Skip to main content

// skills

Skills — your team's playbook, executed by AI

Skills are reusable instruction sets that teach Claude how to handle a specific CygNet task. Triggered by slash commands. Built on the same portable Agent Skills standard Anthropic published as an open format — so a skill that solves a CygNet problem in Narya can be invoked through other Skills-aware clients too. The user-facing unit of reusable knowledge, backed by WAIS underneath.

// anatomy

What a skill is

  • YAML frontmatter — name, description (with trigger phrases), category, metadata. Loaded into the system prompt at all times — about 50 tokens per skill.
  • Markdown body — step-by-step instructions, tool call guidance, common mistakes to avoid, example outputs. Loaded on demand when the skill is invoked.
  • Optional references/ — additional documentation files loaded only if Claude needs deeper context during execution. Three-tier progressive disclosure keeps token usage tight.

// built-in

CygNet skills shipped today

These encode domain knowledge the LLM gets wrong without explicit guidance. Each one was created because a real CygNet task tripped the system enough times to deserve a permanent fix.

/hierarchy
hierarchy
Navigate facility hierarchy for a device — resolves device → facilities → walks the parent attribute chain. Created because the LLM consistently showed device hierarchy instead of facility hierarchy.
/audit-alarms
audit-alarms
Audit alarm configurations across all points linked to a device or facility. Surfaces inconsistencies before they bite in production.
/compare-config
compare-config
Compare point configuration between two facilities to find differences. The fastest way to spot a forgotten attribute on a sibling device.
/device-census
device-census
Survey all device types and counts across the CygNet site. Foundation for naming-convention reasoning and bulk operations planning.
user skills

Your team's skills, on disk

Create new skills directly from Narya Command. Edit the SKILL.md in the built-in editor. Stored in %LocalAppData%\NaryaCommand\skills\. User skills override built-in skills by name, so your team's naming conventions and patterns take precedence over our defaults.

skills + WAIS

The flywheel: use → correction → lesson → skill instruction

  • Scoped retrieval. Skill metadata scopes WAIS retrieval. When /audit-alarms activates, WAIS pulls site-specific alarm conventions and relevant lessons.
  • Execution feedback. Skill invocations feed evidence and corrections back to WAIS, tagged with the skill name.
  • Lesson promotion. When a WAIS lesson reaches sufficient reinforcement, the system suggests promoting it into a skill instruction. Manual approval keeps quality high.
  • Site-specific layers. WAIS auto-generates site-specific context blocks per skill — your built-in skills get customized by your conventions.

Read about WAIS →

// ready?

Trigger a skill in Narya Command

Type /hierarchy in chat. Claude loads the skill instructions, runs the right connector actions in the right order, and reports back. WAIS captures whatever the team learns along the way.