Skills

Skills

The extension bundles 11 behavioral skill rulesets. Skills are injected into the agent via the resources_discover event hook.

Skill List

Skill	Description
agentic-clarification	Iterative Q&A loop to resolve ambiguity, outputs a Context Brief
agentic-plan-crafting	Executable implementation plans with no placeholders
agentic-milestone-planning	Multi-agent review → milestone DAG decomposition
agentic-run-plan	Plan execution with dependency-ordered task processing
agentic-review-work	Independent post-execution verification
agentic-simplify	Parallel code review for reuse opportunities and quality
agentic-systematic-debugging	Reproduce-first, root-cause-first debugging workflow
agentic-karpathy	Behavioral guardrails for implementation agents
agentic-rob-pike	Measurement-driven performance optimization discipline
agentic-clean-ai-slop	6-pass cleanup of LLM-specific code patterns
agentic-long-run	Multi-day execution orchestration through milestones

Skill Loading

Skills are registered via resources_discover and merged with any other discovered skills (e.g., user-level skills).

⚠️ If duplicate skill names exist, the first discovered skill is kept. Extension override is not guaranteed. This is why the superpowers skill must be removed — it defines skills with the same names.

Workflow Routing

Skills are connected by the workflow system:

/clarify → Context Brief → /plan → Implementation Plan → Execute
                                     ↓
                              /ultraplan → Milestone DAG → agentic-long-run

The Complexity Assessment in the Context Brief determines routing:

• Simple (score 5-8) → agentic-plan-crafting
• Complex (score 9-15) → agentic-milestone-planning
• Borderline (score 8-9) → user chooses