Skills

Skills

Skills are on-demand workflows invoked with /skill-name in a Claude Code or OpenCode session. They load specialized instructions and context for specific tasks.

How to Use

Type the skill name with a / prefix in your AI session:

/tdd
/debugging-protocol
/git

The skill's instructions load into the agent's context, providing a structured workflow for the task.

Engineering Skills

Skill	Trigger	Description
/tdd	test-first, red-green-refactor	RED-GREEN-REFACTOR via vertical slices: one failing test, one implementation, repeat
/code-review	review this, check quality	Systematic review: correctness, safety, performance, clarity, consistency
/debugging-protocol	bug, broken, unexpected	Build feedback loop first, then reproduce, hypothesize, and fix
/refactoring-protocol	refactor, clean up	Green tests first, one extraction at a time, structure OR behavior never both
/improve-codebase-architecture	architecture, coupling	Find deepening opportunities where interface is nearly as complex as implementation
/zoom-out	how does this fit	Higher-level map of unfamiliar code section
/setup-project-context	set up context, glossary	Bootstrap a CONTEXT.md domain glossary for the project
/grill-me	stress-test, grill	Interview relentlessly about plan/design until every branch is resolved
/grill-with-docs	challenge plan against docs	Sharpen terminology against domain model, update CONTEXT.md inline
/prototype	prototype, throwaway	Throwaway prototypes: terminal app for logic or UI variations for visual design

Domain Skills

Skill	Trigger	Description
/rails	rails	Models, controllers, views, testing, frontend, infrastructure. Context7-style: loads relevant reference on demand
/docker	docker	Dockerfiles, Compose, multi-stage builds, service patterns. Context7-style

Workflow Skills

Skill	Trigger	Description
/git	commit, PR	Analyze changes, chunk into logical commits, conventional commit messages
/obsidian-memory	save, remember	Save/retrieve agent memory via agent-notes CLI
/ingest	ingest, archive, document	Scan and archive sources (files, folders, URLs); adapts to active memory backend (Wiki vs Obsidian)
/brainstorming	brainstorm, compare options	Explore multiple approaches before committing to one
/caveman	caveman, less tokens, be brief	Ultra-compressed communication mode (cuts token usage ~75%)
/handoff	handoff, fresh session	Compact conversation into a handoff document for a fresh agent session
/to-prd	prd, requirements	Synthesize conversation into a Product Requirements Document
/to-issues	issues, vertical slices	Break a plan/PRD into vertical-slice issues (HITL/AFK classification)
/write-a-skill	write skill, create skill	Meta-skill for creating new skills with proper structure

How Skills Work

Each skill is defined as a SKILL.md file in a dedicated directory under agent_notes/data/skills/. During installation, these are symlinked to the CLI's skills directory (e.g., ~/.claude/skills/).

Skills can:

• Load domain-specific instructions and context
• Define step-by-step workflows
• Reference external documentation (Context7 pattern)
• Include shared blocks via <!-- include: block_name -->

Adding Custom Skills

Skills are plain markdown files. To add a custom skill:

1. Create a directory under agent_notes/data/skills/<skill-name>/
2. Add a SKILL.md file with the skill's instructions
3. Run agent-notes build && agent-notes regenerate

The skill will be available as /<skill-name> in your next session.