The three runs
- 399 s
- USD 0.99, bike brand brief
- 477 s
- USD 0.86, bike brand brief
- 406 s
- USD 0.87, bakery brief
Each run was a headless Claude Code session on Sonnet in a fresh folder that held only brief.md and the Briefatto Board skill. The cost is Claude Code's own report at list API price, for the whole session from brief to kit.
What one run makes
- Claude Code reads the brief and draws three directions with its real copy.
- It picks one and runs
board freeze, which writes DESIGN.md as a numbered version. - It makes a 1440x900 website hero plus six formats: a 1080x1080 post, a 1080x1920 story, a 1200x628 link card, a 300x250 banner, a 1240x1754 A4 poster and a 1200x630 OG image.
- It runs
board checkon every piece and fixes what fails. In the 399-second run, all seven pieces passed in 46 turns, with no step that blocked it.
Agents find the board on their own
In our unaided runs, six fresh agents, three Claude Code sessions and three Codex sessions, got only a fictional client's brief and the words "Make our brand and launch kit." All six found Briefatto Board on their own through the AGENTS.md and CLAUDE.md pointer that npx briefatto-board install writes, and opened its skill among their first actions.
One of those kits, Afterlight Matinees, went to three blind Fable judges, and all three said send it as supplied.
The board adds no model cost
The Briefatto Board command line calls no model API and needs no API key. The agent you already run writes DESIGN.md and the pieces; the board captures, renders, records and checks them.
npx briefatto-board installQuestions and answers
How long does Claude Code take to make a brand kit with Briefatto Board?
About 400 seconds for a seven-piece kit: 399 s, 477 s and 406 s in three runs, each from a brief to a frozen DESIGN.md.
How much does a design kit cost in API usage?
Under USD 1 at list price in each of the three runs: USD 0.99, USD 0.86 and USD 0.87.
Does Briefatto Board add its own model costs?
No. It calls no model API and needs no API key; the only cost is your agent's session.
Will my agent find the board without being told?
Yes. All six fresh agents in our unaided runs found it on their own through the AGENTS.md and CLAUDE.md pointer.
Does it work with agents other than Claude Code?
Yes. It works with any coding agent that reads AGENTS.md and runs shell commands, including Codex, Cursor, GitHub Copilot and Gemini CLI.