linkedin-audit
Skill1 starsChanged 23 days agoMITView on GitHub
Look back at what you posted on LinkedIn for one business and say, with real numbers, what to repeat, what to drop, and how sure we are. A local script ranks each post against your own median (not raw impressions), then finds what the top posts share by type, topic, hook pattern and length, and hands three changes to next week's plan. Use it when someone says "audit my LinkedIn", "what's working on my LinkedIn", "which posts did best", "why are my posts flopping", "review last month's posts", or pastes or exports post numbers. For why one pasted post beat another on its writing, use linkedin-viral. Never scrapes; you export or type the numbers.
In this repository15 other files it gives its agents.
- founder-contextSkill
- humanizeSkill
- linkedin-carouselSkill
- linkedin-commentSkill
- linkedin-dmSkill
- linkedin-hooksSkill
- linkedin-inboxSkill
- linkedin-planSkill
- linkedin-postSkill
- linkedin-profileSkill
- linkedin-replySkill
- linkedin-repurposeSkill
- linkedin-viralSkill
- talking-head-cardsSkill
- transcribeSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 39
Never log in, scrape, or browse LinkedIn for the numbers.
- Distinctive: line 57
Write it for a busy founder:
- Very specific: line 68
Comments: if comments.md or warm-leads.csv exist, note which posts started real conversations or leads.
- Very specific: line 70
Hooks: if hooks.md exists, compare hook scores with results and say whether the scoring looks useful for this audience.
- Distinctive: line 73
Never fill a missing number, never round a small sample into a trend, and never compare with "industry benchmarks" unless the user gives a …
- Distinctive: line 95
Never scrape or log in.
- Distinctive: line 96
Never invent or estimate a missing number.
- Distinctive: line 97
Never mix businesses in one ranking without labels.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
linkedin-audit
Turns what was posted into what to post next. Every number comes from the user's own analytics. Missing numbers stay missing.
Tool paths: <this skill> is the folder this SKILL.md is in (the "base directory" you
were given). Sibling skills sit next to it, so <this skill>/../linkedin-hooks/ is the
hooks skill. Run tools with full paths built from that folder, not from the current
working directory, or they will look missing.
1. Business and numbers
Find the business: a named one, else founder/context.md, else ~/.founder/active
and ~/.founder/businesses/<slug>.md. If none exist, ask once and offer
founder-context. Audit one business at a time. If the user wants an account-wide view,
label every post with its business.
Get the numbers:
~/.founder/social/linkedin/<slug>/posts.csv(written by linkedin-post), with impressions, reactions, comments, reposts, and saves or sends if they have them, typed in from LinkedIn analytics; or- a CSV they export from LinkedIn's post analytics; or
- numbers pasted in chat. Put them into a CSV with the columns
date,type,topic,pattern,first_line,chars,impressions,reactions,comments,reposts,saves,sends.
Never log in, scrape, or browse LinkedIn for the numbers. If the user has none, explain where to find them (each post's analytics, or the Analytics tab) and stop.
2. Run the math
python3 <this skill>/scripts/audit.py <posts.csv>It computes, per post: engagement rate (reactions + comments + reposts + saves + sends, over impressions), the rate compared with the account's own median, reach compared with the median, and a weighted rate where comments count 3x and reposts, saves and sends 2x. It groups by type, topic, hook pattern and length band. Rows with no impressions are listed as "no numbers yet" and left out. Do not do this math by eye. If Python is not available, say so and show your arithmetic step by step.
3. Read the results
Write it for a busy founder:
- Top 3 and bottom 3 posts, copied from the script's ranking, in its order (bottom 3 are the last three rows). Do not re-rank by eye or retype numbers; paste the script's lines. Say which rate you ranked by (the script sorts by weighted rate).
- What the top posts share: type, topic, pattern, length, day. Only claim a pattern if at least 2 of the top posts share it and the bottom ones mostly do not.
- How sure we are: under 8 posts with numbers, say "early hints, not findings". Name other things that could explain a result (a holiday, a repost by a big account, a different time).
- Comments: if
comments.mdorwarm-leads.csvexist, note which posts started real conversations or leads. That matters more than reactions. - Hooks: if
hooks.mdexists, compare hook scores with results and say whether the scoring looks useful for this audience.
Never fill a missing number, never round a small sample into a trend, and never compare with "industry benchmarks" unless the user gives a source.
4. Three changes
End with exactly three changes for the next plan, each tied to evidence:
1. More <x>: posts 3 and 7 did 2.4x and 1.9x the median rate, both <x>.
2. Less <y>: the 3 <y> posts sat under 0.6x.
3. Try <z>: not enough data on <z>; plan 2 posts to find out.5. Save and hand off
Save to ~/.founder/social/linkedin/<slug>/audit.md with the date, the script output, the
reading, and the three changes. Use the Write tool or Bash, then confirm the file exists
(for example ls -l on it). linkedin-plan reads this file. Tell the user:
"linkedin-plan will use these three changes next week."
Never
- Never scrape or log in.
- Never invent or estimate a missing number.
- Never mix businesses in one ranking without labels.
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
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.

