<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Robin Beraud-Sudreau</title>
    <description>The latest articles on DEV Community by Robin Beraud-Sudreau (@robinberaud).</description>
    <link>https://dev.to/robinberaud</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3992645%2F05600fad-ea1a-406a-b264-8e6525d46f78.png</url>
      <title>DEV Community: Robin Beraud-Sudreau</title>
      <link>https://dev.to/robinberaud</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/robinberaud"/>
    <language>en</language>
    <item>
      <title>Chapters and quotes from a transcript, without rewatching your video</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Thu, 19 Nov 2026 10:02:37 +0000</pubDate>
      <link>https://dev.to/robinberaud/chapters-and-quotes-from-a-transcript-without-rewatching-your-video-109m</link>
      <guid>https://dev.to/robinberaud/chapters-and-quotes-from-a-transcript-without-rewatching-your-video-109m</guid>
      <description>&lt;p&gt;You finish a 40-minute video and the chores begin: scroll the timeline to write chapters, hunt for a line worth posting, remember what you actually said in minute 23. It is slow, and it is the part most people skip.&lt;/p&gt;

&lt;p&gt;I wrote a small Hermes skill, &lt;code&gt;transcript-to-chapters&lt;/code&gt;, to take over the mechanical half. You give it the subtitle file your editor or platform already exports (SRT or VTT). A tiny script with no dependencies and no API key reads the timestamps, compares neighbouring stretches of speech and flags where the vocabulary changes, which usually means a new topic. It also lists sentences that could work as quotes.&lt;/p&gt;

&lt;p&gt;The agent does the part that needs judgement: it writes short, honest chapter titles, moves a boundary to the sentence that really starts the topic, and drafts a short summary using only what was said. Quotes are copied word for word with their timestamp, never reworded.&lt;/p&gt;

&lt;p&gt;Before you paste anything, it checks the rules YouTube applies to chapters: the first one starts at 0:00, there are at least three, and each lasts at least ten seconds.&lt;/p&gt;

&lt;p&gt;Example result for a baking video:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;0:00 Starter and flour basics
2:10 Folding the dough
3:00 Preheating and baking
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Honest limits: the boundaries are suggestions based on word counts, so short or uniform talks may need manual fixes, and auto-captions contain mistakes you should check. I tested it on a small sample and a self-test, not yet on hours of real recordings.&lt;/p&gt;

&lt;p&gt;The skill is MIT licensed in the hermes-skills library.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/media/transcript-to-chapters" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/media/transcript-to-chapters&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>media</category>
    </item>
    <item>
      <title>Your analytics export says more than you think (if you read it honestly)</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Wed, 18 Nov 2026 10:02:03 +0000</pubDate>
      <link>https://dev.to/robinberaud/your-analytics-export-says-more-than-you-think-if-you-read-it-honestly-4a9k</link>
      <guid>https://dev.to/robinberaud/your-analytics-export-says-more-than-you-think-if-you-read-it-honestly-4a9k</guid>
      <description>&lt;p&gt;Every creator has that moment: you download the CSV from your dashboard, open it, and stare at forty rows of numbers. Which posts worked? Is the channel actually improving, or did one lucky video skew everything?&lt;/p&gt;

&lt;p&gt;I built a Hermes skill, creator-analytics-digest, for exactly this. You give your agent the exported file. A small script does the counting, because arithmetic is where chatty assistants make mistakes. It finds the title, date and metric columns on its own (English, French and Spanish headers work), ignores "Total" rows, and understands values like 1,200, 4.1% or 2.4K.&lt;/p&gt;

&lt;p&gt;Then the agent writes the digest: what worked, what lagged, the trend, and three follow-up ideas, each tied to a real row. The skill tells it to label reasons as guesses, since an export cannot tell you why people clicked.&lt;/p&gt;

&lt;p&gt;A short example from a fictional knitting channel:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Median views: 1,500&lt;/li&gt;
&lt;li&gt;Standout: "Sweater pattern", 3,500 (more than twice the median)&lt;/li&gt;
&lt;li&gt;Median of the second half of the period: 91% above the first half&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The suggestion: a follow-up with another concrete pattern, flagged as a hypothesis to test, not a rule.&lt;/p&gt;

