Select Language / Selecione o Idioma:
FrameCode VibeWork Γ© um framework de governanΓ§a documental e tΓ©cnica para desenvolvimento de aplicaΓ§Γ΅es assistido por IA. Ele reduz perda de contexto entre sessΓ΅es ao combinar planos formais, changelogs, auditorias, troubleshooting, decisΓ΅es arquiteturais, snippets reutilizΓ‘veis e uma LLM Wiki mantida em Markdown.
O framework usa um ciclo de vida explΓcito para garantir que mudanΓ§as sejam justificadas, planejadas, implementadas, validadas e registradas.
graph TD
A["Fase 0: Briefing"] -->|"DefiniΓ§Γ£o de escopo"| B["Manifesto e Escopo"]
B -->|"SolicitaΓ§Γ£o de mudanΓ§a"| C["Plano de alteraΓ§Γ£o"]
C -->|"ExecuΓ§Γ£o assistida por IA"| D["ImplementaΓ§Γ£o"]
D -->|"Rastreabilidade"| E["Changelog e versionamento"]
E -->|"ValidaΓ§Γ£o"| F["Auditoria e release"]
F -->|"Aprendizado"| G["Wiki / memΓ³ria tΓ©cnica"]
G -->|"PublicaΓ§Γ£o documental"| H["DocumentaΓ§Γ£o pΓΊblica (Wiki + GitHub Pages)"]
H -->|"Contexto acumulado"| C
Nenhuma alteraΓ§Γ£o funcional, visual, estrutural ou documental deve ser aplicada sem plano correspondente em Plans/.
Toda alteraΓ§Γ£o em arquivo versionado deve ser registrada em changelogs/Vx.y.z.md, com plano relacionado, arquivos afetados, validaΓ§Γ£o e riscos residuais.
A pasta wiki/ segue o padrΓ£o LLM Wiki: fontes brutas, pΓ‘ginas sintetizadas, Γndice, log, links internos, estados de confianΓ§a e lint periΓ³dico.
Portabilidade e Reuso: O conhecimento acumulado na
wiki/(padrΓ΅es, decisΓ΅es, troubleshooting) pode e deve ser portado e reutilizado em novos projetos para acelerar o desenvolvimento assistido por IA e manter a consistΓͺncia tΓ©cnica entre diferentes aplicaΓ§Γ΅es.
A pasta snippets/ guarda componentes e padrΓ΅es prontos para adaptaΓ§Γ£o, com galeria visual e tokens CSS.
A pasta governance/ preserva templates genΓ©ricos. Os documentos preenchidos do projeto ficam na raiz. A instanciaΓ§Γ£o e as regras de renomeaΓ§Γ£o estΓ£o em INSTANTIATION.md.
A pasta skills/ armazena procedimentos tΓ©cnicos de alta densidade carregados sob demanda pelo agente de IA β nunca prΓ©-carregados no prompt. Isso preserva a janela de contexto enquanto disponibiliza checklists especializados quando necessΓ‘rios.
Habilidades ativas: obsidian-markdown, git-conventional-commits, wiki-lint, release-checklist.
π .
βββ π AGENTS.md
βββ π CONTEXT_MAP.md
βββ π MANIFEST.md
βββ π README.md
βββ π INSTANTIATION.md
βββ π SCOPE.md
βββ π STACK.md
βββ π DESIGN.md
βββ π WORKFLOW.md
βββ π PLANNING.md
βββ π VERSIONING.md
βββ π TROUBLESHOOTING.md
βββ π TESTS.md
βββ π SECURITY.md
βββ π DATA.md
βββ π AI.md
βββ π REFACTORING.md
βββ π RELEASE.md
βββ π AUDIT.md
βββ π ARCHITECTURAL_DECISIONS.md
βββ π FILESYSTEM.md
βββ π Plans/
β βββ π pending/
β βββ π in_progress/
β βββ π completed/
β βββ π discontinued/
βββ π changelogs/
βββ π troubleshooting/
βββ π decisions/
βββ π audits/
βββ π briefings/
βββ π wiki/
β βββ π schema.md
β βββ π index.md
β βββ π log.md
β βββ π templates/
β βββ π sessions/
β βββ π patterns/
β βββ π decisions/
β βββ π releases/
β βββ π failures/
βββ π skills/
β βββ π README.md
β βββ π obsidian-markdown/
β βββ π git-conventional-commits/
β βββ π wiki-lint/
β βββ π release-checklist/
βββ π snippets/
βββ π governance/
Plans/: ciclo de vida das alteraΓ§Γ΅es.changelogs/: histΓ³rico formal por versΓ£o.troubleshooting/: registros de falhas, hipΓ³teses e validaΓ§Γ΅es.decisions/: ADRs formais.audits/: relatΓ³rios de auditoria.briefings/: registros de descoberta e Fase 0.wiki/: memΓ³ria tΓ©cnica compatΓvel com Obsidian (padrΓ΅es, decisΓ΅es, falhas, releases, sΓnteses).skills/: catΓ‘logo de habilidades executΓ‘veis do agente de IA β carregadas sob demanda (JIT).snippets/: biblioteca de componentes e padrΓ΅es reutilizΓ‘veis.governance/: templates genΓ©ricos do framework.CONTEXT_MAP.md: mapa compacto de carregamento seletivo por tipo de sessΓ£o.
