fix(init): remove triple confirm prompt on Windows CP950 terminals (#602)

[sergio-sisternes-epam]

Summary

Fixes #602 — On Windows systems with non-UTF-8 console encoding (CP950, CP936, CP932), apm init in an existing project showed the overwrite confirmation prompt three times.

Root Cause

Rich Confirm.ask() would:

1. Prompt once
2. Fail to decode input on CP950, retry internally (prompt Integrate copilot runtime #2)
3. Raise an exception — the except block called click.confirm() again (prompt Will there be MCP coverage? #3)

Fix

Simplified the overwrite confirmation to always use click.confirm() directly. This is a simple y/n prompt where cross-platform reliability matters more than Rich styling. The interactive setup section retains its own Rich/click fallback pattern.

Changes

• src/apm_cli/commands/init.py — Replace Rich Confirm.ask fallback with direct click.confirm() (-9 lines, +1 line)
• tests/unit/test_init_command.py — Add 2 regression tests
• CHANGELOG.md — Add fix entry

Testing

All 17 init tests pass. Two new regression tests verify:

• The prompt appears exactly once
• click.confirm is the function called (not Rich)

[Copilot]

Pull request overview

This PR fixes a Windows non-UTF-8 terminal (e.g., CP950) regression where apm init could show the overwrite confirmation prompt multiple times by removing the Rich Confirm.ask() + fallback pattern and always using click.confirm() for the overwrite prompt.

Changes:

• Replace Rich overwrite confirmation (and fallback) with a single click.confirm() call in apm init.
• Add regression tests ensuring the overwrite prompt appears once and that click.confirm is the mechanism used.
• Add a Fixed entry to CHANGELOG.md.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
src/apm_cli/commands/init.py	Simplifies overwrite confirmation prompt to use click.confirm() directly to avoid Rich input-decoding retries on Windows code pages.
tests/unit/test_init_command.py	Adds regression coverage for single prompt emission and validates click.confirm usage.
CHANGELOG.md	Documents the fix under ## [Unreleased] / ### Fixed.

[danielmeppiel]

APM Review Panel — PR #647

[!] Verdict: REQUEST CHANGES — three must-fix items before merge

Re-running the panel verdict with stricter gating: the items below are all same-PR required, not optional follow-ups. The fix itself is correct and well-tested — but the CHANGELOG drift and the residual CP950 bugs in two other call sites mean we'd be shipping a "fixes #602" PR that leaves #602's bug class half-fixed, plus a duplicate changelog line.

Already approved earlier by @danielmeppiel — superseding with REQUEST CHANGES so the items below land before merge.

What's excellent

• Symptom-targeted test. assert result.output.count("Continue and overwrite?") == 1 (test file, line 70) tests the bug directly — a future Rich-confirm regression will fail loudly. Rare and excellent test design.
• Right-sized fix in scope. Drops Rich for one binary y/n gate where styling adds zero signal; keeps Rich on the interactive setup wizard where it does. "Pay only for what you touch" applied correctly.
• High-impact for a small diff. APAC enterprise contributors are a real adoption segment — a triple-prompt on the very first command (apm init) is the kind of bug that turns into a "this tool is broken" tweet.

Required changes (must land in this PR)

1. [x] Drop the duplicate CHANGELOG entry for #622. Diff line 10 adds:

   Propagate headers and environment variables through OpenCode MCP adapter with defensive copies to prevent mutation (#622)

   That entry is already shipped in the v0.8.12 section at CHANGELOG.md:70. PR fix: propagate headers through OpenCode MCP adapter #622 was merged on 2026-04-09. This is a merge-base artifact — please remove the line from the [Unreleased] section. Only the #602 line for this PR should remain. Shipping with this duplicate would put #622 in the changelog twice.

