Guide safe, incremental refactoring that improves code quality without changing behavior.

Auto-invokes for every Claude turn during active T4 task execution. Emits append-only trace events (file_read, output_claim, tool_call, decision) to .claude/plans/traces/{TASK_ID}.jsonl. Source of grounding-discipline checks. Disable: AF_TRACE_DISABLED=true.

React Native gotchas and decision criteria. Covers FlatList optimization, storage hierarchy, and platform-specific pitfalls.

React gotchas and decision criteria covering stale closure bugs in useEffect, conditional rendering traps (falsy 0/empty string), state management selection (useState vs Context vs Zustand vs TanStack Query), and hooks rules. Use when writing React components, debugging re-renders, choosing state management, or encountering useEffect infinite loops.

Python gotchas and decision criteria. Covers async pitfalls, FastAPI/Django patterns, and type hint traps.

Evaluate how you use Claude Code — analyze prompt patterns, feature utilization, and get improvement suggestions. Trigger: /prompts:evaluate, prompt analysis, usage evaluation, how am I using Claude

Load project conventions and generate session context. Use EXPLICITLY when needed, not automatically.

5 techniques for different problem types. Use when stuck or facing complex challenges.

On-demand wrapper over scripts/preflight/run-all.sh. Runs Tier 1 bash linters against a file, command, or arbitrary content. Returns 0 pass / 1 warn / 2 fail. Used by /aura-frog:preflight check and contributors authoring custom hooks.

On-demand plan-tree validator. Runs all 8 invariants from spec §6.7 (parent existence, children integrity, no orphans, valid status, monotonic revision, test_ref existence, DAG no-cycles, freeze_reason). Refuses commits on violations.

Route hierarchical-planning intents to the correct backing script. Use when the user invokes /aura-frog:plan (with or without subcommand), mentions plan verbs (expand/next/replan/promote/archive/freeze/thaw/undo/status/conflicts), or types a plan-vocabulary bare word with .claude/plans/active.json present. Owns verb table, intent classifier, and 3-stage routing pipeline.

Loads minimum hierarchical plan context (.claude/plans/) for current execution focus. Auto-invokes every session when plans exist. Stays under 800 always-loaded tokens regardless of plan tree size.

Compresses a completed plan-tree branch into .claude/plans/archive/{NODE_ID}.summary.md. Removes the original node files (preserved by checkpoint history). Reduces always-loaded surface area as features ship.

Ultra-compact Phase 1 requirements output. HARD CAP: 500 tokens.

Loads permanent_memory.md summary lines (≤120 always-loaded tokens, hard cap 200). Surfaces durable wisdom from past Epics into the current session. Silent if .claude/memory/ does not exist.

Identifies and resolves performance bottlenecks across frontend (Core Web Vitals, code splitting, lazy loading), backend (N+1 queries, async processing), and database (EXPLAIN ANALYZE, indexing strategies) layers. Use when the user reports slow code, latency issues, memory leaks, needs to speed up an application, or wants to benchmark and profile performance.

Systematic performance profiling and optimization. Use when performance issues are reported or suspected. Measure first, optimize second. Applies Pareto principle — find the 20% of code causing 80% of slowness, fix that, not the rest.

Node.js gotchas and decision criteria. Covers async pitfalls, Express/NestJS patterns, and common mistakes.

Next.js 14+ gotchas and decision criteria. Covers server/client boundary, caching strategy, and data fetching patterns Claude commonly gets wrong.

Detect and navigate monorepos correctly. Use when working with repos containing multiple packages (pnpm/yarn/npm workspaces, Turborepo, Nx, Lerna). Ensures commands run in correct package scope, dependencies are routed properly, and cross-package changes are coordinated.

Guide safe database and code migrations with zero-downtime strategies.

On-demand audit of MCP usage. Reads .aura/security/mcp-audit.jsonl, surfaces blocked calls, rate-limit hits, suspicious input patterns. Companion to mcp-call-gate hook (which produces the audit log).

Analyze collected learning data from Supabase to identify success patterns, failure patterns, optimization opportunities, and agent performance trends.

Load agent definitions on-demand to reduce context usage. Only loads full agent when needed.