&lt;p&gt;Nothing leaves your machine: no API key, no upload, only Python's standard library. Limits are stated in the skill: a few dozen rows give noisy results, and I only tested it on small fictional files, not on every platform's real export.&lt;/p&gt;

&lt;p&gt;Try it on your last export.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/data-analysis/creator-analytics-digest" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/data-analysis/creator-analytics-digest&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>dataanalysis</category>
    </item>
    <item>
      <title>"#collab" is not a disclosure: check your sponsored posts before they go live</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Tue, 17 Nov 2026 10:01:33 +0000</pubDate>
      <link>https://dev.to/robinberaud/collab-is-not-a-disclosure-check-your-sponsored-posts-before-they-go-live-fi6</link>
      <guid>https://dev.to/robinberaud/collab-is-not-a-disclosure-check-your-sponsored-posts-before-they-go-live-fi6</guid>
      <description>&lt;p&gt;You write a caption for a brand deal, add "#collab #partner" with ten other hashtags, and hit publish. Is the sponsorship obvious to your audience? Often it isn't. A vague tag, or a clear one that only appears after "see more", is easy to miss.&lt;/p&gt;

&lt;p&gt;I built a small Hermes skill, sponsored-post-disclosure-check, to catch that before publishing. You give it your caption and (ideally) the rules from your brief or contract. It then reports:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;whether a clear word such as "ad" or "sponsored" is present, or only vague ones&lt;/li&gt;
&lt;li&gt;how early it appears, per platform, using limits you can edit&lt;/li&gt;
&lt;li&gt;whether it is lost inside a block of hashtags&lt;/li&gt;
&lt;li&gt;whether the brand is named&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Example on a caption with a 159-character lead-in and a hashtag pile:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Verdict: WARN
[WARN] late: First clear term '#ad' starts at character 159 ... within the first 125
[WARN] buried: The disclosure hashtag sits inside a block of 7 hashtags
[MANUAL] native_label: Check by hand: turn on the platform's paid-partnership label
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then the agent rewrites the caption so the disclosure comes first, in your own voice.&lt;/p&gt;

&lt;p&gt;Honest limits: the default terms and thresholds are placeholders, not official rules, and it only reads text, not videos or labels. It is a checklist, not legal advice. Replace the defaults with your brief or local guidance. It needs no API key, only Python 3.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/sponsored-post-disclosure-check" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/sponsored-post-disclosure-check&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>Answer your comments faster without sounding like a bot</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Mon, 16 Nov 2026 10:02:39 +0000</pubDate>
      <link>https://dev.to/robinberaud/answer-your-comments-faster-without-sounding-like-a-bot-26g6</link>
      <guid>https://dev.to/robinberaud/answer-your-comments-faster-without-sounding-like-a-bot-26g6</guid>
      <description>&lt;p&gt;When a video does well, the comment section turns into homework. Some people ask real questions, some say thank you, a few are unhappy, and plenty is spam. Answering all of it in one sitting makes replies sound flat.&lt;/p&gt;

&lt;p&gt;The comment-reply-assistant skill for Hermes splits the job in two. A small script sorts an exported comments file (CSV or plain text) into questions, praise, criticism, spam and other. It also flags the comments that need a careful human answer. Then the agent drafts a reply for each row in your voice, using a few replies you have written before as a reference.&lt;/p&gt;

&lt;p&gt;Nothing is posted. Every row stays marked &lt;code&gt;draft&lt;/code&gt; until you approve it, and anything the agent cannot know (a price, a date, a personal detail) is left as a &lt;code&gt;[YOU: fill in ...]&lt;/code&gt; placeholder instead of an invented answer. Critical comments get a calm draft and a warning, and abusive ones get no reply at all.&lt;/p&gt;

