A collection of agent skills for coding CLIs — Claude Code, Codex, Gemini CLI, Copilot, and others.
Skills are markdown-only instructions that shape how your coding agent writes, reviews, and architects code. No runtime dependencies, no build steps — just drop them in and go.
Skills | Installation | Usage
| Skill | Description |
|---|---|
| agentmd | Generate a single canonical AGENTS.md plus minimal CLI-specific shims that @-import it for tools that don't read AGENTS.md natively (Claude Code, Gemini CLI, Qwen Code). Based on ETH Zurich research showing that auto-generated context files decrease performance while minimal human-written ones improve it. |
| low-complexity | Automatically enforce low Cognitive Complexity (SonarSource) and Cyclomatic Complexity in all generated code. Activates on every code write/modify — no explicit trigger needed. |
| vertical-slice-architecture | Organize code by feature/use-case instead of technical layers. Works with any language and app type (web API, CLI, event-driven, etc.). |
| codex-review | Iterative code review debate between Claude Code and Codex CLI until both reach consensus. Supports plan-mode review with implementation contracts. |
| codex-discuss | Iterative non-code debate between Claude Code and Codex CLI on any topic — diet, fitness, writing, decisions, strategy, brainstorming. Same iterative consensus mechanic as codex-review, adapted for open-ended subjects (content inline, no filesystem navigation). |
| test-namer | Write expressive, behavior-focused tests following Vladimir Khorikov's testing principles. Plain English names, test behavior not implementation. |
| dual-testing | Go dual testing strategy: integration tests (testcontainers) verify full-chain wiring for happy paths, unit tests (testify/mock) verify error handling logic. Avoids duplicating happy paths across layers. |
| explain-pr | Explain a freshly opened Pull Request or Merge Request back to you in plain language — what changed, why, and how — with the relevant code snippets inline. Fixed, predictable structure; reads the local git diff, so no gh/glab required. |
| pr-quiz | Formative quiz right after a PR/MR ships (chains automatically from explain-pr): three mandatory question roles (behavior / invariant / diagnosis), one at a time, graded on your initial answers with file:line evidence. Opens gaps in a per-project comprehension map but never certifies — that takes a cold or delayed study session. |
| study-session | Measured comprehension sessions over real code, two modes: deep (45-60 min — hidden rubric, cold reconstruction, guided investigation, novel transfer task with an executable prediction) and checkpoint (10-15 min spaced retention check). Grades four dimensions on pre-correction answers; tracks outcome + evidence-freshness per subsystem in a per-project comprehension map. |
| como-si-fuera-de-boca | Explain anything — code, errors, concepts, or non-technical topics — in the simplest plain-language way (ELI5), with a natural Río de la Plata (Argentine) voice that puts clarity before humor. On-request only: triggers on "explicámelo como si fuera de Boca", "bajámelo un cambio", "en criollo", "ELI5", "explain like I'm 5", etc. |
| ste | Write, rewrite, explain, or check anything in ASD-STE100 Simplified Technical English — documents, MR descriptions, or something just discussed in conversation. Four compliance levels (Verified / Marked / Clean / Raw); Verified checks every word against the official Issue 9 dictionary (auto-downloaded, cached, structurally validated) via a deterministic gawk checker with TN/TV annotations — never from memory. Drafts in controlled English first, then translates preserving the STE structure. |
| claude-review | Reverse counterpart of codex-review, for sessions where Codex is the local agent: iterative code review debate between Codex CLI and Claude Code (headless claude -p in read-only plan mode) until both reach consensus. Supports plan review with implementation contracts. |
| claude-discuss | Reverse counterpart of codex-discuss: iterative non-code debate between Codex CLI and Claude Code on any open-ended topic — diet, writing, decisions, strategy. Same consensus mechanic, with the topic content inlined. |
| agent-context-auditor | Audits and produces a repository's agent context files (AGENTS.md, CLAUDE.md, copilot-instructions and other pointers) from local evidence, user intent questions, and per-rule omission cost. Also judges a candidate rule before it gets written. Skill text in Spanish (Río de la Plata). |
| c4-model | C4 model architecture documentation as diagrams-as-code, faithful to the official c4model.com notation (rules distilled from the complete site). Selects the notation for clarity: C4-PlantUML where the target renders it (verified against GitLab.com's PlantUML server — Graphviz layout, native legend), Mermaid flowchart/sequence conventions as zero-dependency fallback; never Mermaid's experimental C4 syntax. Evidence-based codebase analysis, canonical model reused across views, per-diagram key/acronyms/scope/evidence, and the official review checklist as a self-review step. |
npx skills add mryll/skillsnpx skills add mryll/skills --skill <skill-name>For example:
npx skills add mryll/skills --skill low-complexity
npx skills add mryll/skills --skill codex-reviewTip
You can install multiple individual skills by running the command once per skill. Only install what you need — each skill is fully self-contained.
Once installed, skills activate automatically based on context. Some examples:
- agentmd — Ask your agent to "generate CLAUDE.md" or "create AGENTS.md"
- low-complexity — Activates automatically whenever code is written or modified
- vertical-slice-architecture — Ask to "use vertical slice architecture" or start building features in a VSA project
- codex-review — Ask to "review with codex" or "validate plan with codex"
- codex-discuss — Ask to "discuss with codex", "iterate this with codex", or "get codex's take on this" (non-code topics: diet, fitness, writing, decisions, etc.)
- test-namer — Activates whenever tests are written, created, or reviewed
- dual-testing — Ask "where should this test go?" or activates when designing test strategy for a Go handler/feature
- explain-pr — Ask to "explicame el PR/MR", "walk me through this PR", or run
/explain-prright after opening a PR/MR. Optional: to fire it automatically after every PR/MR, add a line like "when you create a PR/MR, invoke explain-pr" to yourAGENTS.md/CLAUDE.md. - pr-quiz — Chains automatically after explain-pr, or ask "quiz me on this change" / "tomame el quiz" after shipping a PR/MR.
- study-session — Ask for a "study session" / "sesión de estudio" on a repo or subsystem you want to truly understand.
- como-si-fuera-de-boca — Ask to "explicámelo como si fuera de Boca", "bajámelo un cambio", "explicalo simple", "en criollo", "ELI5", or "explain like I'm 5" when you want something dumbed down on request.
- ste — Ask to "reescribí esto en STE", "explicame esto en STE", "rewrite this in Simplified Technical English", "redactá este warning en STE", or "check this against ASD-STE100" — works mid-conversation on anything just discussed. Choose Verified (dictionary-checked), Marked, Clean, or Raw; translation on request.