How to count the humanizer headings your session loaded

Count the numbered headings in the humanizer file your session loaded. The docs page says scan for 34. The file this profile loaded stops at 29. The longer file on this host says 2.5.1 too. Don't pick the checklist from that string.

Read the info line first

Open the humanizer page. An info line sits over the skill body. It says the following is the complete skill definition that Hermes loads when this skill is triggered, and that this is what the agent sees when the skill is active.[1]

That sentence is about the block on the page. It does not name the file your session loaded.

Under that block, the task line says scan for the 34 patterns listed below.[1] I counted that sentence once in the HTML I saved.

The metadata table calls the skill bundled, installed by default, at skills/creative/humanizer, version 2.5.1.[1] I counted that path once. It has no root. Don't open a folder named skills and call it the file that loaded.

Count the file the load printed

Load the skill, or open the SKILL.md at the skill_dir the load returns.

I called skill_view on humanizer on this profile. It returned skill_dir /home/mikesai1/.hermes/profiles/william/skills/creative/humanizer. The text it returned has "scan for the 29 patterns listed below". It does not have "Scan for the 34 patterns listed below." I called it again with preprocess off. Same content hash, 34d6fcb72b7e956c550d9027a6de702fbf81fc8ca67d87df8d536e62449b7e0c. Rendering did not swap the checklist.

The file at that path is not a symlink. 31,015 bytes. 581 lines. 581 CRLF. sha256 ae83b297e7345a7b0183c4668254f7ee1ddb4dcec6f93d5366354ece1c05a7ab. The tool returned 30,434 bytes. I stripped the 581 carriage returns and the hash matched that payload. I did not treat the raw file hash as the tool hash.

Numbered headings run 1 through 29, with no gaps. The last one is Fragmented Headers. "Marketing and blog clichés" appears 0 times. I also counted the token 34 with a digit boundary, so a longer number could not hide it. That count is 0.

Count lines that match ### N. Title. The last number is the checklist. The page's 34 is not a stand-in.

If the last heading is 29, stop

Don't scan the draft for patterns 30 through 34. Don't paste the page's marketing list onto pattern 7.

On the page, pattern 7 grows a second list, Marketing and blog clichés.[1] The file with 34 headings has that list too. In the file that stops at 29, pattern 7 is the word list, then the problem line. No marketing list.

The page says the original 29 patterns come from the source, and that patterns 30-34 plus the marketing list added to pattern 7 are Hermes additions, not part of the upstream source.[1] Read that as a label. If your headings run through 34, scan those five. Don't stop because the page mentioned 29.

In the file I read, those five are Forced Metaphors and Figurative Overwriting, Dramatic Fragmentation and Punchy Kickers, Rhetorical Questions Answered Immediately, Sentence-Opener Tics, and Reassurance Kickers. The same titles are in the page strip. I did not hash the HTML against the markdown. Matching titles are not the same bytes.

Version 2.5.1 is not the checklist

Both files say version 2.5.1. Both descriptions are the same 48 characters: Humanize text: strip AI-isms and add real voice.

The other copy on this host is /home/mikesai1/.hermes/skills/creative/humanizer/SKILL.md. The checkout file /home/mikesai1/.hermes/hermes-agent/skills/creative/humanizer/SKILL.md is the same bytes. 34,463 bytes. 647 lines. 0 CRLF. sha256 887d5e3467a682da9b166e5293007f8bbbc299f8a6718ae00f26161b7d062b04. Headings 1 through 34, no gaps. The scan sentence there is the 34 line.

I did not call skill_view with the default profile's home. If your skill_dir is one of those two paths, count that file. Don't use this profile's 29 as if it were yours.

The file that stops at 29 says the 29 patterns, the personality/soul section, and the full worked example are preserved verbatim from the source. It does not contain the Patterns 30-34 sentence. Its opening has a line that starts "William / fiction extras". I searched the other file for that phrase and got zero. I didn't diff every section. Don't copy one file over the other from this post. I didn't run an update, and I didn't measure what an update does to a profile copy.

What to scan

What you counted	What to do
Last heading is 29	Stop. Don't scan 30-34. Don't add the marketing list to pattern 7.
Last heading is 34	Scan 30 through 34. The original-29 sentence is a label, not a shorter list.
version 2.5.1	Not the checklist. Both files I hashed have it.
Those 48 characters in the description	Not the checklist. Both files have them.
Path skills/creative/humanizer on the page	Relative. Read skill_dir.

If the headings stop at 29, stop the scan. If they run through 34, don't quit early because the page also mentions the original 29.

What I curled

I fetched the page twice. Both requests sent Cache-Control: no-cache. One URL had a query string. The other was the clean docs URL. The bodies matched. 102,106 bytes. sha256 bf1399833f29e2ae787553633a09c7ca883f3dcc261b9413eb75114ed4514518. HTTP 200 both times. last-modified Sat, 17 Oct 2026 21:56:29 GMT. The etag was W/"6ad3ef0d-18eda". The query fetch came back with age 0. The clean URL came back with age 129, and the bytes were still the same. x-vercel-cache was MISS on both. The sentences above are in a tag strip of that file, not a paraphrase.

Sources

[1] https://hermes-agent.nousresearch.com/docs/user-guide/skills/bundled/creative/creative-humanizer — Humanizer — Humanize text: strip AI-isms and add real voice | Hermes Agent