&lt;p&gt;A tiny example, with four made-up comments:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;"Can you do one on lighting?" becomes a question, with a draft that says yes and leaves the date for you.&lt;/li&gt;
&lt;li&gt;"Great video, I learned a lot!" becomes praise, with a short and specific thank-you.&lt;/li&gt;
&lt;li&gt;"Too long, clickbait title." becomes criticism, marked for extra care.&lt;/li&gt;
&lt;li&gt;"sub4sub follow me" becomes spam, no reply.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The sorting is based on simple keyword rules in English and French, so it will misread irony and mixed messages. Treat it as a first pass, not a judge. It needs no account and no API key, only Python 3.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/comment-reply-assistant" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/comment-reply-assistant&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>The brand pitch that doesn't need hype: write it from facts you can show</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Sun, 15 Nov 2026 10:03:04 +0000</pubDate>
      <link>https://dev.to/robinberaud/the-brand-pitch-that-doesnt-need-hype-write-it-from-facts-you-can-show-3n7i</link>
      <guid>https://dev.to/robinberaud/the-brand-pitch-that-doesnt-need-hype-write-it-from-facts-you-can-show-3n7i</guid>
      <description>&lt;p&gt;Most pitches to brands fail for the same reasons: they are long, they say "I love your brand" without saying why, and they add big numbers nobody can check. The reader has a full inbox and decides in seconds.&lt;/p&gt;

&lt;p&gt;I built a small Hermes skill, brand-deal-pitch, that fixes the structure and keeps you honest. You give your agent the facts: why this brand specifically, the audience figures you want to quote (each with a date and source), one to three content ideas, what the brand receives, and your rate if you choose to share it. It assembles a short email, plus a follow-up for a week later.&lt;/p&gt;

&lt;p&gt;The useful part is what it refuses to do. It will not invent a rate. It will not build the email without a specific reason for the brand. It stops on leftover placeholders and future dates, and warns about words like "guaranteed" or "viral", figures older than 90 days, numbers with no dated source, and a missing line saying paid content is labelled as an ad.&lt;/p&gt;

&lt;p&gt;Example, with fictional data: Alex makes finance videos and pitches a budgeting app. The email gives a real reason (using its envelope feature since spring), quotes "12,400 YouTube subscribers (as of 2026-11-01, YouTube Studio)", proposes one video plus clips, states the rate Alex provided, and ends with one ask: a 15-minute call.&lt;/p&gt;

&lt;p&gt;No API key, no technical skills: your agent runs one small Python script. It sends nothing and can't verify your numbers, so read the result before it goes out.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/brand-deal-pitch" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/brand-deal-pitch&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>Your media kit shouldn't contain a number you can't prove</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Sat, 14 Nov 2026 10:01:57 +0000</pubDate>
      <link>https://dev.to/robinberaud/your-media-kit-shouldnt-contain-a-number-you-cant-prove-3g5h</link>
      <guid>https://dev.to/robinberaud/your-media-kit-shouldnt-contain-a-number-you-cant-prove-3g5h</guid>
      <description>&lt;p&gt;Brands ask for a media kit, and most creators spend an evening on fonts. The worse part is the numbers: followers from last spring, an engagement rate remembered from "somewhere", a rounded-up view count. A brand that checks your profile finds the gap, and it costs you trust.&lt;/p&gt;

&lt;p&gt;I built a small Hermes Agent skill that handles the part that matters: the facts. You give the agent your real figures, and it produces a clean one-page HTML media kit that you print to PDF from your browser.&lt;/p&gt;

&lt;p&gt;The rules are strict on purpose:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Every audience figure needs a date and a source ("YouTube Studio", "Instagram Insights"). No date, no build.&lt;/li&gt;
&lt;li&gt;Figures older than 90 days trigger a warning.&lt;/li&gt;
&lt;li&gt;Leftover placeholders like TODO stop the build, so you never send a kit with "TODO bio".&lt;/li&gt;
&lt;li&gt;Sections with no data simply don't appear.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A check looks like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;WARN: platforms[1]: figure is 112 days old, refresh before sending
wrote media-kit.html (1 warning(s))
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The kit shows each figure as "12,400 subscribers, as of 2026-11-01, source: YouTube Studio", so brands see how fresh your data is. There's also a line for your disclosure policy, such as "paid content is always labelled as an ad".&lt;/p&gt;

&lt;p&gt;It uses only Python's standard library, needs no account or internet, and ships with a self-test. It can't verify your numbers are true; that stays on you. It only makes sure they are dated, sourced, and copied exactly as you gave them.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/media-kit-builder" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/media-kit-builder&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>A 30-day content calendar you can actually stick to</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Fri, 13 Nov 2026 10:02:42 +0000</pubDate>
      <link>https://dev.to/robinberaud/a-30-day-content-calendar-you-can-actually-stick-to-1b2n</link>
      <guid>https://dev.to/robinberaud/a-30-day-content-calendar-you-can-actually-stick-to-1b2n</guid>
      <description>&lt;p&gt;Most content calendars die in week two. Not because ideas run out, but because the plan was too ambitious or lopsided.&lt;/p&gt;

