Add powershell (windows) support

[cnukaus]

Enhanced Powershell support:

I added Windows command resolution helpers, added codex detection, expanded the “no CLI found” message to include Codex, and changed npm-installed CLIs to prefer <name>.cmd on Windows before spawning.
That fixes the exact failure mode where PowerShell can run codex but child_process.spawn(\"codex\") cannot.

I also updated the CLI surface and docs, now advertises codex in --cli

[cnukaus]

@microsoft-github-policy-service agree

[Alan-Jowett]

Thank you for this contribution, @cnukaus! 🎉 Adding Codex CLI support and fixing the Windows .cmd shim resolution issue with child_process.spawn() is a great improvement — this is a real pain point for Windows users.

The resolveSpawnCommand() approach is well thought out, the test coverage is solid, and the documentation updates are thorough. I've left a few review comments below — mostly around spec sync and a test edge case. Really appreciate you taking the time to submit this!

[Alan-Jowett]

Thanks again for this well-structured PR! A few items to address before merge:

IMPLEMENTATION_PLAN.md — please remove (must fix)

This appears to be a development log from the implementation process. PromptKit doesn't maintain implementation plans at the repo root, and this would add noise to the project structure.

Spec files need codex added (should fix):

requirements.md has several references to the supported CLI list that need codex added:

• REQ-CLI-010 — detection order (copilot → gh copilot → claude)
• REQ-CLI-011 — valid --cli values (copilot, gh-copilot, claude)
• REQ-CLI-017 — per-CLI spawn commands (missing codex entry)
• ASSUMPTION-003 — accepted --cli values

validation.md should document the new TC-CLI-072A test case alongside existing TC-CLI-072.

[what-cloud]

Updated routing .cmd launches through cmd.exe in cli/lib/launch.js:67 while keeping
argument handling explicit, so it does not fall back to the old shell: true behavior.

The No supported LLM CLI found on PATH lines are still expected from the negative-path tests. The important part is that the spawn EINVAL should now be gone.

[Alan-Jowett]

Hi @cnukaus / @what-cloud — thanks for the follow-up commits! A few items still need attention before we can merge:

1. IMPLEMENTATION_PLAN.md — still present (must fix)
   This development log shouldn't live at the repo root. Please remove it from the PR.

2. cli/specs/design.md line 148 — stale detection description
   The text still references execFileSync with where/which, but the implementation now uses direct PATH scanning. Since you're already editing this section, it would be great to correct this.

3. cli/specs/requirements.md — character encoding corruption
   All em-dashes (—, U+2014) in this file have been replaced with triple question marks (???, U+003F × 3) — 30 occurrences. This appears to be an editor or locale encoding issue. The base branch has the correct em-dashes; the PR branch has ??? in their place. The same issue also affects a few comments in cli/tests/launch.test.js.

   You can verify by searching for ??? in the file, or comparing the raw bytes against the upstream version.

Everything else looks good — the spec updates for codex, the test fixes, and the Windows .cmd shim logic are all solid. Just these three items to tidy up. Thanks again for the contribution! 🙏

[cnukaus]

@Alan-Jowett

1. the file was removed
2. updated
3. can you re-check? I copied it again from your latest master

Hi @cnukaus / @what-cloud — thanks for the follow-up commits! A few items still need attention before we can merge:

1. IMPLEMENTATION_PLAN.md — still present (must fix)
   This development log shouldn't live at the repo root. Please remove it from the PR.
2. cli/specs/design.md line 148 — stale detection description
   The text still references execFileSync with where/which, but the implementation now uses direct PATH scanning. Since you're already editing this section, it would be great to correct this.
3. cli/specs/requirements.md — character encoding corruption
   All em-dashes (—, U+2014) in this file have been replaced with triple question marks (???, U+003F × 3) — 30 occurrences. This appears to be an editor or locale encoding issue. The base branch has the correct em-dashes; the PR branch has ??? in their place. The same issue also affects a few comments in cli/tests/launch.test.js.
   You can verify by searching for ??? in the file, or comparing the raw bytes against the upstream version.

Everything else looks good — the spec updates for codex, the test fixes, and the Windows .cmd shim logic are all solid. Just these three items to tidy up. Thanks again for the contribution! 🙏