PR comment bodies posted from Windows PowerShell 5.x can be mojibake-corrupted

[abeltrano]

Summary

Templates that post comments to GitHub PRs (respond-to-pr-comments, review-pull-request) and to Azure DevOps PRs are vulnerable to silent character corruption (mojibake) when the agent runs on Windows PowerShell 5.x and the comment body contains non-ASCII characters (em-dashes, smart quotes, accented names, currency symbols, etc.).

Reproduction

Observed during work on PR #254. The PR description was edited via:

gh pr view 254 --json body --jq .body | Out-File -Encoding utf8 pr-body.md
# ...sed-style replace...
gh pr edit 254 --body-file pr-body.md

After the round-trip, em-dashes (—, U+2014) appeared in the GitHub UI as דÇù-style garbage and apostrophes (') as ''.

Root cause

Two interacting issues on Windows PowerShell 5.x:

1. Pipe decoding: gh ... | Out-File decodes the UTF-8 byte stream from gh as the console codepage (Windows-1252 by default), then re-encodes when writing — producing classic UTF-8 → CP1252 → UTF-8 mojibake.
2. Out-File defaults: Out-File and Set-Content default to Windows-1252 (not UTF-8). Out-File -Encoding utf8 writes UTF-8 with a BOM, which some APIs misinterpret.

Both bash / zsh / PowerShell 7+ default to UTF-8 and are not affected by the same defaults.

Impact

Any template that drafts user-voice text and posts it via gh api, gh pr edit, or az rest is at risk:

• respond-to-pr-comments (GitHub + ADO reply paths)
• review-pull-request (inline comments + overall review summary)
• Future templates that post comments, descriptions, or release notes

The risk is amplified by the new human-voice-fidelity protocol (PR #254), which encourages calibrated voice — including punctuation habits like em-dashes for users who use them, and accented names in greetings.

Proposed fix

Add an "Encoding for external posts" subsection to protocols/guardrails/operational-constraints.md:

• Always write comment / reply / description bodies to a temp file before passing to gh / az rest.
• On Windows PowerShell 5.x, use [System.IO.File]::WriteAllText($path, $content, [System.Text.UTF8Encoding]::new($false)) — never Out-File or Set-Content for non-ASCII bodies.
• Never round-trip existing comment text through gh pr view --jq | Out-File for editing — write fresh content in clean UTF-8.
• bash / zsh / PowerShell 7+: default UTF-8 is fine.

The protocol is already included by both PR-posting templates, so this single edit fixes all current consumers and any future ones that include the protocol.