agentleFS
CSPulseCSPulse/customer-success-skills/skills/email-critic/SKILL.md

email-critic

Skill2 starsChanged 3 months agoMITView on GitHub

Use when the user says "would you send this", "review this email", "tighten this", "does this sound right", "how will the customer take this", or pastes a draft with no instruction. Checks the draft against the call or thread it came from, then returns a plain verdict and a tightened version. Never a numeric score. Not for turning a recording into emails (use call-recap) or a message to a whole list (use customer-update).

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the user pasted a draft, go.

    line 13
  2. Distinctive:

    Never wait. With no source, review the writing only and say so under What I could not check.

    line 19
  3. Distinctive:

    Verify the fix is executable: If the rewrite relies on a meeting happening, a person being available or a file existing, check that against …

    line 29
  4. Distinctive:

    Rewrite only what you named: Keep the user's voice and every fact, owner and date.

    line 62
  5. Distinctive:

    Pull out what you cannot verify: instead of laundering it into cleaner prose, and flag it.

    line 63
  6. Distinctive:

    A chase must change shape: If this is the third ask, repeating it harder fails.

    line 65
  7. Distinctive:

    Bad news is a call first: If the customer has not heard it yet, say the email should be the record of a conversation, not the conversation.

    line 66

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Email Critic

The user walks away knowing whether to send the draft, what is wrong with it, and a version that fixes only that.

Start here

If the user pasted a draft, go. Better with the call transcript or the thread it replies into. If neither is there, ask once, in one message, for:

  1. The source: transcript, notes or the customer email being answered
  2. Who it goes to and what the user needs back
  3. Anything the user already owes this person

Never wait. With no source, review the writing only and say so under What I could not check.

Do this

  1. Name what the email is for in one line: recap, nudge, ask, escalation, bad news, forwardable note. If you cannot name it, that is the verdict.
  2. Check facts against the source first. Every date, number, name, owner and commitment. Flag anything upgraded ("let's look at that" became "as agreed"), anything claimed as sent that you cannot see, anything the user did not commit to. Load references/by-email-type.md for the section matching this type.
  3. Check what the customer will notice missing: the objection they raised, the question they asked, an owner, a date.
  4. Check the ask. One ask, near the top, small enough to answer from a phone, addressed to a named person.
  5. Read it as the recipient. Would they reply today, or file it? Does it sound like a partner or a vendor chasing?
  6. Give the verdict, scaled to the draft: Send it / Nearly there / Needs work / Do not send. Then the tightened version.
  7. Verify the fix is executable. If the rewrite relies on a meeting happening, a person being available or a file existing, check that against the source too.

Output

Verdict: [Send it / Nearly there / Needs work / Do not send] - [one sentence why]

Purpose: [what this email is for]

Wrong or unverified
- [claim] | source says: [what actually happened] | fix: [change]

What to change
- [specific problem] -> [specific fix]

Tightened version
[subject]
[body, preserving every fact, owner and date]

What I kept and why: [anything that could have been cut but was not]

What I could not check
- [gap]

Send it: one sentence and any tiny inline fix, no rewrite. Do not send: say what the email should be before rewriting. A filled example is in examples/sample.md.

Rules

  • Truth before tone. A clumsy sentence costs nothing; a wrong date or an invented commitment becomes the customer's record of what was agreed.
  • No score, ever. A number from the model about to do the rewrite starts driving the edit (cutting words to justify a concision figure) instead of fixing the draft.
  • Do not manufacture criticism. A review that always finds three problems stops being believed. If it is good, say send it.
  • Every criticism names its fix. "Too long" is noise; "cut paragraphs two and three, they restate one" is feedback.
  • Rewrite only what you named. Keep the user's voice and every fact, owner and date. A recap becomes the record, so an added next step becomes a real obligation.
  • Pull out what you cannot verify instead of laundering it into cleaner prose, and flag it.
  • No house style the user has not stated. Judge against this conversation, not generic taste about punctuation or banned words.
  • A chase must change shape. If this is the third ask, repeating it harder fails; remove the blocker for them or ask what is in the way.
  • Bad news is a call first. If the customer has not heard it yet, say the email should be the record of a conversation, not the conversation.

Related (only if installed)

customer-update when the draft goes to a list. one-pager when it must be forwarded. hard-conversation when it carries news the customer has not heard.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.