atlas
Turn a stated need into a full system design by mining the Atlas knowledge graph.
Open source Open in the app JSON README (API)
About
Turn a stated need into a full system design by mining the Atlas knowledge graph.
Details
- Kind
- Plugins
- Topic
- AI, RAG & memory
- Publisher
- a5c-ai
- Origin
- gemini
- Category
- ferramentas
- Version
- 6.0.3
- Last push
- 2026-08-10T22:01:08Z
- Repository state
- ativo
- Language
- JavaScript
- Added
- 2026-08-30 14:13:39
- Updated
- 2026-08-30 14:13:39
- Origin id
a5c-ai/atlas-gemini
README
# atlas Turn a stated need into a full system design by mining the Atlas knowledge graph. Atlas is a knowledge graph of AI-agent stacks, products, capabilities, processes, data models, and the relationships between them. This plugin makes that graph a first-class design tool inside Gemini CLI: ask for a system and Atlas mines the graph to turn the stated need into a layered, evidence-backed design. ## Installation — Gemini CLI ```bash npm install -g @a5c-ai/atlas-gemini-cli atlas-gemini-cli install --global ``` Restart Gemini CLI to pick up the installed plugin. ## What's Included - **Skills**: atlas, atlas-graph-query - **Commands**: (directory) - **MCP**: the Atlas knowledge-graph server (`atlas`), wired natively into Gemini CLI's own MCP config — no manual setup. ## Atlas MCP The plugin injects the Atlas MCP server into Gemini CLI's native MCP config format automatically. By default it points at: ``` https://atlas-staging.a5c.ai/api/mcp ``` Override the endpoint at runtime with the `ATLAS_MCP_URL` environment variable. Once the harness is running, the `mcp__atlas__atlas_public_*` tools (search, record, neighbors, kinds, clusters, stats, wiki pages) are available to the skills and commands. ## How it works - The **`atlas` skill** turns a stated need into a layered system design by mining the graph — when to use Atlas, how to anchor a search, expand neighbors, and synthesize a design. - The **`atlas-graph-query` skill** documents the MCP tool surface so any agent (sub-agents included) can query the graph without re-deriving conventions. - The **commands** (`(directory)`) run interactive, graph-driven workflows — systems discovery, process mining, data mining, and nuance collection. They drive iterative, TDD-style runs whose orchestration is delegated to the Babysitter run lifecycle, so each phase is checkable and resumable. > The commands use Babysitter purely as the orchestration runtime; the design > work itself is driven by the Atlas graph via the MCP tools above. ## Verification ```bash babysitter harness:discover --json | grep gemini ```