claude-code-skill

Designs a custom Agent team for a specific project or workflow, following Claude Code's internal architecture patterns: minimum-permission agents, single-responsibility prompts, orchestrator-based routing. Use this skill whenever the user asks to "design agents for my project", "set up a multi-agent system", "create a team of Claude agents", "build an agent workflow", or wants to delegate different parts of their codebase to specialized agents. Also activate when someone has a complex project with clearly separable concerns and asks how to structure Claude's work.

Audits Claude Code project configuration for prompt cache health. Detects cache-busting patterns across CLAUDE.md files, MCP server configs, and settings.json. Produces a scored health report with ranked recommendations. Use when the user asks to reduce API costs, improve cache hit rate, optimize token spending, diagnose expensive sessions, or investigate why Claude Code sessions feel slow or pricey.

Enhanced /compact that doesn't just summarize — it extracts and persists key decisions, eliminated approaches, new patterns, and current blockers into MEMORY.md before compressing the conversation. Use this skill whenever the user says /compact, "compress context", "compact the conversation", or when the context window is getting full and you want to preserve institutional knowledge from this session. It ensures that after compression, future sessions can reconstruct the reasoning behind current state, not just the state itself.

Plans context window budget for long tasks before they start — estimates token consumption by phase, identifies when autocompact will trigger, and recommends where to place manual /compact checkpoints. Based on Claude Code's autoCompact.ts thresholds and circuit breaker logic. Use this skill when the user is about to start a large task (refactoring a big codebase, processing many files, long research session), asks "will this fit in context?", "when should I compact?", "how do I avoid hitting context limits?", or when a session has already hit warnings about context usage.

Generates a delegation rules document for multi-agent workflows, covering what to keep at orchestrator level vs. what to delegate, and how to write self-contained sub-agent prompts. Use this skill when the user asks to "design multi-agent systems", "delegate work to sub-agents", "improve agent prompt quality", "design orchestrator/worker patterns", "write better sub-agent prompts", or "avoid duplicating work in agents". Also activate when someone is building an orchestrator that spawns workers and wants to avoid re-doing work the sub-agents already handle.

Restructures a chaotic or overgrown MEMORY.md into a clean 3-layer architecture based on how Claude Code's autoDream system organizes memory: a lightweight pointer index (always loaded), topic files (loaded on demand), and an archive layer (historical context, not proactively loaded). Use this skill whenever the user says "clean up MEMORY.md", "reorganize my memory files", "MEMORY.md is getting too long", "fix my memory structure", or when you observe that MEMORY.md exceeds 200 lines, contains full paragraphs instead of pointers, or mixes index entries with topic content.

Audits and restructures CLAUDE.md and system prompt configuration for maximum prompt cache efficiency, following Claude Code's internal SYSTEM_PROMPT_DYNAMIC_BOUNDARY pattern. Use this skill when the user asks to "optimize my CLAUDE.md", "reduce token costs", "improve cache hit rate", "restructure my system prompt", or "why is my context so expensive?" Also activate when someone mentions their CLAUDE.md has grown unwieldy or when working on a project where Claude is called frequently and token costs matter.

End-of-session memory distillation — extracts key decisions, eliminated approaches, new discoveries, and current blockers from the current conversation and writes them to MEMORY.md topic files. Based on Claude Code's autoDream background consolidation service. Activate when the user says "dream", "/dream", "save session memories", "distill this session", "what should I remember from this session", or when a long productive session is ending and the user wants to preserve what was learned.