convez.ai
SkillAutomaçãoAvançado
bcmyguest

Personal Skills — Agentes e Skills Reutilizáveis

bcmyguest/personal-skills

30+ skills reutilizáveis para agentes de IA — diagnóstico, setup, refatoração

Ver no GitHub
InicialPluginScore · de 0 a 100

Repositório de skills portáveis no formato aberto que qualquer agente de coding moderno (Claude Code, Cursor, etc.) consegue ler e executar. Inclui 30+ skills prontas para diagnóstico, desenvolvimento, experimentos, DevOps e refatoração. Instale com `npx skills add` e reutilize em múltiplos agentes via symlink ou cópia.

Instalação

$npx convez-ai bcmyguest/personal-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
42%
Último commit
há 4 dias
Versão
—
Issues abertas
1
Contribuidores
5

Casos de uso

  • Diagnosticar e corrigir falhas rapidamente com paths otimizados
  • Scaffolding automatizado de repos (Python uv, React+TS, Rust)
  • Revisar diffs e PRs antes de fazer merge com auditoria de IA
  • Planejar e rodar experimentos controlados com monitoramento
  • Refatoração de frontend como engenheiro sênior

Veredito

Coleção madura e bem estruturada de skills reutilizáveis que eleva o nível de automação de agentes de IA. Exige conforto com CLI e fluxo de trabalho de agentes, mas economiza tempo significativo em operações repetidas de DevOps, refatoração e testes.

README

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

personal-skills

Portable agent skills — one SKILL.md per skill, in the open skills format that any modern coding agent reads (Claude Code, Codex, opencode, Cursor, …). The repo itself is the package; GitHub is the registry.

Install

npx skills add bcmyguest/personal-skills          # pick skills + agents interactively
npx skills add bcmyguest/personal-skills --all    # every skill -> every detected agent
npx skills add bcmyguest/personal-skills -a claude-code -a codex -a opencode
npx skills update                                  # pull later updates

skills symlinks each skill into your agent's native skills directory (single source of truth, easy updates), or copies with --copy where symlinks aren't supported. See vercel-labs/skills.

Without installing, point an agent at the file: "Follow skills/diagnose-failure/SKILL.md from bcmyguest/personal-skills."

Skills

SkillWhat
diagnose-failureFind and fix the actual cause of a failure, with fast and strict paths
sanity-check-numbersDecide whether a metric, score, or cost is real before trusting it
audit-agent-diffCheck an agent-written diff for overreach before it lands
land-changeCommit, push, PR, and merge exactly the intended change
port-changesBring selected work across branches, PRs, or archives
match-ui-to-referenceConverge a UI on a design through numbered deltas and screenshots
resolve-design-decisionsSettle costly-to-reverse choices in dependency-ordered question rounds
research-optionsCompare options with primary sources and spikes
independent-reviewGet a verdict from a reviewer that did not write the artifact
brief-agentWrite a self-contained brief for the next agent or session
plan-experimentPre-register a comparison before spending on it
run-experimentRun an approved plan, monitor it, and keep it resumable
write-evidence-reportPublish results that say exactly what the evidence supports
uv-setupScaffold a Python repo with uv from shipped templates — CI, auto-releases, PyPI Trusted Publishing
uv-developCorrective uv operations — targeted upgrades, --locked vs --frozen, venv repair
pre-commit-setupStand up pre-commit from shipped fragments — hygiene + Conventional Commits + hook-update policy
release-setupRelease automation from Conventional Commits — canonical git-cliff config + GitLab tag/release pipeline
use-templateStart a repo from your GitHub template repos (Rust CLI, npm, Python, docs site, skills pack)
watchlistTrack upstream GitHub issues, PRs, and releases
assign-mr-reviewersAssign CODEOWNERS reviewers to your open GitLab MRs
rigsInstall OpenRig, hand a goal to its operator, and follow a coding-agent team's work (vendored from OpenRig 0.6.7)
react-ts-setupScaffold a React + TypeScript repo with Vite + full toolchain
rust-setupScaffold a Rust repo from shipped templates — CI, auto-releases, crates.io Trusted Publishing
senior-frontend-refactorRefactor frontend code as a senior engineer
add-ansible-roleAdd a tool-install role to a personal Ansible playbook
debug-lemonadeDiagnose lemonade-server on the Strix Halo box
security-auditFind and adversarially verify a codebase's 1–3 worst security vulnerabilities (capped Sonnet multi-agent workflow)
efficiency-auditFind and verify a codebase's 1–3 worst efficiency problems with concrete cost analyses (capped Sonnet multi-agent workflow)

