For creatorsDirector Lite, a free skill
Ask your own assistant what to post next, and get sources instead of guesses.
A post made on a hunch teaches you nothing when it flops. Director Lite reads what your niche published this month, borrows what already works in other fields, and briefs one post down to the cover and every slide, with a source under every claim.
In our blind test its briefs carried 0.3 made-up numbers each. The same model without it carried 2.5.
Working elsewhere
A ceramics studio’s glaze tests
Each glaze experiment is documented as a repeatable test: the result, how the glaze was mixed and applied, the firing program, smooth and textured versions.
Your version
Agent Lab #014. Task, model, context files, permissions, tools, how it was checked, the outcome, and what changed on the next run.
Word for word from the README. The slides are the brief’s own plan, drawn to size.
The detail
What you get, where, and how long
- What you get
- One post to make first, briefed to the cover and the slides. Ideas that work in other niches, rewritten for yours. A numbered list of every source.
- Where it runs
- Claude (Opus or Sonnet), or GPT-5.6 Sol, with web search on. On Gemini, switch on Search grounding first.
- How long
- A standard brief reads about 25 pages. In our tests that took the model five to eleven minutes.
Price
Free
Free, under the PolyForm Internal Use licence. Use it for your own work and your clients'. Don't resell it, rebrand it or republish it.
What it looks like
Invented numbers per brief, blind-graded
- With Director Lite
- 0.3
- Same model, asked to try its hardest
- 2.5
Blind test · same model · same creator · every brief and grade ships with the skill
Free, for your own work and your clients’
Get Director Lite
Leave your email and the link to the skill comes straight to your inbox. It is free under the PolyForm Internal Use licence.
Questions
What people ask first
- What does Director Lite cost?
- Nothing. Clone it from GitHub and add it to Claude, Codex, or any assistant that can browse.
- Which assistants does it work best on?
- Claude (Opus or Sonnet) and GPT-5.6 Sol with web search on. It needs to browse, because it reads what is live this month.
- Does it make up numbers?
- Rarely. In a blind test, with the same model and the same creator on both sides, its briefs carried 0.3 invented numbers each against 2.5 from the model on its own. Every brief and every grade is published on GitHub. When it cannot find something, it writes "unknown".
Plainly
Where it works best
It works best on Claude (Opus or Sonnet) or GPT-5.6 Sol with web search on, for someone who will make the post it briefs. You get what to make, how it should look and where each idea came from. What a post earns once it is out is up to your audience, so the promise stops at the brief.
It reads the web as text, so it cannot see covers. Meliura Director looks at the covers themselves. Meliura Director