Skills

Ten skills for AI coding agents, one install: isolated code review, independent verification before "done", model routing with resumable Codex lanes, a bounded audit pass, codebase readiness scoring, spaced-repetition teach-back, audience-calibrated explanations, an expert stress test, a handoff prompt for a second agent’s independent review and mobile screen concepts.

AGENTS / TOOLINGDeveloper toolsOpen source · public

Markdown
TypeScript
Bash
SQLite
Apache-2.0

$ gh repo view mneves75/skills

Back to work

MN://skills — System route

$ trace "skills"

repo
readiness scan
routing policy
teach-back review

The problem

When an AI coding agent underperforms in a repository, the repository is usually the reason, and there is no shared way to say so.

The constraint

The skills have to work across agents rather than bind to one vendor, and the assessment has to produce a verdict a human can argue with.

The approach

Each skill is a SKILL.md procedure the agent loads when the request matches its description; three ship a small script (an isolated review runner, a resumable Codex lane wrapper, an SM-2 SQLite store). A Bun/TypeScript assessor scores a repository on nine pillars and a maturity level, locally. CI validates every skill against the Agent Skills spec, runs the hook and typecheck gates, and fails on stale generated site files.

The outcome

Apache-2.0 (three adapted skills keep MIT); `npx skills add mneves75/skills` links the ten skills into Claude Code, Codex, Cursor, OpenCode, pi and any tool that reads SKILL.md. The site carries a how-to with recorded sessions and a sample readiness report.

MN://WORK

More work

See all work