A Staff Engineer's Playbook, Repackaged for LLMs

What it does

agent-skills is a library of 23 Markdown-based skill files that guide AI coding agents through structured engineering workflows. Each skill acts as a system prompt with steps, verification gates, and anti-rationalization tables covering phases from spec and planning to testing, review, and shipping. The pack also includes specialist agent personas and quick-reference checklists for security, performance, and accessibility.

The interesting bit

The project treats prompt engineering as organizational discipline. Rather than one-shot instructions, it bakes in senior-engineer habits like spec-before-code, thin vertical slices, and five-axis code review. The doubt-driven-development skill is particularly unusual: it forces an adversarial self-review with a CLAIM → EXTRACT → DOUBT → RECONCILE → STOP loop, optionally escalating to another model when stakes are high. It is essentially teaching your agent to second-guess itself.

Key highlights

• 23 lifecycle skills that map to the full development loop, activated by slash commands or context (e.g., UI work auto-triggers frontend-ui-engineering)
• Structured workflows with acceptance criteria, change-sizing norms (~100 lines), and explicit “stop-the-line” rules
• Adversarial quality gates including doubt-driven-development and code-review-and-quality with Staff Engineer standards
• Pre-built personas for security auditing, test engineering, and staff-level code review
• Plain Markdown format compatible with Claude Code, Cursor, Copilot, Gemini, Windsurf, and others

Verdict

Worth exploring if you use AI agents for production work and want them to follow consistent engineering hygiene. Skip it if you are looking for executable tooling; this is essentially a curated prompt and rules library, not a runtime framework.