Matt Pocock Skills V1.2 Boosts AI Coding Efficiency

Version 1.2.0 introduces several improvements to the engineering skills repository, building on its already popular foundation.

The project has achieved significant recognition, reaching 204K stars on GitHub and ranking as the 24th most-starred repository of all time. A new dedicated documentation site is now live at aihero.dev/skills, featuring a flow guide and an AI coding dictionary to help users master skills and understand deeper coding concepts.

By integrating into the Claude Code plugin marketplace, Matt Pocock's skills now feature simplified installation and automatic updates delivered through a read-only bundle.

OpenAI YAML files are now included for every skill, ensuring compatibility with the Codex UI, and sidecar files enforce 'allow implicit invocation false' to prevent hidden skills from cluttering the agent's context window in certain harnesses.

Matt Pocock noted that Opus 5 often produces overly verbose and unclear output, leading to frustration for users.

To counter this, the new 'wait what' skill was developed to compel the AI to use simple, declarative language, instructing the agent to ground its responses in the ubiquitous language defined in context.md.

This new skill is designed for instances where the AI agent generates unusable or confusing output, providing a quick solution for users.

Users can simply input 'Wait, what?' to prompt the agent to provide a better, more accurate alternative response.

The original 'Grill Me' skill was inefficient, asking only one question per turn, which resulted in frustrating and repetitive interactions during the final stages of a session.

The updated system now organizes questions into a graph structure where a single critical query can unlock subsequent groups of questions, improving efficiency.

This design allows Grill Me to follow the graph, asking only currently available questions in rounds, which streamlines the interaction process.

The updated interface uses emojis to improve visual navigation for users, making it easier to follow question flows.

Users can now answer multiple questions in a single turn, significantly speeding up interactions as the system continuously pushes the 'frontier' of questions forward upon completion of previous rounds.

This refinement helps maintain concise documentation, makes model behavior more predictable, and crucially, aids in breaking down configuration files into modular skills to prevent front-loading issues.

This skill prioritizes control and security by avoiding direct agent-based cloud access via computer use, instead handling sensitive tasks like securely saving API keys or pushing to GitHub secrets.

The 'Wizard' skill converts a previously 'incredibly painful' process into what is described as a 'weirdly joyful' experience due to its interactive and guided nature.

This feature enables collaboration with non-technical stakeholders who may not be 'AI native' and acts as a bridge for teams without integrated agent platforms like Slack or Microsoft Teams.

The skill has been personally used by Matt Pocock to design a garden office by syncing decisions between his wife and the AI, serving as a temporary solution until agent collaboration becomes more intuitive.

Matt Pocock noted that the 'Wait What' skill will be essential for managing unexpected Opus model behaviors, and he encouraged users to check the GitHub releases for v1.2.0 to view the full change log.