agentleFS
prateeks367prateeks367/voicemoat-skills/skills/audit-before-i-publish/SKILL.md

audit-before-i-publish

Skill0 starsChanged 42 days agoMITView on GitHub

Run a finished draft through one pre-publish pass and rank what to fix. Use when someone says a post is ready, asks for a final check, asks whether to post something as it stands, or pastes a draft and asks what is wrong with it. Do not use it to write a post from nothing, do not rewrite a whole draft when one line is the problem, and do not predict performance.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    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:

    One idea: If there are two, the second is next week's post, not a paragraph.

    line 42
  2. Distinctive:

    Only propose a change when the current ending is actually costing something.

    line 59
  3. Distinctive:

    Read it back for the tells: hedging that was not in the first draft, sentences all drifted to the same length, vocabulary this person does …

    line 64
  4. Distinctive:

    Never make the second call on your own.

    line 103

≈ “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

Audit it before I publish

Editing makes a draft better. An audit decides whether it is ready, which is a different question with a different answer. Most drafts that get stuck do not need another pass of polish. They need one specific thing fixed, and the job here is to find which one and then stop. Work the steps in order.

Step 1: read it in the order a reader meets it

The author already knows what the post says, so they read it as a whole. Nobody else does.

  • The first line on its own, everything else hidden. Does it make the second line necessary? If the honest answer is "it sets up the post", that is a no.
  • Whatever sits above the fold. LinkedIn hides most of a post behind "see more". Everything above that cut is the advert for the rest, so judge it as an advert rather than as an introduction.

Step 2: check the claim before you check the craft

This comes first because it is the only failure that is expensive. A dull post costs you one day. A wrong one costs you the reason people read you.

  • Every number. Can it be sourced, and would you be happy to name that source in the comments? A figure everyone repeats is not a figure anyone checked. If it will not hold, cut it or attribute it.
  • Every story. Is it yours, and did it happen the way it is written? A secondhand anecdote told in the first person loses everyone who recognises it.
  • The promise. Does the opening commit to something the body delivers? A hook the post does not pay off costs more than a flat hook, because the reader who felt the gap does not come back.
  • Anyone identifiable. A colleague, a client, a number from inside a company. Did they agree to be in it?

Step 3: check the shape

  • One idea. If there are two, the second is next week's post, not a paragraph. Name it and take it out.
  • The line breaks. A wall of text loses readers who would have read the same words with air around them.
  • The length. Long is fine when every paragraph earns its place, so never say "this is too long". Say which paragraph is carrying nothing.

Step 4: check the ending, and be willing to leave it alone

Most posts end by asking a question nobody answers. Name which ending this one has before suggesting anything:

  • A real question the reader already has an opinion about
  • A generic prompt, the "thoughts?" ending, which reads as asking a favour
  • A summary that repeats what they just read
  • Nothing at all, which is more often right than people expect

Only propose a change when the current ending is actually costing something. "Add a call to action" is advice, not a diagnosis.

Step 5: check it still sounds like the person who wrote it

Read it back for the tells: hedging that was not in the first draft, sentences all drifted to the same length, vocabulary this person does not use. Point at the lines. "It sounds a bit AI" is not usable.

Step 6: rank, then stop

Give at most three fixes, ordered by what they cost, and say which checks came back clean. Then say plainly whether you would post it as it stands, and be willing to answer yes. An audit that always finds three problems is a habit, not an audit.

What this will not do

It will not tell you whether the post will perform. Reach depends on timing, audience, format and luck as much as on the words, and a checklist implying otherwise would be selling something.

It also will not give you a number. Reading a draft against five posts you pasted is a read-through, not a measurement, and a score out of 100 would claim a precision it does not have.

What changes when VoiceMoat is connected

Steps 1 to 4 are craft, and craft is the same for everybody. Step 5 is the one that needs to know you, and on its own it is guessing:

  • get_voice_profile returns a profile trained on your own published posts, held separately for LinkedIn and for Twitter, rather than a sketch built from whatever you happened to paste.
  • score_voice_match checks the draft against that profile in two parts: Style, how many of your measurable habits it matches (N of M checks), and Stance, whether what it says fits your views (On voice, Check one line or Off voice), with the lines that are off. The assistant may show it as a card. This is the step the free version structurally cannot do, so it is where the audit stops being an opinion.
  • improve_post takes the draft through a deeper pass on the occasions when step 6 turns up three real fixes rather than one.
  • publish_post sends it once you are happy, behind a preview and a one-time confirmation.

Never make the second call on your own. Show the person the exact preview the first call returns and wait for a clear yes before you send the one-time code back. Never make both calls in one step. The assistant may show the preview as a card with its own Post now or Schedule button. A press on that button is the person's yes and sends the code itself, so do not send it again.

VoiceMoat is at voicemoat.com. The connector needs the Pro or Enterprise plan.

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.