NetFabric/NetFabric.Numerics / .agents/skills/agent-graph-orchestration
agent-graph-orchestration
VerifiedDesign and scaffold multi-agent squads/swarms using the graph orchestration pattern (nodes = agents, edges = dependencies/data flow). USE FOR: designing a coordination topology for a multi-agent squad (sequential pipeline, parallel fan-out/aggregation, branching, feedback loop, cyclic); building an orchestrator agent that delegates to specialist agents; setting up an adversarial/tribunal code or document review with independent reviewers; scaffolding a squad's agent files for a target coding-agent harness (e.g. Copilot CLI or Claude Code custom agents). DO NOT USE FOR: authoring a single agent's frontmatter/prompt body (use copilot-cli-custom-agents, claude-code-custom-agents, or the target harness's own skill); SKILL.md authoring (use create-skill); ad-hoc multi-agent chat with no explicit topology.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.8 KB |
| references/graph-concepts.md | markdown | 2.5 KB |
| references/harness-implementation.md | markdown | 7.7 KB |
| references/model-selection.md | markdown | 6.3 KB |
| references/scaffolding-workflow.md | markdown | 16.0 KB |
| references/topologies.md | markdown | 9.8 KB |
references/graph-concepts.md
View on GitHub# Graph Concepts for Multi-Agent Systems
Grounded in "Graphs Meet AI Agents" (arXiv:2506.18019): graphs organize agent planning, execution, memory, and coordination by representing entities as nodes and their relationships as edges — structurization that lets agents reason over explicit dependencies instead of unstructured text.
## Task Dependency Graph (TDG)
Used at design time to decompose a complex task into subtasks with dependencies.
| Element | Meaning |
| --- | --- |
| Node | A subtask decomposed from the overall task |
| Edge | A dependency — one subtask's output feeds another's input |
| Shape | Typically a DAG, so no subtask waits on its own (indirect) result |
A TDG is the blueprint for graph topology: turn each subtask node into an agent node, and each dependency edge into a graph edge.
## Agent Coordination Graph (ACG)
Used at runtime to model how multiple agents communicate once the TDG (or another topology) is chosen.
| Element | Meaning |
| --- | --- |
| Node | An agent (its features/role) |
| Edge | A communication path — who passes messages/results to whom |
| Task-specific relationship | Edges mirror the TDG's subtask dependencies (this skill's default) |
| Environment-specific relationship | Edges mirror a shared environment's structure (e.g., agents near each other in a simulated space) rather than task dependencies |
## Why graphs over ad-hoc multi-agent chat
| Capability | Without a graph | With a graph |
| --- | --- | --- |
| Planning | Implicit ordering inferred per turn | Explicit dependency order, reusable across runs |
| Execution | Tool/agent calls interleaved unpredictably | Deterministic dispatch order from edges |
| Memory | Flat conversation history | Structured per-node results addressable by node ID |
| Coordination | Every agent sees every message | Each node only receives what its edges deliver |
## DAG vs. cyclic graphs
Most TDGs are acyclic (a subtask can't depend on its own output). Cyclic graphs are valid for iterative patterns (draft → review → revise) but require an explicit exit condition and an iteration bound — an unbounded cyclic graph can run indefinitely. See [topologies.md](topologies.md#4-feedback-loop-cyclic) for the bounding mechanism.
## When topology optimization matters
The survey also covers *learning* optimal graph topology (edge weights, GNN-based topology search) for large agent populations. That is out of scope for hand-scaffolded squads of a handful of agents — pick a topology from [topologies.md](topologies.md) directly instead of trying to learn one.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |