oss.sarwagya.wtf

Ontoly Agent Skills

Ontoly Agent Skills are portable SKILL.md packages that teach coding agents how to use Ontoly before searching source files.

Ontoly Agent Skills are portable SKILL.md packages that teach coding agents how to use Ontoly before searching source files.

The skill layer contains workflow only. It does not contain compiler logic, query logic, framework logic, SDK generation, or AI reasoning. Every claim about a repository must come from Ontoly artifacts, the Query Engine, or Ontoly MCP.

LLM Enhancement is mandatory every time Ontoly is used by an LLM. Official Skills declare ontoly.enhancement: "LLM Enhancement" and preserve the graph-first evidence, confidence, and fallback rules in LLM Enhancement.

Install

Prerequisites:

  • Install Ontoly.
  • Build a graph for the repository with ontoly build ..
  • Verify the runtime can list capabilities with ontoly mcp --list.
  • Use a published repository install after release, or the local --copy flow while validating a source checkout.
pnpm add -D @0xsarwagya/ontoly-cli@rc
ontoly build .
ontoly mcp --list
ontoly mcp

Then install a skill folder with an Agent Skills compatible installer. After the repository is public, a single skill can be installed with:

npx skills add 0xsarwagya/ontoly --skill architecture-review

For local release testing, use the repository path:

npx skills add /path/to/ontoly --skill architecture-review --copy

Each installed skill is self-contained. Its reference/ directory includes the standard workflow, graph, MCP, best-practice, fallback, and capability guidance.

Catalog

Start with the Skills Overview for the full skill list, source links, install commands, and capability mapping.

Release assets:

Workflow

Every official installed skill points to reference/workflow.md:

  1. Confirm the installed workflow declares LLM Enhancement.
  2. Verify .ontoly/SoftwareGraph.json.
  3. Run ontoly build . if the graph is missing.
  4. Check graph trust with ontoly coverage ..
  5. Use Ontoly MCP.
  6. Invoke the task-specific capability.
  7. Inspect files only when Ontoly cannot answer.
  8. Cite node ids, relationship types, source spans, graph hash, and confidence.

Commands

ontoly skills list
ontoly skills validate
ontoly skills doctor

ontoly skills validate writes:

validation/skills/report.md
validation/skills/report.json
validation/skills/agent-evaluation.md
validation/skills/agent-evaluation.json

Versioning

Each skill declares:

  • skill version
  • minimum Ontoly version
  • required MCP capabilities
  • mandatory enhancement type, LLM Enhancement
  • category
  • deprecation status

Compatibility is validated against both source skills and installed artifacts before release.

← NestJS SupportOpenAPI Frontend →