convez.ai
Skill
heyimcarlos

agent-skills

heyimcarlos/agent-skills

Ver no GitHub
InicialPluginScore · de 0 a 100

Carlos's agent skills, subagents, and slash commands for Claude Code, Gemini CLI, and OpenAI Codex CLI. Claude plugin marketplace.

Instalação

$npx convez-ai heyimcarlos/agent-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 2 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

agent-skills

Skills and subagents for AI coding agents — one source of truth, installed natively in Claude Code, Gemini CLI, OpenAI Codex CLI, and 50+ other agent CLIs through vercel-labs/skills.

Skills encode the workflows a senior engineer follows: research before coding, spec before plan, plan before implement, verify before ship. Packaged so the same workflows trigger across every tool you drive.

Quickstart

npx skills@latest add heyimcarlos/agent-skills

Pick which skills to install and which agents to install them into. The CLI auto-detects supported agent CLIs on your machine and writes SKILL.md files into each one's native skill path.

Claude Code (full plugin)

npx skills add installs skills only. To also pick up the research subagents in agents/, install via Claude Code's plugin marketplace:

/plugin marketplace add heyimcarlos/agent-skills
/plugin install agent-skills@agent-skills

Layout

skills/
  engineering/    # daily code work — git, PRs, planning, implementation, debug
  qrspi/          # the QRSPI workflow as prefixed qrspi-* setup, create, and iterate skills
  misc/           # reserved for rarely-used skills
agents/           # Claude Code subagents (codebase + thoughts + web research)
.claude-plugin/
  plugin.json     # single-plugin manifest read by `npx skills` and Claude Code

QRSPI workflow

Each stage is its own skill, with an iterate-* sibling for surgical adjustments without restarting the workflow.

StageSkillPurpose
Questionqrspi-create-research-questionsTurn a ticket into targeted questions before any code is read
Researchqrspi-create-researchDocument the codebase as-is, grounded in real files
Specqrspi-create-design-discussion → qrspi-create-structure-outlineSynthesize research into architectural decisions and a phased outline
Planqrspi-create-planConvert outline into a rigid step-by-step plan with dual verification
Implementqrspi-create-worktreeLaunch an isolated implementation session from an approved plan
Setupqrspi-setupInitialize the local thoughts/ workspace used by the QRSPI skills

QRSPI skills use a qrspi- name prefix so they are easy to find and select during installation. They set disable-model-invocation: true — the human invokes each phase deliberately rather than letting the model auto-trigger them.

Every doc-producing QRSPI skill (all of the above except qrspi-create-worktree) accepts an optional --output=html flag. When present, the skill emits a reveal.js slide deck alongside the canonical markdown — color, SVG diagrams, annotated code, comparison tables. Iterate skills detect the sibling .html and refresh both. See skills/qrspi/HTML-OUTPUT.md for the conventions and per-phase slide structures.

Reference

Engineering

Code-work skills — git, PRs, planning, implementation, debug.

  • commit — Create focused git commits with user approval
  • debug — Debug issues by investigating logs, database state, and git history.
  • defragance — Distill the taste, idioms, and operating habits of strong comparables into project guidance.
  • deslop — Remove AI-generated code slop and clean up code style.
  • find-comparables — Research mature comparable repositories, language standards, and books to extract best practices for a project or rewrite.
  • get-pr-comments — Fetch and summarize review comments from the active pull request.
  • review-and-ship — Review the current branch for bugs, intent fit, and test coverage; run or write tests; commit focused work; open or update a PR.
  • thermo-nuclear-code-quality-review — Run an extremely strict maintainability review for abstraction quality, giant files, and spaghetti-condition growth.

QRSPI

The Question → Research → Spec → Plan → Implement workflow as explicit skills.

  • qrspi-create-research-questions — Analyze a ticket to generate targeted research questions before any codebase investigation begins.
  • qrspi-iterate-research-questions — Refine research questions by adding, removing, or reframing them based on user feedback.
  • qrspi-create-research — Document codebase as-is using parallel sub-agents, guided only by the research questions document.
  • qrspi-iterate-research — Deepen or correct codebase research based on user feedback using targeted sub-agents.
  • qrspi-create-design-discussion — Synthesize research and ticket into architectural decisions, surfacing options and patterns for human review.
  • qrspi-iterate-design-discussion — Refine design options and converge on an architectural decision through collaborative discussion.
  • qrspi-create-structure-outline — Translate an approved design into a phased structural outline with vertical slices and verification checkpoints.
  • qrspi-iterate-structure-outline — Adjust phase ordering, boundaries, or verification checkpoints in the structure outline.
  • qrspi-create-plan — Convert an approved structure outline into a rigid step-by-step implementation plan with dual verification criteria.
  • qrspi-iterate-plan — Surgically adjust the implementation plan when scope changes or implementation hits a mismatch.
  • qrspi-create-worktree — Create an isolated git worktree and launch an implementation session from an approved plan.
  • qrspi-setup — Set up the local thoughts/ workspace and ignore it from git.

Subagents (Claude Code only)

Specialist research personas in agents/, invoked from skills via the Agent tool. Only installed by the Claude Code plugin path; npx skills doesn't install subagents into other CLIs (they don't have the concept).

Editing

Canonical content lives in skills/<bucket>/<name>/SKILL.md and agents/<name>.md. After adding or removing a skill, update .claude-plugin/plugin.json and the reference section above.

See AGENTS.md (alias CLAUDE.md) for repo conventions, skill anatomy, and orchestration rules.

License

MIT.

Ver o original no GitHub

Bom para

Relacionados