docs: add Windows command references for skills

[neubig]

Summary

• add inline Windows/PowerShell notes only beside short POSIX-shell examples
• add scoped references/windows.md files only for skills with repeated Unix-style command blocks
• keep each reference focused on that skill's commands: env vars, curl, temp paths, tarballs, grep/sed, Linux package installs, SSH permissions, or gh quoting as applicable
• regenerate skills/index.js from the edited SKILL.md files

Validation

• npm run build:skills
• $env:PYTHONUTF8='1'; uv run --with pyyaml python scripts/sync_extensions.py --check commands catalog coverage
• verified every SKILL.md references/windows.md pointer resolves to a local file
• git diff --cached --check

Note: full sync_extensions.py --check is not useful from this Windows checkout because the symlink check reports existing .claude-plugin / .codex-plugin entries as non-symlinks. The coverage warnings shown by the scoped check are pre-existing and non-blocking.

Issue

Fixes #331

[neubig]

@all-hands-bot please review this PR.

[enyst]

@OpenHands Do a /codereview on this PR. Investigate deeply first and check the linked issue too. Read agentskills.io format so you understand.

Post your review with gh api as a gh review. Feel free to use event approved or comment or request-changes.

[openhands-ai]

I'm on it! enyst can track my progress at all-hands.dev

[enyst]

🟢 Taste Rating: Good taste - This keeps the Windows guidance scoped to the commands that need translation, uses focused references/windows.md files where repeated shell patterns would otherwise bloat SKILL.md, and preserves the AgentSkills progressive-disclosure model.

I investigated the linked issue (#331), the full PR diff, the generated catalog, and the AgentSkills specification/best-practices docs. The added references/ files are consistent with the spec's optional reference resources and the guidance to keep detailed material on-demand and focused. The inline notes are short and placed beside the relevant POSIX snippets rather than turning each skill into a cross-platform manual.

Validation I ran:

• npm run build:skills
• uv run --with pyyaml python scripts/sync_extensions.py --check commands catalog coverage
• uv run --with pyyaml python scripts/sync_extensions.py --check
• GITHUB_TOKEN=... python scripts/sync_openhands_sdk_skill.py --check
• Markdown fence/CR checks and local references/windows.md pointer checks across the changed files

No blocking issues found. The only AgentSkills frontmatter mismatches I found are pre-existing nested plugin skill names in plugins/cobol-modernization/.../cobol-modernization-overview and plugins/migration-scoring/.../migration-scoring-overview; this PR did not introduce those.

[RISK ASSESSMENT]

• [Overall PR] ⚠️ Risk Assessment: 🟢 LOW
  Documentation-only change plus generated catalog normalization. Main user-facing risk is inaccurate Windows command translation, and the new examples are scoped, conventional PowerShell equivalents for env vars, curl aliasing, temp paths, here-strings, redirects, and Windows-native install paths.

VERDICT:
✅ Worth merging: The structure matches AgentSkills progressive disclosure and the generated artifacts validate cleanly.

KEY INSIGHT:
The PR solves the real portability gap without polluting every skill's always-loaded context.

This review was created by an AI agent (OpenHands) on behalf of @enyst.

[openhands-ai]

Since my last summary, there have been no additional changes or actions.

• The PR review request was completely addressed: I investigated the PR, linked issue, and AgentSkills format, ran relevant validation, and posted the GitHub PR review via gh api.
• No repository files were modified or committed.
• No extraneous changes were introduced.