Layout: one folder per skill under skills/, each with a SKILL.md plus any resources it needs (references/, templates/, config.json).

writing-with-clarity moved to bcmyguest/claude-memory; fast-hypothesis and reproduce-bug are folded into diagnose-failure.

Skill chains

Workflows are wteam teams in wteam/teams/, one team.dot per team: each worker step and reviewer names its skill with skill=, and a ticket walks every step before it commits once. wteam/ mirrors the wteam config dir, so it also carries the artifact-review phase and its persona roles. Import it:

scripts/install-wteam.sh   # copies wteam/ into ${WTEAM_CONFIG_DIR:-~/.config/wteam}, then lists teams and phases
wteam run --pairs 1 --team experiment --slug <slug>

Hand-offs outside a team come from each skill's ## Next section. After editing a team or a ## Next section, regenerate and check:

python3 scripts/chains.py            # writes CHAINS.md and this section
python3 scripts/chains.py --check    # fails on stale output, a missing skill, or a team hand-off no ## Next states
python3 scripts/chains.py --check --wteam   # also parses every team with wteam
python3 scripts/test_chains.py       # check-mode and routing-example tests
TeamPath
changethe ticket's skill → audit-agent-diff (review)
experimentplan-experiment → independent-review (review) → gate user approves spend → run-experiment → sanity-check-numbers (review) → write-evidence-report → independent-review (review)
flowchart LR
  audit_agent_diff["audit-agent-diff"]
  land_change["land-change"]
  bugbot(["Bugbot"])
  brief_agent["brief-agent"]
  next_agent{{"start next agent"}}
  diagnose_failure["diagnose-failure"]
  refine(["refine (bcmyguest/wteam)"])
  independent_review["independent-review"]
  resolve_design_decisions["resolve-design-decisions"]
  match_ui_to_reference["match-ui-to-reference"]
  plan_experiment["plan-experiment"]
  spend_approval{{"user approves spend"}}
  run_experiment["run-experiment"]
  port_changes["port-changes"]
  research_options["research-options"]
  watchlist["watchlist"]
  to_spec(["to-spec (mattpocock/skills)"])
  sanity_check_numbers["sanity-check-numbers"]
  write_evidence_report["write-evidence-report"]
  audit_agent_diff -->|"clean diff"| land_change
  audit_agent_diff --o|"alternative: general bug finding on the same diff"| bugbot
  brief_agent -.->|"brief"| next_agent
  diagnose_failure -->|"fix and its live check"| audit_agent_diff
  diagnose_failure --o|"alternative: when the failure is an agent pipeline behaving badly rather than a crash"| refine
  independent_review -.->|"verdict"| resolve_design_decisions
  independent_review -.->|"approved artifact"| land_change
  match_ui_to_reference -->|"screenshots and the resolved delta list"| audit_agent_diff
  plan_experiment -->|"pre-registered plan"| independent_review
  plan_experiment -->|"pre-registered plan"| spend_approval
  plan_experiment ==>|"approved plan"| run_experiment
  port_changes -->|"applied subset and the left-out list"| audit_agent_diff
  research_options -->|"findings or ADR"| independent_review
  research_options -->|"facts per option"| resolve_design_decisions
  research_options -.->|"pinned items that wait on an upstream issue"| watchlist
  resolve_design_decisions -.->|"decision record"| independent_review
  resolve_design_decisions -.->|"decision record"| to_spec
  run_experiment -->|"result rows"| sanity_check_numbers
  run_experiment -.->|"pipeline arm that stalls or loops"| refine
  sanity_check_numbers -->|"wrong number and its traced definition"| diagnose_failure
  sanity_check_numbers -->|"verified numbers"| write_evidence_report
  write_evidence_report -->|"draft"| independent_review
  classDef external stroke-dasharray: 4 3
  classDef action fill:#fff4d6,stroke:#b8860b
  class bugbot,refine,to_spec external
  class next_agent,spend_approval action

Solid arrows are transitions a skill states, dashed arrows are suggestions, and circle-ended arrows are alternatives. CHAINS.md draws each team and lists every hand-off.

