Skip to content
Back to skills

Email Coaching

ASecurity

Rubric for giving honest, specific, voice-aware feedback on the user's draft emails

  • 2 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 2, 2026
ai-agentsgo

Security analysis

A100/100

Scanned October 2, 2026

npx -y skills add Folken2/nuvel --skill email-coaching --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Email Coaching?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Email Coaching
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/folken2-email-coaching/badge)](https://www.skillsdirectory.com/skills/folken2-email-coaching)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: email-coaching
description: Rubric for giving honest, specific, voice-aware feedback on the user's draft emails
when_to_use: The user asks for feedback, review, critique, or "make this better" on a draft they're composing. Also fires when they paste a draft and ask "thoughts?"
---

# Coaching the user's draft

Coaching = grounded, specific, short, voice-aware. Generic advice ("be clearer", "more concise") is worse than silence.

## The mandatory prelude

Before saying anything about a draft, always:

1. `get_current_compose` — read the actual draft. Never coach from memory or assumption.
2. `analyze_draft` — get objective metrics (word count, hedge count, passive count, opener/sign-off, longest sentence).
3. `recall_writing_style` — load the user's voice rules.

If `recall_writing_style` returns `{"status": "empty"}`, say so explicitly: "I haven't learned your style yet — feedback will be generic until you send a few." Then coach only on the objective metrics; skip voice claims.

## The rubric (in priority order)

For each draft, walk these in order and stop when you have 2-3 concrete points:

1. **The ask is unclear.** Can the recipient tell, in one read, what they're supposed to do? If not, the rest doesn't matter. Point at the sentence that should carry the ask.
2. **Voice mismatch.** Compare opener, sign-off, contraction count, and sentence length to `recall_writing_style`. Call out one specific drift, not "tone feels off".
3. **Hedge density.** `hedge_count` ≥ 3 in a short email is usually weakness. Quote the hedges. Suggest the assertive rewrite.
4. **Sentence load.** Any sentence > 30 words is a candidate to split. `longest_sentence_words` from the analyzer tells you which.
5. **Apology inflation.** `apology_count` ≥ 2 usually means the user is over-apologizing. Replace with a direct statement of what changed.
6. **Structural drift.** `has_opener` / `has_signoff` mismatches the user's usual pattern (per style memory).

## How to deliver

- Lead with **one** strongest observation. Not a list.
- Quote the exact line you're commenting on. No paraphrasing.
- Offer a concrete rewrite, not a direction. "Try: '…'" not "make it more direct."
- If the draft is good, say "looks good — send it" and stop. Don't manufacture issues.
- Never coach length-of-message: 4 lines is fine, 40 lines is fine, if it serves the ask.

## Failure modes to avoid

- Generic platitudes ("great point!", "consider being more concise").
- Stacking 6 points — the user won't act on 6.
- Re-writing the whole email when they asked for feedback. Edit ≠ rewrite. If they want a rewrite, they'll ask for "draft this for me".
- Inventing voice rules that aren't in the style memory.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.