Improve line-number accuracy in radius-code-review skill

[brooke-hamilton]

Description

Improves the radius-code-review skill to eliminate incorrect line numbers, which is a frequent issue with the review scripts. An alternate approach is to remove the generated script altogether.

Changes

.github/skills/radius-code-review/SKILL.md:

• Anchor-based comments: Every review comment now records an anchor (a unique code snippet on the target line). The anchor — not the integer — is the source of truth.
• New "Line-number accuracy (mandatory)" section: Instructs the agent to never type a line number from memory and to look up each line via grep -n, and to ensure the cited line is on the RIGHT side of the diff.
• Mandatory line-number verification (Step 3): Adds a step to print actual file content at every cited line and confirm it matches the comment's anchor before generating the script.
• Runtime anchor resolution (Step 5): Adds a line_for shell helper that resolves line numbers from anchors at script build time and fails loudly if an anchor matches zero or more than one line, so a wrong/stale anchor aborts the script instead of mis-placing a comment.
• Checklist updates: Adds items requiring anchors and line-number verification.

Type of change

• This pull request fixes a bug or improves an existing skill.

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

[Copilot]

Pull request overview

This PR updates the Radius code-review skill to reduce incorrect inline-review line numbers by requiring anchor snippets, verification, and runtime line resolution.

Changes:

• Adds anchor-based review comment formatting and mandatory line-number accuracy guidance.
• Adds a verification step to compare cited lines against anchors before script generation.
• Updates the generated review-posting script template to resolve line numbers from anchors.