacklog#7600