convez.ai
AgenteDesenvolvimento WebAvançado
MullaAhmed

dummyIndex

MullaAhmed/dummyIndex

Indexação inteligente de codebases para Claude Code com planejamento e orquestração multi-agente

Ver no GitHub
BomPluginScore · de 0 a 100

Motor de contexto persistente para repositórios que transforma qualquer codebase em uma pasta `.context/` navegável pelo Claude Code, sem necessidade de grep. Combina extração determinística por AST com um conselho multi-agente (desenvolvedor, arquiteto, críticos) que adiciona julgamento semântico.

Instalação

$npx convez-ai MullaAhmed/dummyIndex

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
9
Forks
0
Acompanhando
0
Saúde
71%
Último commit
há 3 dias
Versão
v0.38.1
Issues abertas
0
Contribuidores
2

Casos de uso

  • Navegação eficiente de grandes codebases dentro do Claude Code
  • Planejamento de features com propostas validadas e toolkits auto-equipados
  • Manutenção de contexto consistente entre sessões de desenvolvimento
  • Orquestração de agentes para implementação de propostas complexas

Veredito

Ferramenta sofisticada e bem estruturada que resolve um problema real: manter contexto navegável em grandes projetos para Claude Code. A abordagem de conselho multi-agente + índice determinístico é sólida, mas exige investimento inicial e compreensão profunda do ecossistema Claude para máximo aproveitamento.

README

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

dummyindex

The persistent context engine for a repo. A Claude Code skill that turns any codebase into a .context/ folder Claude can navigate without grepping — deterministic AST extraction plus a multi-agent council (dev, architect, critics) that fills in the judgment.

pip install --user dummyindex          # or: uv tool install dummyindex
dummyindex install                     # one-time, user-global
cd /path/to/your/repo
claude                                 # open Claude Code in your repo
> /dummyindex <path>                   # e.g. /dummyindex ./src

The bootstrap above (pip install + dummyindex install) is the only time you touch the terminal — after that your interface is the slash commands inside Claude Code, and the rest of the CLI is the agent's deterministic backbone (the skill and council invoke it; you don't run it by hand).

After the first run, every future Claude Code session in this repo consults .context/ before reading source at random. A SessionStart hook surfaces what drifted since the last update, and the session reconciles the index in place.


What it is

dummyindex runs in two modes per repo. Setup mode (one-time): /dummyindex builds .context/, installs hooks, and writes the CLAUDE.md managed block. Ongoing mode (every session): the spine plans, builds, and evolves — /dummyindex-plan turns a feature request into a consistency-checked proposal and auto-equips the project-tuned toolkit in .claude/ for it, /dummyindex-build drives the proposal through those equipped agents (and warns if the repo isn't equipped instead of silently falling back), and /dummyindex-remember saves cross-session memory to .context/session-memory/. /dummyindex-equip is the standalone way to (re)equip or evolve the toolkit.

Core principle: dummyindex stays the spine — it never writes production code itself. It plans, equips .context/-grounded tooling into .claude/, and orchestrates; the generated tooling + dispatched agents do the writing.


Install

User-global (one-time):

pip install --user dummyindex        # or: uv tool install dummyindex
dummyindex install                   # copies skill into ~/.claude/skills/dummyindex/

Per-repo (no global state):

cd /path/to/your/repo
dummyindex install --scope project   # writes .claude/skills/dummyindex/SKILL.md in this repo

To remove:

dummyindex uninstall                 # or: --scope project [--dir PATH]

Quickstart

Inside a Claude Code session opened in your repo:

/dummyindex                          # ingest + council, install hooks (setup mode)
/dummyindex ./src                    # scope to a subdirectory
/dummyindex-plan "add rate limiting" # NL → proposal, then auto-equips the toolkit for it
/dummyindex-build                    # drive the proposal's checklist through the equipped agents
/dummyindex-equip                    # standalone: (re)equip or evolve the toolkit (plan auto-equips)
/dummyindex-remember                 # save cross-session memory

CLI — the agent's deterministic backbone (no LLM cost). The skill and council run these; you don't type them by hand. The only terminal commands a human runs are the install bootstrap above. Shown here for transparency:

dummyindex ingest .                  # build .context/ backbone + CLAUDE.md block
dummyindex context query "how does auth work"   # ranked feature shortlist
dummyindex context rebuild --changed .          # quick deterministic backbone refresh
dummyindex context reconcile .                  # what drifted since the last reconcile (commit-anchored)
dummyindex context hooks status .              # check hook health
dummyindex context --help            # full command list

Token usage (reads Claude Code transcripts, no LLM cost) — a human checks this via the /tokens slash command, which wraps dummyindex usage:

dummyindex usage                     # current chat: context window + dedup session totals
dummyindex usage daily               # per-day totals across every project (also: session|monthly|blocks)

Full command reference: docs/COMMANDS.md — every slash command and CLI command in one place.


Docs

  • docs/COMMANDS.md — every command (slash + CLI) in one page
  • docs/README.md — docs index (guide + reference + internal)
  • docs/guide/ — twelve conceptual docs: architecture, data model, CLI surface, lifecycle, retrieval model, and more

License

MIT — see LICENSE.

Ver o original no GitHub

Bom para

Relacionados