GigiKit Guides

Activating & Using Skills

Skills activate via slash commands in a Claude Code session or by name in agent delegation prompts. This guide covers the activation syntax, available flags, and common patterns for combining skills in a workflow.

Basic Activation

Type a skill’s slash command in any Claude Code session:

/gk:cook "Add pagination to the users list endpoint"
/gk:plan "Design a caching layer for API responses"
/gk:test "Run integration tests for the payments module"

Claude loads the skill’s SKILL.md, reads the instructions, and executes the defined workflow. If a skill requires clarification, it presents an AskUserQuestion prompt rather than guessing.

Passing Flags

Core skills like /gk:cook and /gk:plan accept flags that change their operating mode:

Cook Flags

# Full interactive workflow with human review gates (default)
/gk:cook "Add search functionality" --interactive

# Skip research phase, go straight to scout → plan → code
/gk:cook "Add search functionality" --fast

# Run implementation phases in parallel across multiple agents
/gk:cook "Add search functionality" --parallel

# Auto-approve all review gates (no human pauses)
/gk:cook "Add search functionality" --auto

# Skip the testing phase entirely
/gk:cook "Add search functionality" --no-test

# Execute an existing plan file directly
/gk:cook plans/260305-0912-search-feature/plan.md

Plan Flags

# Auto-detect complexity and pick the right planning mode (default)
/gk:plan "Add OAuth login" --auto

# Fast planning — skip research, go straight to architecture
/gk:plan "Add OAuth login" --fast

# Hard mode — 2 researchers, red-team review, thorough validation
/gk:plan "Add OAuth login" --hard

# Generate two alternative approaches for comparison
/gk:plan "Add OAuth login" --two

Smart Intent Detection

/gk:cook detects intent from natural language — you often don’t need explicit flags:

What You TypeDetected Mode
/gk:cook "quick fix for the login bug"fast
/gk:cook "trust me, just add the endpoint"auto
/gk:cook "add auth, payments, and notifications in parallel"parallel
/gk:cook "add user profile page no test"no-test
/gk:cook plans/260305-0912-feature/plan.mdcode (execute plan)

Chaining Skills in a Workflow

Skills are designed to hand off to each other. The most common production chain is:

# 1. Research the problem space
/gk:research "JWT authentication patterns in Express.js"

# 2. Plan the implementation
/gk:plan "Add JWT auth based on research findings"

# 3. Execute the plan
/gk:cook plans/260305-0912-jwt-auth/plan.md

# 4. Verify and review (cook handles this automatically,
#    but you can invoke manually for ad-hoc work)
/gk:test "Run auth test suite"
/gk:code-review

In practice, /gk:cook orchestrates this entire chain automatically. You invoke the sub-skills manually only for focused, one-off tasks.

Common Skill Combinations

Debugging a Failing CI Build

/gk:debug "Auth tests failing in CI but passing locally"
/gk:test "Run full test suite and compare with CI output"

Explaining Complex Code

/gk:preview --explain "how the JWT refresh token rotation works"
/gk:preview --diagram "authentication flow from login to protected route"

UI Work

/gk:ui-ux-pro-max "Design the user profile settings page"
/gk:test ui http://localhost:3000/profile

Database Design

/gk:databases "Design schema for multi-tenant SaaS with row-level security"

Using Skills in Agent Prompts

When the orchestrator delegates to a sub-agent, it references skills by name in the prompt:

Spawn tester sub-agent:

"Use the /gk:test skill to run the full test suite for the auth module.
Activate /gk:debug if any tests fail and root cause is unclear.
Activate /gk:chrome-devtools for visual regression on the login page.

Work context: /Users/dev/my-app
Reports: /Users/dev/my-app/plans/reports/"

Next Steps