Agent Mode / Copilot CLI delivered identical user prompt twice in one session

## Summary

VS Code Agent Mode / Copilot CLI agent host delivered the same user prompt twice into one Copilot CLI session. The second delivery happened after the agent had already completed the requested implementation, so the agent treated the repeated prompt as a fresh request until it re-checked session state.

This looks like duplicate user-message delivery at the VS Code / Agent Host / Copilot CLI boundary, not duplicated model output.

## Environment

- VS Code Agent Mode with Copilot CLI runtime
- Remote SSH workspace on Linux
- Agent session producer: `copilot-agent`
- Copilot agent version recorded in the session log: `copilotVersion: 1.0.70-0`
- Model selected in that session: `gpt-5.5`
- Context tier: `long_context`

## What happened

A user sent a `/bots-plan-review-fix ...` prompt once, watched the agent implement and verify the changes, and then the identical user prompt was delivered again later in the same session.

Sanitized local session-log evidence:

- First user message timestamp: `2026-07-27T05:25:15.846Z`
- Duplicate user message timestamp: `2026-07-27T05:46:26.157Z`
- Both user message payloads had the same sanitized SHA-256 prefix: `ef69989333a53295`
- Both payloads had the same length: `2240` characters
- The implementation was completed and recorded before the duplicate arrived (`2026-07-27T05:46:06Z` in local continuity receipts)

The skill/context wrapper for the same command was also duplicated immediately after each prompt:

- First skill-context timestamp: `2026-07-27T05:25:23.922Z`
- Duplicate skill-context timestamp: `2026-07-27T05:46:42.094Z`
- Same sanitized SHA-256 prefix: `f6f5ae7334a82577`
- Same length: `3316` characters

## Expected behavior

A user prompt submitted once should create one `user.message` in the Copilot CLI agent session. If the user sends another message while a request is in progress, VS Code should either steer the existing turn or queue exactly the new message, not replay a previously delivered prompt after completion.

## Actual behavior

The exact same prior prompt was delivered again as a fresh `user.message` in the same session, after the original task was already completed.

## Impact

For coding-agent sessions this is risky: the duplicate prompt can cause repeated side effects unless the agent independently detects idempotency/stale state. In this case the agent avoided re-implementation only after checking local continuity and current code state.

## Notes

I did not attach raw session logs because they contain private workspace prompts and file paths. The evidence above is from the local Copilot CLI session event log, reduced to timestamps, content hashes, lengths, and version metadata.

A likely mitigation on the user side is setting:

```json
"chat.requestQueuing.defaultAction": "steer"
```

But this issue is about the host delivering a previously submitted prompt again, so it would be useful to know whether queued request replay can duplicate already-completed turns.

## Repository note

I first tried to file this in `microsoft/vscode-copilot-release`, but GitHub reports that repository is archived/read-only for new issues. Filing here because the symptom is at the VS Code Agent Mode / Agent Host delivery layer.