2. [x] Fix the same bug in src/apm_cli/commands/init.py:212. This is in the same file as the fix, immediately after the prompt this PR corrects. The interactive setup section still calls Rich Confirm.ask("Is this OK?") and will triple-prompt on CP950/CP936/CP932 — meaning bug: apm init asks "Continue and overwrite?" three times on Windows CP950 terminals #602's reproduction (apm init in a Windows non-UTF-8 terminal) will still triple-prompt, just on the next prompt instead of the first. We can't merge a "fixes bug: apm init asks "Continue and overwrite?" three times on Windows CP950 terminals #602" PR that only fixes the first of two prompts in the same command flow. Please replace Confirm.ask() here with click.confirm() (same one-line change as the rest of this PR), and extend the regression test to cover this prompt too.

3. [x] Fix the same bug in src/apm_cli/commands/deps/cli.py:455. Rich Confirm.ask("Continue?") in the deps clean flow — same triple-prompt failure mode on CP950. apm deps clean is one of the first 5 commands a Windows user is likely to run. Same one-line replacement; add a minimal test.

A _safe_confirm() helper in _helpers.py would also be reasonable (same fix, three call sites — meets the codebase's "abstract when 3+" rule), but it's not required — three direct replacements are fine if simpler. Architect's call.

Cleanup also wanted (low-effort, same PR)

4. [!] Delete the now-dead _lazy_confirm() helper. After this PR removes the only two callers (the import on diff line 22 and the use replaced by diff line 38), _lazy_confirm() at src/apm_cli/commands/_helpers.py:103 has zero callers in src/. Architect verified. One-line deletion. With the two new replacements above, this helper becomes dead permanently — drop it now while you're in the area.

Optional polish

5. [i] Inline comment to lock in intent. A one-liner above confirm = click.confirm("Continue and overwrite?") like # Plain click.confirm: Rich Confirm.ask triple-prompts on CP950 terminals; see #602 would prevent a well-meaning future PR from "modernizing" this back to Rich. Cheap, valuable.

Merge call

Land items 1-4 in this PR, then re-request review. The bug fix itself is excellent — the test design alone is worth merging. But the scope is half of #602, and we should ship the full reproduction fix in one PR rather than chase the same bug across follow-ups. With the three call-site changes plus the CHANGELOG cleanup, this becomes a definitive close on the CP950 confirmation-prompt class.

Growth angle

• APAC contributor signal. This bug was caught and fixed by an external contributor (Sergio). A Windows + CP950 reproduction means someone in the APAC enterprise dev pipeline tried APM and cared enough to fix it — leading indicator worth tracking.
• Cross-platform parity narrative. With fix(auth): thread dep_ref.port into credential resolution (#785) #788 (auth port-threading parity) and fix: VS Code adapter fails to configure HTTP/SSE remote MCP servers (#654) #656 (cross-runtime adapter parity) both landing, this is the third recent PR landing "APM works the same regardless of where/how you run it." Worth amplifying once shipped completely.
• Release-notes one-liner (post-fix): "apm init and apm deps clean no longer triple-prompt on Windows non-UTF-8 terminals (Taiwan/HK/Japan code pages)." — naming both commands and the affected regions resonates.

Thanks @sergio-sisternes-epam — happy to re-approve as soon as the four required items land. The fix you've already shipped here is the template; extending it to the two siblings is mechanical.

_{Reviewed via the apm-review-panel skill: python-architect, cli-logging-expert, devx-ux-expert; convergent verdict, escalated to REQUEST CHANGES on follow-up scope.}

[danielmeppiel]

Escalating to REQUEST CHANGES per the updated panel verdict above (#647 (comment)). Three required same-PR items:

1. Drop the duplicate #622 CHANGELOG line (already shipped in v0.8.12 at CHANGELOG.md:70).
2. Fix Confirm.ask() at src/apm_cli/commands/init.py:212 — same file, same command, same bug class.
3. Fix Confirm.ask() at src/apm_cli/commands/deps/cli.py:455 — same triple-prompt failure mode on CP950.

Plus a low-effort cleanup (delete dead _lazy_confirm() helper in _helpers.py:103, zero callers after this PR).

The fix you've already shipped here is correct and well-tested — the test design (output.count(...) == 1) is exemplary. Extending it to the two sibling call sites is mechanical and gives us a definitive close on the CP950 confirmation-prompt class in one PR rather than chasing it across follow-ups.

Happy to re-approve as soon as the four items land. This supersedes my earlier approval.