Skills

Skills

Skills extend agent behavior for cross-cutting concerns. They apply regardless of stack or project type.

Always Active

code-simplifier

Applied to every piece of code written or reviewed.

Rule	Description
No speculative abstractions	Solve today's problem only
Functions max 30 lines	One thing per function
Flat over nested	Guard clauses and early returns
Descriptive names	No Base, Abstract, Manager, Handler
Comments explain why	Never what

critical-review

A 4-phase adversarial protocol during implementation:

1. Pre-Flight — Identify happy path, failure points, boundary conditions
2. Adversarial Review — RESILIENCE (silent failures), EFFICIENCY (code smells), FRICTION (happy-path friction)
3. Self-Correction — Each weakness refactored and re-verified in isolation
4. [VALIDATION] — Edge cases covered, justification, residual risks

verbosity

Controls response length per turn. Set at install with --verbosity, stored in ~/.claude/memory/verbosity.md. Re-injected every turn by the verbosity-remind hook so it never drifts.

Level	Behavior
MIN (default)	One sentence. [CHANGES] tag with file list.
INFO	Up to 5 bullets. [CHANGES] + [REASON].
VERBOSE	Full explanation, prose allowed. All tags.

memory-first

Before reading any file, walk this chain and stop at the first hit:

1. Memory — project.md + personal.md
2. Grep / Glob — pattern search
3. Targeted read — last resort, always offset + limit, max 150 lines

agent-delegation

Keeps the main context clean:

• Sub-agents handle exploration and parallel work
• Return summaries of max 200 words
• Raw file contents never enter the main context

Retired

ui-ux-pro-max

Earlier releases described this as installed from GitHub for frontend stacks. No shipped code does either. Filed as [FEAT-037] for replacement.

Skill Files

Each skill lives in skills/<name>/SKILL.md with its full ruleset and activation conditions.