What v0.3 didn't solve
In my May case study, I described a system with 8 role folders, 8 generic skills, and a Qdrant vector memory you'd switch on beyond about 50 pieces of content published per month. It worked. But as I duplicated it for new client accounts, 3 points of friction kept coming back.
The first: the brand doctrine was a file you could consult, and nothing forced you to go through it. A poorly used skill could produce correct content without ever reopening 01-brand. The second: the editorial calendar lived in a plain Markdown file, with no approval status. There was no way to tell at a glance what was an idea, a draft, or published. The third, and the biggest: Qdrant required a vector database to run, monitor and debug, for a level of use almost none of my accounts ever reached.
None of these points of friction was a blocker on its own. Together, they slowed down every new deployment. Version 2 tackles all 3 head-on.
Copilot becomes Cockpit
The first change, and the most visible: the name. The /start-copilot command becomes /start-cockpit, and the copilot-setup skill becomes cockpit-setup. The GitHub repository stays put: it remains at github.com/Littlpinguin/marketing-copilot-template.
The new name carries meaning. “Copilot” suggested an autopilot, a system that would make decisions for you. “Cockpit” better describes what it really is: a flight deck. The instruments are there and the brand doctrine is wired in, while you stay at the controls. It's also the name I chose for the paid engagement, the AI Marketing Cockpit. Both share the same philosophy: the free MIT-licensed template is the version you fly yourself, and the engagement is the version I fly with you.
What was added
The v2.0.0 overhaul goes all the way to the core of the system.
| Dimension | v0.3 | v2.1 (current) |
|---|---|---|
| Skills | 8 generic | 45 (49 being integrated) |
| Specialized agents | Not applicable | 10 |
| Brand doctrine | A file to consult | Injected as step 0 of every skill |
| Editorial calendar | Plain Markdown | Approval statuses (idea → draft → to review → approved → published) |
| Optional modules | None | 7, activated with /modules |
The first 43 skills of v2.0 come partly from community sources under MIT or Apache licenses (claude-seo, ui-ux-pro-max, taste-skill, frontend-design, claude-ads, claude-blog), adapted to the per-brand folder system, and partly from skills I had already proven in production (slide decks, carousels, image generation). Version 2.1 adds 2 more (45 in total) with the full port of the n8n module: n8n-builder, n8n-audit, n8n-debugger. A 4th wave is in progress and not yet officially released: 4 web animation skills (animation-gsap, animation-animejs, animation-lottie, animation-scroll-reveal) taken from the third-party repo claudedesignskills, under the MIT license. Once they're merged, the count will reach 49.
The 7 optional modules cover what goes beyond pure content production: video, n8n automations, a performance dashboard, acquisition, market monitoring, scheduled social publishing via Postiz, and a password-protected client space. Each one is activated on demand with /modules, without weighing down the base install for those who don't need it.
The 06-graphic-design/presentations/ module also received a catalog of 52 HTML slide layouts, with a full-screen mode and a mandatory Playwright quality check before PDF export. The repo's public demo now uses a complete fictional brand, “Meridian Conseil”: 52 layouts, 10 landing pages, 10 interactive lead magnets, and a dashboard covering 4 months. Enough to get a feel for the output before you've even configured your own brand.
What was removed
This is the part that rarely gets documented in a product changelog, and yet it's often the most revealing. Version 2.0 removes Qdrant outright: no more vector database, no more semantic memory pipeline. The /seed-corpus and /connect-qdrant commands go with it. Memory now relies solely on the brand files, read directly.
Qdrant worked. Removing it, after presenting it as the way forward, simply reflects how it was used: almost nobody needed it, and every extra connection is an extra point of failure for a gain most accounts never reached.
In my May case study, I put Qdrant's useful threshold at around 50 pieces of content published per month. In practice, none of the accounts I manage has stayed above that threshold for several consecutive months. The maintenance cost (monitoring indexing, debugging vector queries, explaining the concept to a client who couldn't care less) far exceeded the real benefit. Simplifying here aligns the tool with how it's actually used, instead of how I imagined it would be.
See the full changelog and the code
54 skills, 14 agents, 7 optional modules: it's all in the repo, under the MIT license.
Open the GitHub repoWhat it changes if you duplicate the template today
If you're starting from scratch, nothing changes in how you get going: you clone the repo and run /start-cockpit in place of the old /start-copilot. The wizard still spends most of its time on brand discovery, now with an explicit access check on sensitive connectors before they're activated.
If you were already on a pre-v2 version, be aware that there's no automatic migration tool yet. The CHANGELOG.md documents every change version by version, but merging an old 01-brand folder by hand with the new folder tree is riskier than it looks. The safest route is still to clone the latest version of the repo and copy your existing brand guidelines into it, rather than merging.
The repo itself hasn't changed address or license: still free, still MIT, still at github.com/Littlpinguin/marketing-copilot-template. For the complete system architecture (modules, skills, connectors), the Cockpit project page remains the up-to-date reference.
Conclusion
A v2 is also about what you remove when actual use disproves the starting assumption. Qdrant was a good idea on paper. Real-world use said otherwise, and the system adapted to reality rather than the other way around.
The name changes and the architecture gets stronger, while the original intent stays intact: you fly the system, and the decisions stay yours.