Skills with no chain: add-ansible-role, assign-mr-reviewers, debug-lemonade, efficiency-audit, pre-commit-setup, react-ts-setup, release-setup, rigs, rust-setup, security-audit, senior-frontend-refactor, uv-develop, uv-setup.

The portable handoff and pickup skills moved to their own repo — bcmyguest/baton (npx skills add bcmyguest/baton).

Claude Code plugins (hook-bearing — not plain skills)

Some tools need a hook, which the portable skills format can't carry. Those are Claude Code plugins and now live in their own repos:

PluginWhatRepo
assisted-bygit-attribution skill plus a PreToolUse hook that enforces AI commit attribution on git commitbcmyguest/assisted-by
skill-injectLocal, model-agnostic automatic skill injectionbcmyguest/skill-injector
claude-memorySession conventions (MEMORY.md) plus /install script for @include injectionbcmyguest/claude-memory
claude plugin marketplace add bcmyguest/assisted-by
claude plugin install assisted-by@assisted-by

Bootstrap my own machine

For my own setup, install-plugins.sh is an idempotent bootstrap that registers the assisted-by, caveman, and anthropic-agent-skills GitHub marketplaces, then installs and enables the plugins listed in plugins.json. It does not install the portable skills/ — use npx skills above for those.

git clone git@github.com:bcmyguest/personal-skills.git ~/personal-skills
~/personal-skills/install-plugins.sh

Layout

Two tracks: portable skills install with npx skills; hook-bearing plugins install through the Claude Code marketplace.

skills/                          # portable agent skills — one folder per skill (npx skills)
  uv-setup/                  SKILL.md + templates/
  uv-develop/                SKILL.md
  pre-commit-setup/          SKILL.md + templates/
  release-setup/             SKILL.md + templates/
  use-template/              SKILL.md
  watchlist/                 SKILL.md
  assign-mr-reviewers/       SKILL.md + scripts/
  react-ts-setup/            SKILL.md + references/
  rust-setup/                SKILL.md + templates/
  senior-frontend-refactor/  SKILL.md
  add-ansible-role/          SKILL.md + config.json + templates/
  debug-lemonade/            SKILL.md
  diagnose-failure/ … write-evidence-report/   SKILL.md each (method skills, see CHAINS.md)
  security-audit/            SKILL.md + scripts/ + evals/
  efficiency-audit/          SKILL.md + scripts/ + evals/
  _fixtures/                 shared eval fixture (vuln-app) for the audit skills
skills.sh.json                   # skills.sh groupings for the skills above
plugins.json                     # external marketplaces + enable list for install-plugins.sh
install-plugins.sh               # personal-machine bootstrap (hook-bearing plugins)
IDEAS.md                         # parked skill candidates and config snippets
wteam/teams/<name>/team.dot      # wteam teams, one per workflow (the source of the chains)
wteam/phases/, wteam/roles/      # the artifact-review phase and its personas
scripts/install-wteam.sh         # copies wteam/ into the wteam config dir
chains/routing.md                # routing examples and near-misses per skill
scripts/chains.py                # checks teams against ## Next; writes CHAINS.md and the Skill chains section
CHAINS.md                        # generated: team diagrams, hand-off graph and table
DECISIONS.md                     # why the method skills look the way they do

License

See LICENSE. All rights reserved; in particular, the contents may not be used as training, fine-tuning, or evaluation data for machine-learning or AI systems.

Ver o original no GitHub

Bom para

Relacionados