&lt;p&gt;I built a small Hermes skill, &lt;code&gt;content-calendar-30-days&lt;/code&gt;, that splits the job in two. A tiny script handles the boring mechanics: pick your posting days, rotate through your three pillars so you never post the same theme twice in a row, alternate platforms, vary formats, and reserve a few "recycling" slots. Then the agent does the thinking part with you: real topics, written as promises to your audience.&lt;/p&gt;

&lt;p&gt;You start by giving it three pillars (say Teach, Story, Proof), your platforms, and how many posts per week you can really sustain. The skeleton looks like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;2026-12-02 | Wed | instagram | Teach | carousel | TODO
2026-12-05 | Sat | linkedin  | Story | text post | TODO
2026-12-12 | Sat | instagram | Story | single image | TODO: new angle on a past post
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Every fifth post is a recycling slot: you point to one of your own older posts that did well and give it a new format or angle. The skill never invents past posts; if you have none to offer, the slot stays yours to fill.&lt;/p&gt;

&lt;p&gt;A final check flags empty topics, duplicates and back-to-back pillars, so you hand over a calendar that is complete. It also reminds you to mark sponsored posts as such.&lt;/p&gt;

&lt;p&gt;It runs locally, needs no key, and does not claim to know your best posting time, because it has no audience data. It is a planner, not an oracle.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/content-calendar-30-days" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/content-calendar-30-days&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>Is your hook too slow? Check a short video script before you film</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Thu, 12 Nov 2026 10:02:10 +0000</pubDate>
      <link>https://dev.to/robinberaud/is-your-hook-too-slow-check-a-short-video-script-before-you-film-57b4</link>
      <guid>https://dev.to/robinberaud/is-your-hook-too-slow-check-a-short-video-script-before-you-film-57b4</guid>
      <description>&lt;p&gt;You wrote a script for a Reel, a TikTok or a Short. It reads well. Then you film it and the first sentence takes six seconds to say, and the whole thing runs a minute over. Re-shooting is the expensive way to find that out.&lt;/p&gt;

&lt;p&gt;I made a small Hermes skill, short-video-script, that handles the part before the camera. You give your agent a topic, the platform and the one thing you want viewers to do at the end. It writes the script in a fixed format: a HOOK, a few BEATs, a TURN (where the promise is kept) and a single CTA. Every line also carries a visual note in brackets, so you know what changes on screen.&lt;/p&gt;

&lt;p&gt;Then a tiny Python file estimates how long each line takes to say, flags a hook that is too long or a missing call to action, and splits the words into short caption chunks you can paste into your editor.&lt;/p&gt;

&lt;p&gt;Here is a real run on a five-line script:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;00:00.00-00:03.60  HOOK   9 words  3.6s
00:03.60-00:07.60  BEAT  10 words  4.0s
...
TOTAL: 20.4s (target 45s)
hook takes about 3.6s to say, limit is 3.0s (9 words)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The hook is only a little over my 3-second limit, so I shorten it before filming. That is the whole point.&lt;/p&gt;

&lt;p&gt;Some honest notes. The timing is an estimate from word count at a fixed pace, so record one take and adjust the pace setting. The skill never makes up numbers or testimonials: if a figure would help, it leaves a placeholder for you. And if the video is a paid partnership, it reminds you to label it.&lt;/p&gt;

&lt;p&gt;Skill and instructions are in the hermes-skills repository, MIT licensed.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/short-video-script" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/short-video-script&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>Ten hooks, one honest filter: choosing your opening line with reasons</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Wed, 11 Nov 2026 10:02:29 +0000</pubDate>
      <link>https://dev.to/robinberaud/ten-hooks-one-honest-filter-choosing-your-opening-line-with-reasons-5ejj</link>
      <guid>https://dev.to/robinberaud/ten-hooks-one-honest-filter-choosing-your-opening-line-with-reasons-5ejj</guid>
      <description>&lt;p&gt;The video is edited, the post is written, and the first line is still blank. You write a hook, doubt it, rewrite it, then pick whichever feels least bad.&lt;/p&gt;

