convez.ai
SkillAutomaçãoIntermediário
Entelligentsia

Skillforge

Entelligentsia/skillforge

Skill packs modulares para potencializar agentes Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Coleção modular de skills para Claude Code e sistemas LLM agentes, incluindo geradores de código, integrações de API, plugins de segurança e referências de padrões de design e engenharia. Permite expandir capacidades de agentes IA com funcionalidades específicas de domínio, desde automação de tarefas até análise de segurança.

Instalação

$npx convez-ai Entelligentsia/skillforge

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
3
Forks
1
Acompanhando
0
Saúde
37%
Último commit
há 3 meses
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Automatizar análise de segurança em plugins recém-instalados
  • Gerar workflows e personas de engenharia customizadas por projeto
  • Integrar APIs externas (Freshdesk, etc.) em fluxos agentes
  • Referenciar padrões de design GoF, RAG e arquitetura de agentes
  • Avaliar eficiência de contexto em sessões Claude Code

Veredito

Skillforge é um ecossistema maduro e bem estruturado de skills prontos para produção. A segmentação entre plugins de automação (security-watchdog), integrações (freshdesk-api) e referências (padrões, LLM patterns) oferece valor real para equipes que já trabalham com Claude Code, embora exija familiaridade com o ecossistema de plugins do Claude.

README

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

Skillforge

Skill packs for Claude Code and other LLM agent systems, published by Entelligentsia (entelligentsia.in).

Available Packages

PackageTypeDescription
forgeMeta-generatorSelf-enhancing AI software development lifecycle — scans your codebase, generates project-specific workflows, personas, templates, and tools
security-watchdogSecurity pluginAuto-scans newly installed/updated Claude Code plugins for prompt injection, malicious hook scripts, and data exfiltration
freshdesk-apiAPI integration skillFreshdesk helpdesk API — tickets, contacts, companies, knowledge base (Solutions), webhooks (1 skill)
doc-reviewWorkflow pluginIn-browser document annotation overlay with Claude-driven apply — annotate HTML/MD/TXT in browser, click Apply, edits land in source (1 skill, 4 commands)
design-patternsReference skillsCanonical software design patterns — all 23 GoF + enterprise/DDD patterns (10 skills)
llm-patternsReference skillsLLM integration patterns — RAG, tool use, agents, guardrails, tool synthesis (9 skills)
harness-engineeringReference skillsAgent harness components — memory/compaction, caching, sandboxing, permissions, telemetry, hooks, routing (13 skills)
meta-webxr-skillsReference skillsMeta Quest PWA XR engineering (8 skills)
threejs-skillsReference skillsThree.js 3D development (10 skills)
lean-ctx-evalEval skillMeasure whether the lean-ctx MCP saves more context tokens than it costs, from your own Claude Code session transcripts (1 skill)

Installation (Claude Code)

/plugin marketplace add Entelligentsia/skillforge

Then install whichever packs you need:

/plugin install security-watchdog@skillforge
/plugin install freshdesk-api@skillforge
/plugin install doc-review@skillforge
/plugin install design-patterns@skillforge
/plugin install llm-patterns@skillforge
/plugin install harness-engineering@skillforge
/plugin install threejs-skills@skillforge
/plugin install meta-webxr-skills@skillforge
/reload-plugins

Forge

Forge has its own repository. See Entelligentsia/forge for installation instructions.

Forge

Forge is different from the reference skill packages. Instead of loading knowledge into context, it generates a complete project-specific engineering practice: agent personas, workflows, templates, review checklists, and tools — all tailored to your stack.

/forge init          # Bootstrap SDLC into your project
/sprint-plan         # Start your first sprint (generated command)
/engineer ACME-S01-T01   # Plan a task (generated command)

See Entelligentsia/forge for the full vision and design.

