skills

Skills for Real Engineers is a repository by Matt Pocock containing agent skills used for everyday software engineering with AI coding agents. It is shared from the author's .agents directory and designed to be small, adaptable, composable, and model-agnostic. The author distinguishes these from larger process-owning approaches like GSD, BMAD, and Spec-Kit, arguing that those take control away from the developer; these skills instead aim to fit into existing workflows. Installation offers two paths: a Claude Code plugin installs the full set as a managed, read-only bundle that updates automatically, while the skills.sh installer copies editable skill files into a project so users can modify them. After installation, a setup skill (setup-matt-pocock-skills) is run once per repository to configure the issue tracker (GitHub, Linear, or local files), triage labels, and documentation location. The repo's stated motivation is fixing common failure modes in coding agents: - Misalignment between what the user wants and what the agent builds, addressed by grilling sessions that interrogate the plan before work begins (grill-me, grill-with-docs). - Verbose agent output, addressed by building a shared domain language documented in CONTEXT.md and ADRs, so agents reason and communicate more concisely. - Code that doesn't work, addressed through feedback loops: static types, browser access, and automated tests, including a red-green-refactor TDD skill and a phased diagnosing-bugs loop. - Architecture decay, addressed through design-focused skills that encourage deep modules and regular codebase architecture surveys. Skills are organized into two main categories: Engineering skills include user-invoked commands such as ask-matt (a router over skills), grill-with-docs, triage, improve-codebase-architecture, to-spec, to-tickets, implement, and wayfinder. Model-invoked engineering skills include prototype, diagnosing-bugs, research, tdd, domain-modeling, codebase-design, code-review, resolving-merge-conflicts, and wizard. Model-invoked skills may be called by the agent automatically when appropriate; user-invoked skills orchestrate workflows and may invoke model-invoked ones. Productivity skills, usable beyond coding, include user-invoked commands like grill-me, handoff, teach, to-questionnaire, and wait-what, plus the reusable interviewing primitive grilling and writing-for-agents for producing agent-readable documentation such as AGENTS.md and CLAUDE.md. The author offers a newsletter for staying updated on changes to the skills. The repository also documents design decisions, including one about shipping as a Claude Code plugin, in an ADR directory. Overall, it is a practical library of repeatable engineering practices encoded as slash commands and agent-callable skills, aimed at developers who want more control over how AI agents write software.