{"slug": "we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-and", "title": "We Built an Open-Source Skill to Improve Coding-Agent Prompts — Now Live in ChatGPT and the MCP Registry", "summary": "The dondego.com team released PromptingCraft, an open-source Agent Skill that turns a development conversation into an execution-ready prompt for coding agents, capturing task requirements, decisions and authorization boundaries. The skill is available in public GitHub and GitLab repositories, published in the ChatGPT Plugin Directory at version 0.3.0, and its optional guidance server is listed in the official MCP Registry, with a separately versioned standalone Skill at 0.4.0. The team notes the skill helps prepare instructions but does not guarantee correct implementation, and that the 0.4.0 ChatGPT package is a candidate rather than a published upgrade.", "body_md": "A coding agent can follow a request faithfully and still miss the decision that mattered: an API must remain compatible, deployment is out of scope, or a human must approve a release. Those details often live across a development conversation rather than in a single task description.\n\nWe built [PromptingCraft.com](https://promptingcraft.com) to help turn that conversation into a prompt a fresh coding agent can execute. The open-source Agent Skill is now available in our public repositories; PromptingCraft.com is also [published in the ChatGPT Plugin Directory](https://chatgpt.com/plugins/plugin_asdk_app_6acde71d1ddc81918d6095dba60b6023), and its optional guidance server is listed in the [official MCP Registry](https://registry.modelcontextprotocol.io/?q=com.promptingcraft%2Fguidance).\n\nA request like this can be a useful start:\n\nAdd pagination to the users API.\n\nBut imagine the surrounding discussion said existing clients must keep working, the agent can run tests, and deployment requires a separate decision. An execution-ready handoff could instead say:\n\nAdd pagination to the users API while preserving existing client compatibility. Inspect the current API and tests first, implement the smallest compatible change, and run relevant tests. Report the behavior and any migration implications. Do not deploy; ask for separate authorization before any production release.\n\nThat is an illustrative example, not a benchmark result. The goal is to preserve the actual discussion, not add requirements the user never gave.\n\nPromptingCraft extracts the task, requirements, decisions and authorization boundaries. It considers applicable provider guidance, drafts a concise prompt, critiques whether the important contract survived, and revises when there is a substantive reason. If the original prompt is already suitable, it can say so rather than rewrite it unnecessarily.\n\nTry asking your skill-capable coding agent:\n\nThe Skill helps prepare instructions; it does not guarantee that a coding agent will implement a task correctly. Check the resulting prompt and the agent's work.\n\nThe standalone Skill contains the prompt-authoring workflow. The optional remote MCP server provides structured access to curated, dated prompting guidance and source provenance. Its tools search guidance, retrieve a record, and check freshness of supported allowlisted sources. It is not a general web search, and the server does not fetch every source on every request. The currently published MCP 0.3.0 service and the newer standalone Skill 0.4.0 are separately versioned.\n\nProvider recommendations and PromptingCraft's own methods are distinguished in the project. Source guidance is evidence to consider, not an instruction from a third-party page to override the user.\n\nFor Claude Code, the public README documents a verified installation using the skills CLI:\n\n```\nDISABLE_TELEMETRY=1 npx -y skills@1.7.4 add revanbcn/promptingcraft-skill --skill promptingcraft --agent claude-code --copy --yes\n```\n\nSee the [README on GitHub](https://github.com/revanbcn/promptingcraft-skill) or the [canonical GitLab repository](https://gitlab.com/revanalex/promptingcraft-skill) for other installation and connection methods. The public MCP endpoint is `https://mcp.promptingcraft.com/mcp`; connection and tool availability depend on the client.\n\nThe ChatGPT listing is version 0.3.0. A 0.4.0 ChatGPT package is a candidate, not a published upgrade. An existing Cursor Directory entry is separate from the Cursor Marketplace; we are not claiming marketplace approval or automatic activation across clients.\n\nOur recent work includes clearer installation instructions, scoped DeepSeek and Kimi source records in the public source, and client compatibility checks. The production MCP currently serves its separately deployed guidance ledger; a source update should not be mistaken for a production deployment.\n\nLearn more at [PromptingCraft.com](https://promptingcraft.com). Developed by the dondego.com team.", "url": "https://wpnews.pro/news/we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-and", "canonical_source": "https://dev.to/revan_dondego/we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-chatgpt-and-the-mcp-27hg", "published_at": "2026-10-28 21:59:52+00:00", "updated_at": "2026-10-28 22:11:37.585230+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "agent-protocols", "ai-products"], "entities": ["PromptingCraft", "dondego.com", "ChatGPT", "MCP Registry", "Claude Code", "GitHub", "GitLab", "Cursor"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-and", "markdown": "https://wpnews.pro/news/we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-and.md", "text": "https://wpnews.pro/news/we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-and.txt", "jsonld": "https://wpnews.pro/news/we-built-an-open-source-skill-to-improve-coding-agent-prompts-now-live-in-and.jsonld"}}