&lt;p&gt;Hook Lab is a Hermes skill that makes this less of a coin flip. You give it your topic, audience and draft. The agent writes ten hooks from different angles (a mistake, a question, a contrast, a specific number) using only facts you provide. A small local script then scores each one:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Specificity:&lt;/strong&gt; could this line only be about your content?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Curiosity:&lt;/strong&gt; does it open a question people want answered?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Promise kept:&lt;/strong&gt; does your draft deliver what the hook says?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The third one is the point. If a hook says "500 euros in 2 days" and your draft never mentions 500, it gets flagged. So do phrases like "guaranteed" or "you won't believe".&lt;/p&gt;

&lt;p&gt;A tiny run:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;1. [7.3] Why my 12 minute video lost viewers at minute 3
2. [7.0] I rewrote my intro 4 times, but the first draft won.
3. [5.0] I made 500 euros in 2 days
     - number(s) in hook not found in body: 500
4. [1.7] Amazing secret hacks nobody tells you!!
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It is a rule-based filter for English text, not a crystal ball. It can't judge your voice or predict views; your analytics still decide. But it helps you compare options and catch over-promising before your audience does. No API key, no account.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/hook-lab" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/hook-lab&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>One article, six posts: repurposing without losing your voice</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Tue, 10 Nov 2026 21:03:02 +0000</pubDate>
      <link>https://dev.to/robinberaud/one-article-six-posts-repurposing-without-losing-your-voice-5bfh</link>
      <guid>https://dev.to/robinberaud/one-article-six-posts-repurposing-without-losing-your-voice-5bfh</guid>
      <description>&lt;p&gt;You spent a week on an article or a video. Then you spend another week rewriting it as a LinkedIn post, a thread, a carousel and a newsletter blurb. Or you paste it into a chatbot and get something smooth, generic, and sometimes with a number that was never in your piece.&lt;/p&gt;

&lt;p&gt;The content-repurposer skill for Hermes tackles both problems. Before any drafting, a small script reads your article or transcript (text, Markdown, SRT or VTT) and builds a "source brief": how you write (sentence length, "you" versus "I", emoji use), the lines that stand well alone, and every number and link that must survive unchanged.&lt;/p&gt;

&lt;p&gt;The agent then drafts each format with firm rules. Use only facts from the source. Reuse your own lines word for word. Leave a placeholder when a format needs a number you did not give. Keep any sponsorship disclosure. You get a fact-check list showing where each number came from, so review takes minutes.&lt;/p&gt;

&lt;p&gt;A tiny example, from a made-up piece about batch filming:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;python3 scripts/source_brief.py my-article.md
- reads as: short punchy sentences; speaks to the reader directly
- Quotable: "Stop filming when the spark is gone."
- Facts: 3 hours, 6 hours
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The LinkedIn draft opens with your own line, not a template.&lt;/p&gt;

&lt;p&gt;No API key, no account, nothing leaves your machine. The voice analysis is a rough hint tuned for English, so read every draft before posting, and check each platform's current limits.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/content-repurposer" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/content-repurposer&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>contentcreation</category>
    </item>
    <item>
      <title>Which of your dependencies are actually outdated? A one-file check</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Mon, 09 Nov 2026 10:04:38 +0000</pubDate>
      <link>https://dev.to/robinberaud/which-of-your-dependencies-are-actually-outdated-a-one-file-check-2a32</link>
      <guid>https://dev.to/robinberaud/which-of-your-dependencies-are-actually-outdated-a-one-file-check-2a32</guid>
      <description>&lt;p&gt;Most "outdated dependency" tools want an install, an account, or a CI integration. Sometimes you just want to open a project you haven't touched in months and see what has moved.&lt;/p&gt;

&lt;p&gt;I wrote a small Python script for that, packaged as a Hermes skill. It uses only the standard library and the public PyPI and npm registry endpoints, so there is no API key and nothing to install.&lt;/p&gt;

