8000 GitHub - alltomatos/skills: Skills for Real Engineers. Straight from my .claude directory. · GitHub
[go: up one dir, main page]

Skip to content
alltomatosPublic
forked from mattpocock/skills

About

Skills for Real Engineers. Straight from my .claude directory.

Resources

Stars

22 stars

Watchers

1 watching

Forks

 
 

Repository files navigation

Skills

Skills for Real Engineers (Fork)

English | Português (Brasil)

skills.sh

This repository is a fork of mattpocock/skills. All credit for the original design and architecture goes to Matt Pocock and the contributors of the original project.

This fork adds GitHub-driven governance, persistent documentation, planning by Issues, automatic framework verification, skill re-deploy and a mandatory QA gate. The idea is to make agent-assisted development fast without losing traceability, feedback and human control.

The skills are small, composable and adaptable. They work across models and agent CLIs. Hack them, adapt them, make them yours.

Use it (no install)

The repository is the source. Paste the link and ask for the skills you want: there is no installer, no permanent clone and no environment variable.

https://github.com/alltomatos/skills
Where you are What to do
Claude Code /plugin marketplace add alltomatos/skills then /plugin install alltomatos-skills@alltomatos-skills
Codex, OpenCode, Cursor, Gemini, Copilot and other CLIs npx skills add alltomatos/skills
Any agent, no CLI Ask: "use the skills from https://github.com/alltomatos/skills - read SKILLS.md and load only the ones I ask for"

SKILLS.md is the full catalog (name, when to use it, path, raw URL and support files). llms.txt is the compact index agents read first. Both are generated from .claude-plugin/plugin.json, which is the single source of truth for which skills ship.

Language

Skill instructions are written in English, and there is a single copy of each skill. The language of the output is a separate matter and follows whoever is using the skill: reply in Portuguese and everything that comes back — PRDs, ADRs, QA reports, mockups — is in Portuguese. Set output_language in your project's AGENTS.md to pin it. The full contract is in AGENTS.md.

Adding or changing a skill

  1. Create the folder under skills/<bucket>/<name>/ with the SKILL.md and its support files beside it.
  2. Add the path to skills in .claude-plugin/plugin.json.
  3. Run node scripts/build-catalog.mjs 8000 to regenerate SKILLS.md and llms.txt.

Skills under skills/personal/ are not distributed and do not appear in the manifest.

First run

With the skills loaded, open a project in your agent and run the architect to drive scope alignment, design system and technical architecture:

architect

(The architect skill covers discovery, active questioning, mockup/ADD generation and preparation of the architectural artifacts before handing off to execution.)

The consuming project must be versioned in Git and have a remote repository on GitHub. In a new project:

git init
git add .
git commit -m "chore: initialize repository"
gh repo create <name> --source . --remote origin --push

If the environment is empty, without Git or without a GitHub remote, the architect/developer must stop and guide that process before implementing anything.

Workflow: closed-loop engineering

  1. Framework update: checks the repository (local clone, when present, or GitHub) and reports new commits. No installer: update by re-adding the marketplace or running npx skills add again.
  2. Governance and Scope: validates Git, GitHub remote, authentication and project state.
  3. Architecture and Design (architect): runs active questioning, designs the solution (ADD/SAD, C4, ADRs, DER schemas), defines design tokens and screen mockups (MASTER.md + overrides), User Guides, and obtains the mandatory HITL approval from the user.
  4. Slicing into Issues (to-issues): cuts the Epics and approved specifications into vertical slices on GitHub with verifiable acceptance criteria and Blocked by dependencies.
  5. Orchestration (developer): coordinates the task queue and the per-Issue development loop, managing branches/worktrees.
  6. Test-Driven Development (tdd): applies a strict Red-Green-Refactor cycle to each issue (with support from query-docs, expo-expert, diagnose).
  7. Functional Quality Assurance (qa-analyst): mandatory pre-security gate; confronts requirements, edge cases, functional tests and acceptance criteria.
  8. E2E Audit and Security (secure-e2e): mandatory pre-PR gate; code audit (OWASP) and offensive/negative tests against browser and API (*.spec.sec.ts).
  9. Delivery (PR): only after QA and Security are green, runs the Pull Request flow with human confirmation.

GitHub is the persistent source of Issues, scope, acceptance criteria, dependencies, review and history. ESTADO_DEVELOPER.md is only the operational view of the DAG.

The Developer

developer is the master skill of this fork (renamed from orchestrator), a concept that does not exist in the original repository. It evaluates, documents, delegates, supervises and expands the engineering flow.

orchestrator keeps working as a compatibility shortcut — it calls exactly the same skill, so installations made before the rename do not break.

Golden rule

The developer does not do heavy work directly. It identifies the problem and delegates to the right skill. If no skill covers the bottleneck, it invokes write-a-skill to create one.

Risk tiers

Tier Use Behavior
T1 typo, lint, simple documentation atomic execution and local verification
T2 tests, small ADRs, limited changes validates the roadmap and reports evidence
T3 schema, authentication, data or public API pauses and requires human approval

Why these skills exist

1. The agent did not do what I wanted

The most common failure mode is misalignment. grill-me, grill-with-docs and grill-feature-with-docs run a question session that turns intent into requirements, domain language and recorded decisions.