Para maximizar a transparΓͺncia de custos de chamadas de APIs de LLMs, o framework mapeia o consumo de tokens estimado para cada cenΓ‘rio de desenvolvimento com base em suas polΓticas ativas:
| CenΓ‘rio Mapeado | Documentos Ingeridos | Custo Inicial (Sem AICC) | Custo por Turno com AICC | Economia com AICC |
|---|---|---|---|---|
| Bugfix / Troubleshooting | AGENTS.md + TROUBLESHOOTING.md + PLANNING.md |
~5.000 tokens | ~1.200 tokens | -76% |
| Nova Funcionalidade | AGENTS.md + SCOPE.md + PLANNING.md + DESIGN.md |
~7.000 tokens | ~1.500 tokens | -78% |
| Componentes / UI | AGENTS.md + DESIGN.md + snippets/README.md |
~4.500 tokens | ~1.000 tokens | -77% |
| RefatoraΓ§Γ£o | AGENTS.md + REFACTORING.md + PLANNING.md |
~8.000 tokens | ~1.800 tokens | -77% |
| Briefing / InstanciaΓ§Γ£o | AGENTS.md + INSTANTIATION.md + BRIEFING.md + MANIFEST.md |
~8.500 tokens | ~2.000 tokens | -76% |
| Release | CONTEXT_MAP.md + skill:release-checklist (JIT) |
~2.500 tokens | ~600 tokens | -76% |
Nota: As estimativas consideram o tamanho mΓ©dio atual dos arquivos de governanΓ§a do framework. 1 token β 4 caracteres em inglΓͺs ou ~3 caracteres em portuguΓͺs.
Use este repositΓ³rio como base para um novo projeto ou mantenha-o como framework central.
git clone https://github.com/Sistema2D/FrameCode-VibeWork.git meu-projeto
cd meu-projetoLeia AGENTS.md e INSTANTIATION.md. A instanciaΓ§Γ£o nΓ£o depende de script automΓ‘tico: renomeaΓ§Γ΅es e substituiΓ§Γ΅es devem ser feitas explicitamente, preservando templates em governance/ e wiki/templates/.
Preencha BRIEFING.md, atualize MANIFEST.md, STACK.md, SCOPE.md e README.md, e registre a alteraΓ§Γ£o por plano e changelog.
Ao solicitar mudanΓ§as, peΓ§a para o agente seguir AGENTS.md. Para consultas, anΓ‘lise e revisΓ£o sem ediΓ§Γ£o de arquivos, plano nΓ£o Γ© obrigatΓ³rio. Para qualquer alteraΓ§Γ£o, o fluxo de plano e changelog Γ© obrigatΓ³rio.
FrameCode VibeWork is a technical and document-based governance framework for AI-assisted application development. It reduces context loss between sessions by combining formal plans, changelogs, audits, troubleshooting, architectural decisions, reusable snippets, and an LLM Wiki maintained in Markdown.
The framework uses an explicit lifecycle to ensure that changes are justified, planned, implemented, validated, and recorded.
graph TD
A["Phase 0: Briefing"] -->|"Scope definition"| B["Manifest and Scope"]
B -->|"Change request"| C["Change plan"]
C -->|"AI-assisted execution"| D["Implementation"]
D -->|"Traceability"| E["Changelog and versioning"]
E -->|"Validation"| F["Audit and release"]
F -->|"Learning"| G["Wiki / technical memory"]
G -->|"Documentation publishing"| H["Public documentation (Wiki + GitHub Pages)"]
H -->|"Accumulated context"| C
No functional, visual, structural, or document change should be applied without a corresponding plan in Plans/.
Every change in a versioned file must be recorded in changelogs/Vx.y.z.md, with related plans, affected files, validation, and residual risks.
The wiki/ folder follows the LLM Wiki standard: raw sources, synthesized pages, index, log, internal links, confidence states, and periodic linting.
Portability and Reuse: The knowledge accumulated in the
wiki/(patterns, decisions, troubleshooting) can and should be ported and reused in new projects to accelerate AI-assisted development and maintain technical consistency across different applications.
The snippets/ folder holds components and patterns ready for adaptation, with a visual gallery and CSS tokens.
The governance/ folder preserves generic templates. Filled project documents reside at the root. Instantiation and renaming rules are in INSTANTIATION.md.
The skills/ folder stores high-density technical procedures loaded on-demand by the AI agent β never pre-loaded into the prompt. This preserves the context window while making specialized checklists available when needed.
Active skills: obsidian-markdown, git-conventional-commits, wiki-lint, release-checklist.
π .