&lt;p&gt;It reads a &lt;code&gt;requirements.txt&lt;/code&gt; or &lt;code&gt;package.json&lt;/code&gt;, takes the pinned version of each package, asks the registry for the latest one, and sorts the result by risk: major updates first, then minor, then patch. Packages it can't judge (loose ranges like &lt;code&gt;&amp;gt;=2.0&lt;/code&gt;, dist tags like &lt;code&gt;latest&lt;/code&gt;) are labelled &lt;code&gt;unpinned&lt;/code&gt; rather than guessed at. A package the registry doesn't know shows &lt;code&gt;lookup-failed (HTTP 404)&lt;/code&gt; instead of crashing the run.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;$ python3 scripts/depwatch.py package.json
PACKAGE      PINNED   LATEST   STATUS
express      4.0.0    5.3.0    major
left-pad     1.0.0    1.3.0    minor
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Add &lt;code&gt;--fail-on-major&lt;/code&gt; and it exits with code 1 when a major bump exists, which is handy in a scheduled job. &lt;code&gt;--json&lt;/code&gt; gives structured output.&lt;/p&gt;

&lt;p&gt;It's deliberately limited: it compares version numbers, not changelogs, so it can't tell you if an upgrade really breaks anything. It skips transitive dependencies and lock files. Treat the output as a to-do list for reading release notes, not as a verdict.&lt;/p&gt;

&lt;p&gt;An offline &lt;code&gt;--selftest&lt;/code&gt; covers the version parsing and classification logic.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/software-development/dependency-update-watch" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/software-development/dependency-update-watch&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>softwaredevelopment</category>
    </item>
    <item>
      <title>Audit a docker-compose file in one command, no PyYAML needed</title>
      <dc:creator>Robin Beraud-Sudreau</dc:creator>
      <pubDate>Sun, 08 Nov 2026 10:02:03 +0000</pubDate>
      <link>https://dev.to/robinberaud/audit-a-docker-compose-file-in-one-command-no-pyyaml-needed-33i3</link>
      <guid>https://dev.to/robinberaud/audit-a-docker-compose-file-in-one-command-no-pyyaml-needed-33i3</guid>
      <description>&lt;p&gt;Most compose files start as a quick experiment and end up in production. Along the way, a few things creep in: a database published on every network interface, a password written straight into &lt;code&gt;environment:&lt;/code&gt;, an image tagged &lt;code&gt;latest&lt;/code&gt;, a container mounted on the Docker socket. Nobody decides these on purpose; they survive because nobody rereads the file.&lt;/p&gt;

&lt;p&gt;I wrote a small Hermes skill, &lt;code&gt;docker-compose-auditor&lt;/code&gt;, that does the line-by-line read. It is one standard-library Python file with a tiny YAML reader built in, so there is nothing to install.&lt;/p&gt;

&lt;p&gt;It reports findings by severity: high for exposed datastores, literal secrets, privileged mode and Docker socket mounts; medium for unpinned images and undeclared volumes; low for missing restart policies and healthchecks. It exits with code 1 when something serious is found, so it works as a CI gate.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/compose_audit.py docker-compose.yml
&lt;span class="c"&gt;# [HIGH  ] db  exposed-datastore  Port 5432:5432 (PostgreSQL) is published on all interfaces...&lt;/span&gt;
&lt;span class="c"&gt;# [MEDIUM] web latest-tag         Image 'nginx:latest' uses the 'latest' tag...&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A reference like &lt;code&gt;${DB_PASSWORD}&lt;/code&gt; is accepted as safe; a literal value is not.&lt;/p&gt;

&lt;p&gt;It has limits, and the skill says so: secret detection is based on variable names, and anchors, merge keys and &lt;code&gt;include&lt;/code&gt; are not resolved. It is a quick first pass, not a replacement for a proper security review.&lt;/p&gt;




&lt;p&gt;Skill (MIT): &lt;a href="https://github.com/RobinBeraud/hermes-skills/tree/main/devops/docker-compose-auditor" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills/tree/main/devops/docker-compose-auditor&lt;/a&gt;&lt;br&gt;
All skills: &lt;a href="https://github.com/RobinBeraud/hermes-skills" rel="noopener noreferrer"&gt;https://github.com/RobinBeraud/hermes-skills&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and reviewed through automated checks.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagent</category>
      <category>ai</category>
      <category>opensource</category>
      <category>devops</category>
    </item>
  </channel>
</rss>
