Introducing skillc: A Development Kit for Agent Skills | govctl Blog

We’re excited to announce skillc v0.1.0 — the development kit for Agent Skills, the open format for extending AI agent capabilities with specialized knowledge and workflows.

What is skillc?

Agent Skills define what a skill is: a folder with a SKILL.md file containing metadata and instructions. skillc provides the how: tooling to create, validate, build, and trace skills throughout their lifecycle.

Where the Agent Skills specification ends, skillc begins.

Installation

cargo install skillc

This installs the skc binary.

Two Workflows

skillc serves two distinct user groups:

For Skill Authors

Create, validate, and test skills locally before publishing with confidence:

skc init my-skill           # Create a new skill
# ... edit SKILL.md ...
skc lint my-skill           # Validate quality
skc build my-skill          # Build for local testing
skc stats my-skill          # See how agents use your skill
git push origin main        # Publish to GitHub

For Power Users

Compile any skill to unlock full-text search and usage analytics:

npx skills add user/skill   # Install a skill
skc build user-skill        # Enable search + analytics
skc search user-skill "api" # Full-text search
skc stats user-skill        # Usage insights

Note: Building is optional for consumers. Published skills work without compilation — building just enables search and analytics.

Key Features

Scaffolding

Create new skills with proper structure from the start:

skc init my-skill           # Project-local skill
skc init my-skill --global  # Global skill

skillc creates the directory with a valid SKILL.md template including required frontmatter.

Linting

17 lint rules validate skill quality before publishing:

SKL1xx  Frontmatter validation (name, description, format)
SKL2xx  Structure rules (size, headings, hierarchy)
SKL3xx  Link validation (file exists, anchors, no escapes)
SKL4xx  File organization (orphan detection)

Example output:

$ skc lint my-skill
error[E300]: SKL102 name-format: name 'My-Skill' contains uppercase characters
warning[W300]: SKL104 name-match-dir: name 'foo' does not match directory 'bar'

Compilation with Multi-Target Deploy

Build skills and deploy to specific agents:

skc build my-skill                  # Deploy to Claude (default)
skc build my-skill --target cursor  # Deploy to Cursor
skc build my-skill --target all     # Deploy to all configured targets

The compiler:

Creates a stub — a compact SKILL.md with gateway instructions
Builds a search index for full-text queries
Deploys to the specified agent’s skill directory via symlink

Full-Text Search

FTS5-indexed search over skill content:

skc search my-skill "borrow checker"
skc search my-skill "config" --limit 5
skc search my-skill "api" --format json

Usage Analytics

Track which sections agents actually read:

skc stats my-skill                    # Summary
skc stats my-skill --group-by sections # Most accessed sections
skc stats my-skill --group-by files    # Most accessed files
skc stats my-skill --group-by search   # Most frequent search terms

This data enables data-driven skill refinement. If agents never read your “Advanced Configuration” section, maybe it should be restructured.

MCP Integration

All read commands are exposed as MCP tools for direct agent integration:

skc mcp  # Start the MCP server

Available tools: skc_outline, skc_show, skc_open, skc_sources, skc_search, skc_stats, skc_build, skc_init, skc_lint.

Add to your agent’s MCP configuration to enable seamless skill access.

Storage Layout

.skillc/skills/     Project-local skills
~/.skillc/skills/   Global source store
~/.claude/skills/   Claude runtime (deployed)
~/.cursor/skills/   Cursor runtime (deployed)

Command Reference

Category	Command	Description
Author	`skc init`	Create new skill or project
	`skc lint`	Validate structure and quality
	`skc build`	Compile and deploy locally
	`skc list`	List all managed skills
Read	`skc outline`	List all sections
	`skc show`	Show section content
	`skc open`	Read file contents
	`skc search`	Full-text search
	`skc sources`	List source files
Analytics	`skc stats`	Usage analytics
	`skc sync`	Merge local logs
Agent	`skc mcp`	Start MCP server

Governed by govctl

skillc development follows strict phase discipline via govctl. The project has 10 normative RFCs:

RFC-0000 — Vision and compatibility guarantees
RFC-0001 — Skill compilation
RFC-0002 — Gateway commands (outline, show, open)
RFC-0003 — Usage analytics
RFC-0004 — Search protocol
RFC-0005 — Error code registry
RFC-0006 — Scaffolding (init)
RFC-0007 — CLI reference
RFC-0008 — Skill authoring standard (lint rules)
RFC-0009 — Configuration

Design Principles

Extend, don’t reinvent — Agent Skills is an open standard. skillc adds tooling; it doesn’t fork the format.
Observability unlocks iteration — You can’t improve what you can’t measure. Instrumented access is the foundation.
Progressive disclosure by default — The spec recommends it; skillc enforces it. Stubs expose structure without content.
Local-first, zero infrastructure — Runs on your machine. Skills are files. No accounts, no network, no lock-in.

What’s Next

v0.1.0 establishes the core workflow: init → lint → build → stats. Future versions will add:

Configurable lint rules via .skillc/config.toml
SARIF output for IDE integration
Additional lint rules (code block languages, heading hierarchy)
Inline directives to suppress warnings

skillc is open source under MIT/Apache-2.0. Check out the GitHub repository to get started, and visit agentskills.io for the full Agent Skills specification.