Reverse-Claude: The Self-Configuring Runtime Intelligence Skill

This skill is Claude Code's mirror. It encodes how Claude Code works internally -- its permission cascade, plugin system, agent orchestrator, context compactor, and configuration surface -- extracted from ~13,000 lines of production source. Point this skill at Claude Code itself to understand, configure, and extend it.

When to Use

• Configuring Claude Code — set up permissions, hooks, CLAUDE.md rules, settings
• Building Claude Code-like systems — command safety, plugin lifecycle, agent orchestration
• Managing LLM context — compaction hierarchies, retry strategies, circuit breakers
• Understanding internals — how permissions cascade, how agents spawn, how compaction works
• Extending Claude Code — writing plugins, skills, hooks, MCP integrations

Module 1: Permissioned Invocation & Runtime Safety

Invariant

No command executes without passing through a multi-layered permission cascade that enforces deny > ask > allow > read-only > passthrough priority, with defense-in-depth against shell injection, env-var bypass, wrapper bypass, and parser differentials.