ToneGuard Skill

Analyzes a message against ToneGuard's style rules and either clears it (no issues) or
flags problems and suggests a rewrite.

Canonical source: /Users/summerrae/claude_code/toneguard/SKILL.md (mirrored into the Claude skills-plugin cache). Edit here first; re-sync the cache copy after any change.

Complementary reference: communication principles

Before analyzing a message, also load /Users/summerrae/claude_code/claude-skills/shared/communication-principles.md. The principles there (audience-centered focus, lead with the conclusion, plain-language simplicity, "I help" framing, preparation, control-your-message) apply on top of ToneGuard's style rules — they do not replace them. The ToneGuard rules below are tactical (em dashes, passive-aggressive triggers, hedging); the communication principles are strategic (audience, structure, medium).

Precedence when the two disagree: ToneGuard style rules win on specific phrasing calls (they are more opinionated and user-specific); communication principles win on structure and framing calls.

Output format

If the message passes: One sentence confirming it's clear and on-tone. No rewrite needed.

If the message has issues:

1. Flagged issues — bullet list of specific problems, each with a brief explanation
2. Suggested rewrite — full revised message
3. What changed — one line per change, e.g. "Removed em dash → period"

Keep the analysis concise. Don't explain rules the user didn't violate.

---

Style Rules (from style-rules.md)

Voice

• Sound like a real person, not a corporate template
• Warm, direct, casual-professional
• Rewrites should read like something Summer would actually send

Sentence structure

• One idea per sentence; split anything you have to read twice
• Lead with the main point; don't bury the ask
• Cut hedging: "What I mean to say is" → just say it
• State commitments clearly and early

Things to flag / avoid

• Em dashes (—) → reads as AI-generated; use periods or commas
• "Would you mind... next time" → passive-aggressive; rewrite as a casual statement
• "It made me feel..." → puts reader on the defensive; state what happened and what you need
• "I noticed you [did X]" → surveillance energy; use passive framing instead
• Questions when you mean statements → "Would you mind checking?" → "Just loop me in"
• Guilt-trip framing → keep asks general, not accusatory
• "Even though I..." → sounds like building a case
• "Is everything okay?" → if you mean "what's going on?", just ask that

Things to do

• Assume good intent; frame as miscommunication, not mistakes
• Name the specific thing (not "the channel" — name it)
• State what you need clearly; every request should make the ask obvious
• "Going forward" > "next time" (forward-looking, not finger-pointing)
• Pair asks with something positive when appropriate

Clear is kind (Brené Brown principles)

• Being clear IS being kind; vagueness to avoid discomfort is unkind
• Name the issue directly; don't hint or soften to the point of confusion
• State expectations explicitly; implied expectations are unfair
• If a message dances around a hard topic, say it directly but kindly
• Some things shouldn't be text — if emotionally complex, suggest a call

Hemingway checks

• Flag sentences over 25 words → split or simplify
• Flag adverbs: "very," "really," "basically," "honestly" → use a stronger verb
• Flag qualifiers: "I think," "sort of," "kind of," "maybe" → commit or cut
• Simpler words: "utilize" → "use," "facilitate" → "help," "commence" → "start"

Grammarly-style checks

• Grammar, subject-verb agreement, wrong word usage
• Wordy phrases: "in order to" → "to," "due to the fact that" → "because"
• Passive voice: "was completed" → "I completed"
• Tentative openers: "I just wanted to..." → drop "just"; "I'm no expert but..." → drop it
• "Does that make sense?" → replace with a clear statement

Confidence and engagement

• Strong, relevant first line
• One or two points max per message
• Active voice throughout
• Flag flat, monotone writing

Slack-specific (if context is Slack)

• Edit ruthlessly; fewer words always
• Use bold for key points and actions
• Numbered lists over bullets
• Embed links in phrases; no naked URLs
• Spell out what the reader would otherwise have to calculate

When NOT to flag

• Message is already clear, warm, and professional → silent pass
• Casual greetings, emoji reactions, quick acknowledgments
• Tone matches context appropriately

---

Analysis workflow

1. Read the message in full
2. Check against each rule category above
3. If no issues: clear it with one sentence
4. If issues found:
   • List only the rules actually violated (don't enumerate all rules)
   • Write a rewrite that fixes all issues without over-engineering the message
   • Summarize changes in a short diff list
5. Never suggest changes that make the message longer without adding meaning