Copilot says complete but the change is missing? Verify the repository.

Treat the repository and verification output as the acceptance evidence. A chat completion message cannot prove that the requested files changed or that tests passed.

Do this first: Compare the requested outcome with the Git diff, expected files, and a focused test. Save the current state before asking Copilot to continue so it does not repeat work that already landed.

What the failure looks like.

The run is reported as complete, but an expected file, code path, or test result is missing or incorrect when checked locally.

Why it happens.

This is a repository-verification workflow, not a claim of a specific Copilot defect. Historical issue #13592 describes an Agent waiting after terminal output, which is a different symptom and is not evidence that Copilot falsely reported completion.

Fix it without losing the task.

Translate the request into one or more observable acceptance conditions.

Inspect Git status and the diff, then confirm the expected files and behavior exist in the worktree.

Run the narrowest relevant test or check and record its exact output. Do not report untested requirements as complete.

Ask Copilot for only the missing change, providing the current diff and failed check. Re-run the acceptance check afterward.

What the next AI should receive.

Acceptance criteria, expected files, actual Git diff, focused test command and output, missing behavior, any correct partial changes, and one next action.

First safe action

Write down the acceptance condition, inspect the exact files and diff, and run only the focused check needed to determine what remains incomplete.

Evidence boundary

A passing test covers only the behavior it exercises; a completion message and a partial diff do not establish full correctness. Keep untested requirements marked as unverified.

Diagnostic checkpoint

Use this page when the output is incomplete or fails acceptance. If the Agent is still waiting after a finished terminal command, use the separate terminal-state guide.

How ShardStitch helps.

ShardStitch can carry the goal, actual diff, failed approaches, and verification state into a fresh session. It cannot certify code quality or replace the project's tests and review.

Source trail.

Copilot completion-state evidence, checked September 24, 2026: Copilot issue #13592 was opened June 29, 2025 for Copilot Chat 0.28.2 and VS Code 1.101.2. The reporter said a terminal command had returned while the Agent run remained waiting; pressing Enter in the terminal appeared to wake it. The issue is closed as a duplicate in the archived issue repository. This supports a historical UI/terminal-state mismatch, not a claim that every completion indicator is wrong; the workaround was reporter-observed, not maintainer-verified.

FAQ.

Can Copilot saying Done be used as proof that the task passed?

No. Verify the changed files and run the relevant check. The chat status records the session state, not the correctness of the repository.

Should I ask Copilot to redo the whole task if one check fails?

Usually not. Preserve correct partial work, provide the failing output, and ask for the smallest missing change with an explicit acceptance test.

Scope note: This guide covers local verification after an AI coding session claims completion. It does not attribute incomplete work to a confirmed Copilot product defect.