LinkedIn Post Validator

Linkedin post audit

by sergebulaev · MIT license · ★ 3,059 on GitHub · updated Sep 18

Run any post draft through the 2026 heuristic checklist.

# LinkedIn Post Audit

Run any post draft through the 2026 heuristic checklist. Catches AI tells, timing/format issues, length violations, and structural weaknesses before publishing.

## When to use

- Before publishing a hand-written or AI-drafted post
- When `linkedin-post-writer` finishes a draft (auto-invoked)
- When a recent post didn't land and the user wants a post-mortem

## Input

- A post draft (plain text)
- Optional: target audience, scheduled time, format (text / carousel / video / image)

## Output

- **Pass/Fail** header

Overview

Run any post draft through the 2026 heuristic checklist. Catches AI tells, timing/format issues, length violations, and structural weaknesses before publishing.

What you get

Pass/Failheader
Blockers(must fix before publishing): em dash density over the cap, paragraphs at 3+ AI markers, reveal bridges, external links in body
Warnings(ship-risky): staccato stacks, sincerity markers, missing referenced numbers, generic close
Score estimatesper-paragraph tell density, approximate first-hour reach fit. No detector score: on 100-300 word text those are noise and the skill does not promise to beat them
Suggested fixesinline rewrites for each issue
Timing recommendationbest window given audience

Manual install (npx, ZIP)

Claude Codenpx skills add sergebulaev/linkedin-skills --skill linkedin-post-audit -g -a claude-code

Claude.ai / desktopZip the folder and upload it under Settings → Capabilities → Skills.

Codex, Cursor, other agentsPoint them at SKILL.md. The other files are plain text.

Connects with Agent Alley MCP · agentalley.io/mcp

Files included

post-audit.md4.9 KB

Read post-audit.md here

sergebulaev/main1 file

post-audit.md4.9 KB

post-audit.md

Show the full text80 lines

linkedin-post-audit/post-audit.md80 lines · 4.9 KB

PreviewCode

Outline

RawView on GitHub

LinkedIn Post Audit

Run any post draft through the 2026 heuristic checklist. Catches AI tells, timing/format issues, length violations, and structural weaknesses before publishing.

When to use

Before publishing a hand-written or AI-drafted post
When linkedin-post-writer finishes a draft (auto-invoked)
When a recent post didn't land and the user wants a post-mortem

Input

A post draft (plain text)
Optional: target audience, scheduled time, format (text / carousel / video / image)

Output

Pass/Fail header
Blockers (must fix before publishing): em dash density over the cap, paragraphs at 3+ AI markers, reveal bridges, external links in body
Warnings (ship-risky): staccato stacks, sincerity markers, missing referenced numbers, generic close
Score estimates: per-paragraph tell density, approximate first-hour reach fit. No detector score: on 100-300 word text those are noise and the skill does not promise to beat them
Suggested fixes: inline rewrites for each issue
Timing recommendation: best window given audience

Checks

Blockers (auto-fail)

Em dash density above ~1 per 100 words (1-2 per post); en dash between clauses; double dash. A single em dash is not a blocker
External link in body (not in first comment)
Post exceeds 3,000 chars (LinkedIn hard limit)
Opens with "In today's fast-paced world...", a reveal bridge ("Here's what", "Stop X, start Y"), or a sincerity announcement ("Let me be honest")
Ends with "What do you think?", "Thoughts?", "Let that sink in."
Any paragraph with 3+ vocabulary / grammar markers, or any negative-parallelism / "The result?" reveal bridge (see ../references/audit-ai-tells.md)
Frames LinkedIn as inferior in a LinkedIn post (algo penalty)

Warnings (flag with suggested fix)

