About
A handoff should carry the task, not change it.
PromptingCraft.com started from a practical problem: the prompt handed to a coding agent often says less — or more — than the discussion that produced it.
The project
PromptingCraft.com is an open-source Agent Skill for Claude Code and Codex, MIT licensed and developed in public on GitLab. It turns an established development discussion into an execution-ready prompt: it extracts the task contract, consults applicable official guidance, drafts, critiques once and checks that requirements and permissions survived.
It is deliberately small: one skill, plain references, deterministic checks. No hosted service, account, telemetry or model router.
Principles
- Your words first. Explicit instructions outrank any guideline.
- Official guidance only. Model-specific advice must come from the model developer’s documentation.
- Labels, not blur. Official, methodology, user requirement and research stay separate.
- Honest limits. No invented sources, dates or results, and no superiority claims.
Who develops it
PromptingCraft.com is developed by the dondego.com team. It is an independent project and is not affiliated with, sponsored by or endorsed by OpenAI or Anthropic. Claude, Claude Code, OpenAI and Codex are trademarks of their respective owners and are named only to describe compatibility.
How this website was researched
For transparency: the tools below informed this website’s design and page structure. None of them is evidence for how to prompt a model — that is reserved for official documentation.
Design references · Mobbin
Sections reviewed on 2026-10-12. Principles were adapted; layouts, artwork and brand identities were not copied.
- Height — “Turn //todo into tasks” heroA transformation stated in the headline itself.
- The Content Architecture — real repository in an editor frameShow the actual artifact, not a mock.
- Antimetal — centred headline over a product windowRestraint: one statement, one demonstration.
- Symbolic — editorial serif with stepped process cardsSerif display type for an editorial voice.
- Clay — bracketed, numbered “How it works” stepsMonospaced step numbers as structure.
- Ploy — row-by-row “today vs. with” comparisonContrast specific behaviours, not slogans.
- Wild — scattered points resolving into a lineVisualise disorder becoming order.
- Stripe — dated, tagged changelog rowsLedger rows with date and tag for provenance.
- Retool — column footer with an oversized wordmarkA confident close to every page.
Search intent · Ubersuggest
United States, English, retrieved 2026-10-12. Estimated monthly searches; used to name pages after what developers actually look for.
| Query | Volume | Where it is answered |
|---|---|---|
| claude code skills | 12,100 | Docs and homepage describe what the skill is and how to install it. |
| agent skills | 8,100 | Homepage positions PromptingCraft.com as an Agent Skill. |
| claude code best practices | 2,400 | Sources page lists official recommendations with links. |
| claude code prompts | 720 | Examples page shows real before/after prompts. |
| openai prompt engineering guide | 480 | Sources page links to OpenAI’s own guides. |
| anthropic prompt engineering guide | 320 | Sources page links to Anthropic’s own guides. |
| codex best practices | 210 | Sources and How it works. |
| codex prompts | 140 | Examples page. |
| how to use claude code skills | 140 | Docs page. |
“ai coding agent prompts” and “agent task specification” showed no measurable volume, so no page targets them.
Illustrations were generated with Higgsfield (GPT Image model) from original prompts and edited for this site; the icon is an original SVG.