Claude Core

Delegates work to bounded subagents and synthesizes their evidence. Use when asked to delegate, fan out, parallelize, or spawn subagents, or when a task splits into 2-5 independent subtasks such as multi-area code search or parallel review. You stay the planner and read every result.

Finds algorithmic complexity and performance hotspots, then optimizes them behind tests without changing behavior. Use for nested loops, repeated scans, N+1 queries, costly re-renders, or when a request mentions hot paths or slow code. Defaults to an analysis-only report.

Fetches, fixes, and resolves unresolved GitHub PR review threads with the gh CLI. Use when addressing reviewer feedback, clearing blocking conversations before merge, or applying GitHub suggested changes. Verifies each finding against current code and resolves only what it fixed.

Author, validate, and install custom Claude Code subagents as .claude/agents Markdown files. Use when the user asks to create a subagent, custom agent, or .claude/agents role; define a reusable reviewer, explorer, planner, or worker persona; set an agent's model or tool allowlist; or turn a recurring workflow into a delegatable role. Produces Markdown agent files with YAML frontmatter (name, description, model, tools) plus a focused system-prompt body, scoped to the smallest tool surface that does the job (read-only by default). Covers naming, description wording for autonomous delegation, model policy (Opus for reasoning/review, inherit for scans), tool scoping, project vs user scope, and a smoke test. Pairs with fan-out, which delegates to these roles. Not for one-off ad-hoc delegation (use fan-out) or authoring Agent Skills (SKILL.md files).