βββ π AGENTS.md
βββ π CONTEXT_MAP.md
βββ π MANIFEST.md
βββ π README.md
βββ π INSTANTIATION.md
βββ π SCOPE.md
βββ π STACK.md
βββ π DESIGN.md
βββ π WORKFLOW.md
βββ π PLANNING.md
βββ π VERSIONING.md
βββ π TROUBLESHOOTING.md
βββ π TESTS.md
βββ π SECURITY.md
βββ π DATA.md
βββ π AI.md
βββ π REFACTORING.md
βββ π RELEASE.md
βββ π AUDIT.md
βββ π ARCHITECTURAL_DECISIONS.md
βββ π FILESYSTEM.md
βββ π Plans/
β βββ π pending/
β βββ π in_progress/
β βββ π completed/
β βββ π discontinued/
βββ π changelogs/
βββ π troubleshooting/
βββ π decisions/
βββ π audits/
βββ π briefings/
βββ π wiki/
β βββ π schema.md
β βββ π index.md
β βββ π log.md
β βββ π templates/
β βββ π sessions/
β βββ π patterns/
β βββ π decisions/
β βββ π releases/
β βββ π failures/
βββ π skills/
β βββ π README.md
β βββ π obsidian-markdown/
β βββ π git-conventional-commits/
β βββ π wiki-lint/
β βββ π release-checklist/
βββ π snippets/
βββ π governance/
Plans/: lifecycle of changes.changelogs/: formal version history.troubleshooting/: records of failures, hypotheses, and validations.decisions/: formal ADRs.audits/: audit reports.briefings/: discovery and Phase 0 records.wiki/: Obsidian-compatible technical memory (patterns, decisions, failures, releases, syntheses).skills/: catalog of executable AI agent skills β loaded on-demand (JIT).snippets/: library of reusable components and patterns.governance/: generic templates of the framework.CONTEXT_MAP.md: compact selective loading map by session type.
To maximize transparency and API call cost-efficiency with LLMs, the framework maps the estimated token consumption for each development scenario based on its active policies:
| Mapped Scenario | Ingested Documents | Initial Load (No AICC) | Continuous Turn Cost (With AICC) | Savings with AICC |
|---|---|---|---|---|
| Bugfix / Troubleshooting | AGENTS.md + TROUBLESHOOTING.md + PLANNING.md |
~5,000 tokens | ~1,200 tokens | -76% |
| New Feature | AGENTS.md + SCOPE.md + PLANNING.md + DESIGN.md |
~7,000 tokens | ~1,500 tokens | -78% |
| Componentes / UI | AGENTS.md + DESIGN.md + snippets/README.md |
~4,500 tokens | ~1,000 tokens | -77% |
| Refactoring | AGENTS.md + REFACTORING.md + PLANNING.md |
~8,000 tokens | ~1,800 tokens | -77% |
| Briefing / Instantiation | AGENTS.md + INSTANTIATION.md + BRIEFING.md + MANIFEST.md |
~8,500 tokens | ~2,000 tokens | -76% |
| Release | CONTEXT_MAP.md + skill:release-checklist (JIT) |
~2,500 tokens | ~600 tokens | -76% |
Note: Estimates are based on the current average size of the framework's governance files. 1 token β 4 characters in English or ~3 characters in Portuguese.
Use this repository as a base for a new project or keep it as a central framework.
git clone https://github.com/Sistema2D/FrameCode-VibeWork.git my-project
cd my-projectRead AGENTS.md and INSTANTIATION.md. Instantiation does not rely on automatic scripts: renaming and replacements must be done explicitly, preserving templates in governance/ and wiki/templates/.
Fill out BRIEFING.md, update MANIFEST.md, STACK.md, SCOPE.md, and README.md, and record the change via plan and changelog.
When requesting changes, ask the agent to follow AGENTS.md. For queries, analysis, and reviews without file editing, a plan is not mandatory. For any modification, the plan and changelog workflow is mandatory.
Abra a pasta raiz como um vault no Obsidian para visualizar links entre decisΓ΅es, falhas, padrΓ΅es, auditorias, releases e sΓnteses da wiki. / Open the root folder as a vault in Obsidian to visualize links between decisions, failures, patterns, audits, releases, and wiki syntheses.
O conceito de LLM Wiki usado como inspiraΓ§Γ£o para a memΓ³ria tΓ©cnica incremental deste framework Γ© creditado a Andrej Karpathy, autor do gist LLM Wiki. / The LLM Wiki concept used as inspiration for the incremental technical memory of this framework is credited to Andrej Karpathy, author of the LLM Wiki gist.
Se este framework for ΓΊtil para o seu trabalho, vocΓͺ pode apoiar o desenvolvimento pelo Buy Me a Coffee: / If this framework is useful for your work, you can support development via Buy Me a Coffee:
Este projeto estΓ‘ licenciado sob a licenΓ§a MIT. Veja LICENSE. / This project is licensed under the MIT license. See LICENSE.