2. The agent is too verbose

A shared language reduces ambiguity, improves file and function names and makes the codebase easier to navigate. grill-with-docs records that language in CONTEXT.md and ADRs. See also the CONTEXT.md example from the original course-video-manager repository.

3. The code does not work

The agent needs feedback loops: types, browser, automated tests and minimal reproductions. tdd applies red-green-refactor and diagnose runs the cycle reproduce, minimise, hypothesise, instrument, fix and regression-test.

4. We built a ball of mud

Agents speed up coding and can speed up entropy too. improve-codebase-architecture promotes continuous design, clear boundaries and a more testable, navigable base.

5. The risk of autonomous development

Roadmap, GitHub Issues, documentation, worktrees and QA form a governance loop that limits scope drift and makes decisions auditable.

Delegation table

Problem Skill
Governance and orchestration developer
System architecture, ADD/SAD, C4, Mockups/UI and ADRs architect
Roadmap and Epics roadmap
Triage and issue management triage
Domain language and ADRs grill-with-docs
A specific module or feature grill-feature-with-docs
Slicing into GitHub Issues to-issues
PRD to-prd
Hard bug or regression diagnose
Code without tests tdd
Third-party APIs query-docs
Expo/React Native apps (EAS, Router, config plugins, upgrade) expo-expert
E2E and security secure-e2e
Quality analysis qa-analyst
Degraded architecture improve-codebase-architecture
Throwaway prototype prototype
MVP bootstrap scaffold-mvp
Missing context zoom-out
MCP servers mcp-builder
SuperKuma deploy and monitoring superkuma-monitoring
Pre-commit hooks with Husky/Prettier setup-pre-commit
Git safety guards in the agent git-guardrails-claude-code
Creating, evaluating and optimizing skills skill-creator
Ultra-compressed communication (token savings) caveman
Unmapped bottleneck write-a-skill
Plan alignment grill-me
Handoff handoff

Skills reference

Engineering

Skills for daily code work: architect, diagnose, grill-with-docs, grill-feature-with-docs, triage, improve-codebase-architecture, tdd, to-issues, to-prd, zoom-out, scaffold-mvp, prototype, roadmap, secure-e2e, qa-analyst, query-docs, expo-expert, mcp-builder, superkuma-monitoring and developer.

architect acts as a senior Software, Solutions and Engineering Architect: designs systems for corporate level (production-readiness), runs scope alignment through active questioning, defines design tokens and screen mockups (MASTER+Overrides pattern) to guide developers, enforces engineering rigor (Clean Architecture, DDD, no N+1, OWASP, zero regression) and produces the key artifacts (ADD/SAD, C4 Model in Mermaid, ADRs, DER/MER schemas, OpenAPI/gRPC contracts and cloud/infrastructure topologies).

orchestrator is a compatibility shortcut: the same skill as developer (old name, kept so existing installations do not break).

expo-expert is stack-specific knowledge (Expo/React Native): CNG and config plugins, Expo Modules API, EAS Build/Submit/Update/Workflows, Expo Router, New Architecture, animation (Reanimated/Skia), Native UI, DOM components, data fetching and SDK upgrades. It teaches the agent to consult the living Expo documentation (llms.txt + .md) instead of trusting outdated training knowledge.

mcp-builder is a guide for building high-quality MCP (Model Context Protocol) servers in Python (FastMCP) and Node/TypeScript, with strict schemas, agent-oriented error handling and evaluation/benchmark scripts.

superkuma-monitoring covers the full SuperKuma monitoring cycle: deploying a new instance (Proxmox VM, Docker Compose + MariaDB), updating/upgrading an existing one, and operating the SuperKuma MCP server to discover a site's infrastructure (AD, Proxmox, VMware, Linux, pfSense, Mikrotik, switches, UniFi, TrueNAS, cameras) and turn it into monitors, tags, notifications and status pages.

Productivity

General workflow skills: caveman, grill-me, handoff, write-a-skill and skill-creator.

skill-creator is a meta-skill for creating, testing, evaluating (qualitatively and quantitatively with evals/HTML viewer) and optimizing the trigger accuracy of new skill descriptions.

Misc

Skills kept for specific cases: git-guardrails-claude-code, migrate-to-shoehorn, scaffold-exercises and setup-pre-commit.

Skills under personal/, in-progress/ and deprecated/ are not distributed.

Documentation governance

The consuming project should keep AGENTS.md or CLAUDE.md, CONTEXT.md or CONTEXT-MAP.md, DEVELOPER-ROADMAP.md, docs/agents/issue-tracker.md, docs/agents/triage-labels.md, docs/agents/domain.md and docs/adr/ when there are relevant decisions.

Credits and origin

This project is a fork of mattpocock/skills. The original architecture, the inspiration and much of the skills came from Matt Pocock. See also the skills newsletter to follow the original work.

About

Skills for Real Engineers. Straight from my .claude directory.

Resources

Stars

22 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages

0