Open‑Source Skill Pack that Helps AI Engineers Tame Large‑Model Code Assistants
The article introduces the open‑source project mattpocock/skills, which equips developers with interactive “grill” commands to interrogate AI code assistants, align expectations, use a shared ubiquitous language, and integrate the skills in under 30 seconds, aiming to close the communication gap between engineers and large‑model generators.
Problem – AI Miscommunication
When developers ask an AI code assistant to implement a feature (e.g., a payment module), the model often generates code that follows the wrong assumptions because the required domain details are not explicitly captured.
Solution – "Grill" Commands
The core skills are /grill-me and /grill-with-docs. Invoking /grill-me makes the AI respond with a series of clarifying questions (e.g., payment states, refund handling, multi‑currency support) instead of emitting code immediately. /grill-with-docs extends this pattern by allowing the AI to reference a supplied documentation set while asking the same questions.
Ubiquitous Language via CONTEXT.md
A CONTEXT.md file defines project‑specific terminology (e.g., “materialization cascade”). The AI reads this file before any interaction, so subsequent dialogs use the defined terms consistently, reducing the need to repeat explanations.
Design Philosophy – Small, Precise, Composable
Each skill is an independent shell script or Markdown directive. Because they are isolated, they can be combined like building blocks on any supported coding agent (Claude Code, Codex, Cursor, etc.). This avoids “vibe coding”, where an AI takes over the entire development pipeline and obscures the source of errors.
Installation – 30 seconds, Zero Barrier
npx skills@latest add mattpocock/skillsRunning the command displays a list of available skills. The user selects the desired skills and the target coding agent; the setup completes in under 30 seconds.
Intended Audience
Developers who experience inaccurate AI‑generated code.
Technical leads who need to coordinate requirements with AI assistance.
Full‑stack engineers seeking higher fidelity from AI code generation.
Users of Claude Code, Codex, Cursor, or similar AI tools.
Project Origin
The skills originate from the author’s personal .claude directory and reflect decades of engineering practice. They are distributed as open‑source scripts rather than theoretical frameworks.
Repository
GitHub: https://github.com/mattpocock/skills
Stars: 34,722 | Today’s increase: +7,429
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.