Skills Index

security-watchdog

Skill / CommandPurpose
/security-watchdog:scan-plugin <plugin-id>Scan any installed plugin for prompt injection, malicious hooks, and data exfiltration
plugin-securityThreat model and heuristics reference — attack taxonomy, severity guide, detection patterns

Runs automatically via SessionStart hook: detects newly installed or updated plugins and prompts Claude to scan before your first request.

design-patterns

SkillPatterns Covered
pattern-selectionEntry point — decision tree mapping pain to pattern
creationalSingleton, Builder, Factory Method, Abstract Factory, Prototype
structuralAdapter, Facade, Decorator, Proxy, Composite, Flyweight, Bridge
behaviouralChain of Responsibility, Command, Strategy, State, Observer, Memento, Mediator, Visitor, Iterator, Template Method
domain-modelingEntity, Value Object, Aggregate, Aggregate Root
data-accessRepository, Unit of Work, Data Mapper, Active Record
service-layerApplication Service, Domain Service, Service Layer
domain-eventsDomain Events, Transactional Outbox, eventual consistency
cqrsCommands, Queries, Read Models, Projections
anti-corruptionAnti-Corruption Layer, Gateway, Strangler Fig

llm-patterns

SkillPain It Removes
pattern-selectionEntry point — decision tree for LLM integration patterns
structured-generationOutput breaks parsers, violates schemas, varies in shape
ragLLM hallucinates, lacks domain knowledge, gives stale answers
tool-useLLM needs live data, calculations, or side effects
agent-loopTask requires autonomous multi-step reasoning
guardrailsOutput contains harmful content, PII, or policy violations
prompt-engineeringPrompts are ad-hoc, untested, unversioned
graceful-degradationModel is down, slow, or over budget
evaluation-harnessNo way to measure quality or detect regressions
tool-synthesisLLM called repeatedly for tasks codifiable as deterministic tools

harness-engineering

SkillConcern
memory-compactionHot/warm/cold tiering; lossy compression that preserves load-bearing facts
session-persistenceAppend-only event log, idempotent replay, fork/branch semantics
prompt-cachingCache breakpoint placement, TTL, silent-regression traps
subagent-orchestrationFork vs fresh, briefing rules, isolation, result merge
tool-sandboxingFilesystem jail, egress allowlist, scoped credentials, safety vs security
permission-gatesRisk classification, scope-bound consent, irreversibility detection
streaming-ioPartial JSON parsing, mid-stream tool dispatch, cancellation
concurrency-controlConflict graphs, intent-order feed, partial-failure semantics
telemetry-tracingSpan trees, cost attribution, replay from trace
hook-systemEvent taxonomy, blocking vs observation, failure isolation
workspace-stateRead-before-edit invariant, external-mutation detection
model-routingCapability/cost routing, distinct from graceful degradation
rate-limitingToken buckets per dimension, fair allocation, 429 with Retry-After

meta-webxr-skills

SkillTrigger
webxr-sessionWebXR session lifecycle, requestSession, feature flags
webxr-renderingXR render loop, reference spaces, frame timing
webxr-inputController input, hand tracking, hit testing
webxr-passthroughAR/MR passthrough, plane/mesh detection
webxr-anchorsPersistent spatial anchors
webxr-layersWebXR Layers API, compositing
webxr-ratkReality Accelerator Toolkit (Three.js wrapper)
webxr-pwa-questPWA manifest, service worker, Meta Quest packaging

threejs-skills

SkillTrigger
threejs-fundamentalsScene setup, cameras, renderer, Object3D hierarchy
threejs-geometryBuilt-in shapes, BufferGeometry, custom geometry, instancing
threejs-materialsPBR materials, shader materials, material properties
threejs-lightingLight types, shadows, environment lighting
threejs-texturesTexture types, UV mapping, environment maps
threejs-animationKeyframe animation, skeletal animation, morph targets
threejs-loadersGLTF loading, texture loading, async patterns
threejs-shadersGLSL, ShaderMaterial, uniforms, custom effects
threejs-postprocessingEffectComposer, bloom, DOF, screen effects
threejs-interactionRaycasting, controls, mouse/touch input, object selection

freshdesk-api

SkillTrigger
freshdesk-apiFreshdesk API: tickets, contacts, companies, knowledge base articles, webhooks, automation

Five reference files cover the full API surface: tickets, contacts/companies, Solutions (knowledge base), webhooks/automation, and SDK examples.

doc-review

CommandPurpose
/doc-review:review-doc <file>Start in-browser annotation overlay on HTML/MD/TXT; runs auto-apply loop
/doc-review:apply-review [<json>]Guided apply of pending review comments back into source
/doc-review:review-doc-stopDrop this session's references; servers exit when last ref drops

Multi-session aware: two sessions reviewing the same doc share one server. Apply button in browser queues edits; Claude applies within ~60s. Stores comments under <project-root>/.doc-review/.

Acknowledgements

Ver o original no GitHub

Bom para

Relacionados