Hook doesn't fit in first 210 chars (mobile …see more cutoff)
Length outside 900-1,300 sweet spot (or 1,500-1,900 for long-form with breaks)
A paragraph that reads machine-flat (4+ sentences all the same length, no clause doing work). Flag that paragraph only; sentence-length variance is not a reach lever on LinkedIn, so never suggest adding variance as a tactic
No odd-precision number with a named referent (a bare number does not clear this)
No named entity
No first-person sensory detail
Stacked or perfectly parallel rule-of-three, or 3+ triads in the post (one natural triad passes)
More than 2 hashtags
User's own product named more than once
Missing reaction-prompting moment: a specific, dated, uncomfortable fact stated flat, or an opinion with stakes. A framed confession ("I'll be honest, this hurt") does not clear this; the frame is the tell
Passive voice >10% 18a. Staccato stacks ("Short. Punchy. Done.", "No X. No Y. Just Z.", "All the X. None of the Y."), one-word paragraphs, more than 2 standalone fragments, or a long/short/long/short seesaw 18b. Hedging stack or sincerity marker mid-post ("perhaps", "it seems", "honestly?", "real talk") 18c. Over-scrubbed: uniformly flat tone, zero em dashes and zero triads in a long post, no reaction or opinion anywhere
First line is not a complete standalone hook (it needs line 2 to make sense). 2026 corpus: every top post front-loads a full hook before the fold.
No blank line after the hook / wall-of-text open. Winners use heavy whitespace: one idea per line, blank line after the hook.
Emoji sprinkled mid-text in a narrative post, or more than 2-3 total in prose. Top posts front-load 1-2 meaningful emoji; serious/contrarian posts use zero. Exempt: structured glossary/list formats (e.g. F15 Explain-to-Kids) where one emoji anchors each line on purpose.
Comment-gate ("comment X and I'll DM you...") in a post whose goal is thought leadership. Organic top performers use zero hard comment-gates; only flag-clear when the post's goal is list-building (then F6 is intentional).
No clear primary goal: the post chases comments, reposts, likes, and saves all at once. Pick one (see ../../../references/hook-formulas.md "Engagement-goal split").

Info (neutral notes)

Suggested posting time given audience
Format recommendation (text / carousel / video) given topic
Similar-hook detection: if this post's first 100 chars match a recent post

Steps

Parse draft into sentences, paragraphs, first-210-char hook.
Run each blocker check; collect failures.
If any blockers, return FAIL with specific fix suggestions; optionally offer auto-rewrite.
If no blockers, run warnings.
Report per-paragraph tell density (markers per paragraph, em dashes per 100 words, fragment count, triad count). Do not estimate a detector score.
Return structured report.

Example

See ../references/audit-examples.md for worked examples.

linkedin-humanizer — aggressive rewrite if audit fails
linkedin-post-writer — regenerate draft using a proven formula

