Skip to content
Agentmart
Submit a tool

graph-system-stability

Use before changing Exograph graph domain types, snapshot construction, relationship extraction, graph queries, Note Root Formats, Workspace Ontology…

Skill by k3nnethfrancis

IndexedApache-2.0★ 5
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/k3nnethfrancis-graph-system-stability

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use before changing Exograph graph domain types, snapshot construction, relationship extraction, graph queries, Note Root Formats, Workspace Ontology interpretation, graph integrity checks, layout, scene logic, WebGPU/Canvas rendering, Graph Pane integration, or graph performance tests. Preserves Markdown ownership, open ontology, evidence, renderer independence, editor latency, deterministic layout, and fallback behavior.

Use cases: documents, design, productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/graph-system-stability

What's inside

Skill name
graph-system-stability
Description
Use before changing Exograph graph domain types, snapshot construction, relationship extraction, graph queries, Note Root Formats, Workspace Ontology interpretation, graph integrity checks, layout, scene logic, WebGPU/Canvas rendering, Graph Pane integration, or graph performance tests. Preserves Markdown ownership, open ontology, evidence, renderer independence, editor latency, deterministic layout, and fallback behavior.
License
Apache-2.0
Instructions
207 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Graph System Stability

Protect Exograph's graph architecture while changing its knowledge model or spatial
view. Apply only the branch of this workflow relevant to the task.

## Required orientation

Read these current sources first:

- `AGENTS.md` and `docs/glossary.md`
- `docs/architecture.md`
- `docs/knowledge-graph.md`
- `docs/note-root-formats.md` and `docs/workspace-ontology.md`
- `evals/graph/README.md`

For graph meaning or projection also read the affected owners:

- `packages/core/src/knowledge-graph.ts`
- `packages/core/src/workspace-graph.ts`
- `packages/core/src/graph-projection.ts`

For layout, scene, rendering, gestures, or labels read only the affected owners:

- `apps/desktop/src/renderer/src/spatialGraphRuntime.ts`
- `apps/desktop/src/renderer/src/graphSceneFoundation.ts`
- `apps/desktop/src/renderer/src/graphLayoutSimulation.ts`
- `apps/desktop/src/renderer/src/graphInteraction.ts`
- `apps/desktop/src/renderer/src/graphPresentation.ts`
- `apps/desktop/src/renderer/src/graphRendererHost.ts`
- `apps/desktop/src/renderer/src/components/SpatialGraphView.tsx`
- `apps/desktop/src/renderer/src/graphWebGpuRenderer.ts`

Keep production graph semantics in Exograph's stable contracts; do not introduce a
second graph application or graph model.

## Classify the change

Choose the narrowest affected layer:

1. **Knowledge Graph** — `WorkspaceGraph` and `KnowledgeGraphSnapshot`: Concepts,
   Properties, Relations, Evidence, resolution, origin, Format/Ontology
   interpretation, and graph queries.
2. **Graph projection** — `GraphTopology`: semantic facts compiled into numeric
   topology, visual classes, weights, and snapshot-qualified cold metadata.
3. **Layout and scene** — deterministic positions, camera, selection, paths,
   picking, focal labels, and mental-map continuity.
4. **Renderer** — WebGPU or Canvas pixels and device recovery only.
5. **Product integration** — Graph Pane, workers, IPC, persistence, editor-load
   isolation, accessibility, and packaged-app behavior.
6. **Verification** — graph contract/integrity tests or the repo-local graph
   performance suite.

Do not solve a lower-layer problem by moving ownership into a higher layer.

## Hard invariants

### Knowledge ownership

- Markdown and frontmatter remain canonical.
- Preserve unknown types, properties, and supported nested YAML values.
- Keep Concept types and Relation predicates open vocabularies.
- Record `document | ontology | inferred` origin, resolution, and inspectable Evidence for Relations.
- Keep semantic similarity, inferred types, and proposed Relations as versioned
  Derived Signals until a user accepts a Markdown change.
- Generic Markdown remains usable without a Workspace Ontology.
- Treat `ontology.yaml` as Candidate source. Only an exact reviewed Keep may
  publish Active derived state; Candidate watcher events never invalidate the
  graph by themselves.
- A Format projects base Concepts and an Ontology may interpret their meaning;
  neither becomes a canonical database, mutation authority, or reason to reject
  unknown data.

### Layer ownership

```text
Markdown → Knowledge Graph → Graph projection → Layout → Scene → Renderer
```

- `WorkspaceGraph` is the one production knowledge-graph boundary and emits
  `KnowledgeGraphSnapshot`.
- `GraphTopology` is the compact presentation projection of that snapshot, not
  a second semantic model. Do not create a parallel production graph.
- Closed numeric kinds may exist inside a compiled Graph View for performance;
  they must not become durable ontology enums.
- Scene and interaction modules own camera, selection, paths, and picking.
- `GraphPresentationCompiler` owns resolved visual classes and label policy.
- Renderers draw resolved numeric state. They do not interpret properties,
  choose relations, run pathfinding, or mutate graph meaning.
- The GPU owns reconstructible render copies only; CPU state remains sufficient
  for recovery and Canvas fallback.

### Responsiveness and stability

- Known Note/editor state paints independently of graph, layout, or index
  freshness.
- Graph extraction, Format/Ontology validation, topology compilation, layout,
  and enrichment stay in derived workers and off keystroke and Note-navigation
  critical paths.
- Input mutates camera or selection synchronously; simulation never blocks a
  gesture.
- The same topology, algorithm version, and seed produce deterministic settled
  output within declared tolerances.
- Unchanged Concepts retain their mental map across graph epochs.
- Rest means rest: no recurring animation frame or worker timer after layout and
  camera motion settle.
- The Canvas fallback preserves graph meaning and interaction state.
- Private Workspace projections never enter public fixtures or tunnels.

### Honest quality

- Keep renderer throughput, layout geometry, product interaction, and graph
  integrity as separate results.
- Never present density, modularity, orphan count, semantic similarity, stress,
  or neighborhood preservation as universal knowledge quality.
- Quality claims name their fixture, profile, sample size, hardware,
  metric definition, and evidence.
- Unsupported metrics remain unsupported; never synthesize a passing zero.

## Change workflow

From k3nnethfrancis/exograph, Apache-2.0.

More from k3nnethfrancis/exograph

Signals

Stars
5
Forks
0
Open issues
48
Last commit (this folder)
Jul 30, 2026 (3 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from k3nnethfrancis

Cite this page

graph-system-stability skill. Agent Mart. https://agentmart.dev/skills/k3nnethfrancis-graph-system-stability (last updated 2026-10-24)