atlas

atlas

This skill turns a stated need into an entire system design by mining the Atlas knowledge graph. It is the brain of the atlas plugin. For non-trivial runs it delegates orchestration to babysitter:babysit using an atlas-specific .a5c process; for simple lookups it queries the graph directly.

1. What Atlas is

Atlas is a knowledge graph of agents, processes, data models, capabilities, workflows, clusters, and wiki pages. You reach it through the mcp__atlas__atlas_public_* MCP tools, which this plugin wires natively into your harness (see the plugin README "Atlas MCP" section; the server URL is overridable via ATLAS_MCP_URL). Never assume the graph's shape — discover it through the tools.

2. When to use

Trigger phrase	Command
design a system, blueprint from a need, graph-driven design	/atlas:discover
process mining, what processes does X need	/atlas:mine-processes
data mining, what data does X need	/atlas:mine-data
collect nuances, edge cases/constraints for X	/atlas:collect-nuances

3. The need → design pipeline (core method)

1. Frame the need — restate the goal as a domain + concrete outcomes. If the request is ambiguous, run a short interview (AskUserQuestion). Per repo policy, interview ONLY when requirements are genuinely unclear.
2. Locate anchors — atlas_public_search(q=<need terms>) to find seed nodes; atlas_public_clusters / atlas_public_kinds to scope the domain.
3. Expand the graph — atlas_public_neighbors(id, depth, edges, kinds) from the anchors to gather the system's parts: components, workflows, processes, data models, capabilities.
4. Read detail — atlas_public_record(id, expandNeighbors) for fields and edges; atlas_public_wiki_page for narrative context.
5. Synthesize the design — assemble the discovered nodes into a layered system design (components, processes, data, integrations, nuances) and write a design artifact under .a5c/atlas/<run>/design.md plus a machine mirror.
6. Converge (TDD) — each phase asserts its own checkable outputs before proceeding (see the atlas processes), iterating until the assertions pass.

4. How to delegate

For any non-trivial run, hand off to babysitter:babysit (via the Skill tool) naming the matching atlas process:

• /atlas:discover → atlas-systems-discovery
• /atlas:mine-processes → atlas-process-mining
• /atlas:mine-data → atlas-data-mining
• /atlas:collect-nuances → atlas-collect-nuances

Do not hand-roll orchestration when a process exists.

5. Guardrails

• No fallbacks (repo rule). If you find yourself writing a fallback, stop and fix the root cause.
• Never invent graph node ids — only reference ids returned by Atlas tools.
• Keep breakpoints sparse; use them only when user input is genuinely critical or ambiguous.
• Do not emit kind: 'shell' subtasks unless the user explicitly asks for a shell-oriented workflow.