1	# LinkedIn Post Audit
2
3	Run any post draft through the 2026 heuristic checklist. Catches AI tells, timing/format issues, length violations, and structural weaknesses before publishing.
4
5	## When to use
6
7	- Before publishing a hand-written or AI-drafted post
8	- When `linkedin-post-writer` finishes a draft (auto-invoked)
9	- When a recent post didn't land and the user wants a post-mortem
10
11	## Input
12
13	- A post draft (plain text)
14	- Optional: target audience, scheduled time, format (text / carousel / video / image)
15
16	## Output
17
18	- Pass/Fail header
19	- Blockers (must fix before publishing): em dash density over the cap, paragraphs at 3+ AI markers, reveal bridges, external links in body
20	- Warnings (ship-risky): staccato stacks, sincerity markers, missing referenced numbers, generic close
21	- Score estimates: per-paragraph tell density, approximate first-hour reach fit. No detector score: on 100-300 word text those are noise and the skill does not promise to beat them
22	- Suggested fixes: inline rewrites for each issue
23	- Timing recommendation: best window given audience
24
25	## Checks
26
27	### Blockers (auto-fail)
28	1. Em dash density above ~1 per 100 words (1-2 per post); en dash between clauses; double dash. A single em dash is not a blocker
29	2. External link in body (not in first comment)
30	3. Post exceeds 3,000 chars (LinkedIn hard limit)
31	4. Opens with "In today's fast-paced world...", a reveal bridge ("Here's what", "Stop X, start Y"), or a sincerity announcement ("Let me be honest")
32	5. Ends with "What do you think?", "Thoughts?", "Let that sink in."
33	6. Any paragraph with 3+ vocabulary / grammar markers, or any negative-parallelism / "The result?" reveal bridge (see `../references/audit-ai-tells.md`)
34	7. Frames LinkedIn as inferior in a LinkedIn post (algo penalty)
35
36	### Warnings (flag with suggested fix)
37	8. Hook doesn't fit in first 210 chars (mobile `…see more` cutoff)
38	9. Length outside 900-1,300 sweet spot (or 1,500-1,900 for long-form with breaks)
39	10. A paragraph that reads machine-flat (4+ sentences all the same length, no clause doing work). Flag that paragraph only; sentence-length variance is not a reach lever on LinkedIn, so never suggest adding variance as a tactic
40	11. No odd-precision number with a named referent (a bare number does not clear this)
41	12. No named entity
42	13. No first-person sensory detail
43	14. Stacked or perfectly parallel rule-of-three, or 3+ triads in the post (one natural triad passes)
44	15. More than 2 hashtags
45	16. User's own product named more than once
46	17. Missing reaction-prompting moment: a specific, dated, uncomfortable fact stated flat, or an opinion with stakes. A framed confession ("I'll be honest, this hurt") does not clear this; the frame is the tell
47	18. Passive voice >10%
48	18a. Staccato stacks ("Short. Punchy. Done.", "No X. No Y. Just Z.", "All the X. None of the Y."), one-word paragraphs, more than 2 standalone fragments, or a long/short/long/short seesaw
49	18b. Hedging stack or sincerity marker mid-post ("perhaps", "it seems", "honestly?", "real talk")
50	18c. Over-scrubbed: uniformly flat tone, zero em dashes and zero triads in a long post, no reaction or opinion anywhere
51	19. First line is not a complete standalone hook (it needs line 2 to make sense). 2026 corpus: every top post front-loads a full hook before the fold.
52	20. No blank line after the hook / wall-of-text open. Winners use heavy whitespace: one idea per line, blank line after the hook.
53	21. Emoji sprinkled mid-text in a narrative post, or more than 2-3 total in prose. Top posts front-load 1-2 meaningful emoji; serious/contrarian posts use zero. Exempt: structured glossary/list formats (e.g. F15 Explain-to-Kids) where one emoji anchors each line on purpose.
54	22. Comment-gate ("comment X and I'll DM you...") in a post whose goal is thought leadership. Organic top performers use zero hard comment-gates; only flag-clear when the post's goal is list-building (then F6 is intentional).
55	23. No clear primary goal: the post chases comments, reposts, likes, and saves all at once. Pick one (see `../../../references/hook-formulas.md` "Engagement-goal split").
56
57	### Info (neutral notes)
58	24. Suggested posting time given audience
59	25. Format recommendation (text / carousel / video) given topic
60	26. Similar-hook detection: if this post's first 100 chars match a recent post
61
62	## Steps
63
64	1. Parse draft into sentences, paragraphs, first-210-char hook.
65	2. Run each blocker check; collect failures.
66	3. If any blockers, return FAIL with specific fix suggestions; optionally offer auto-rewrite.
67	4. If no blockers, run warnings.
68	5. Report per-paragraph tell density (markers per paragraph, em dashes per 100 words, fragment count, triad count). Do not estimate a detector score.
69	6. Return structured report.
70
71	## Example
72
73	See `../references/audit-examples.md` for worked examples.
74
75
76	## Related skills
77
78	- `linkedin-humanizer` — aggressive rewrite if audit fails
79	- `linkedin-post-writer` — regenerate draft using a proven formula
80