liujiarui0918
UserBatteries-included Claude Code setup: 33 skills / 22 agents / 25 commands / 11 hooks / 8 MCPs + one-click installer for Windows & macOS
Categories
Indexed Skills (54)
condition-based-waiting
Use when waiting for something async. NEVER use Start-Sleep blindly — wait on conditions.
defense-in-depth
Use when a critical invariant matters. Single-layer protection is a tech-debt smell.
document-handling
Use when reading/writing docx/pdf/pptx/xlsx files. Prefer programmatic libs over shelling out.
extracting-from-images
Use when user pastes a screenshot / image. Extract text, layout, intent — don't just describe pixels.
incremental-context-building
Use when entering a large unfamiliar codebase. Build understanding in layers, not all-at-once.
lean-context
Use to keep context lean. Read only what you need. Compress aggressively.
mcp-builder
Use when building a new MCP server. Covers stdio vs HTTP, tool schema, error handling.
prompt-engineering
Use when helping user write an LLM prompt. Cover structure, examples, output format, anti-patterns.
receiving-code-review
Use when receiving review feedback. Reason about each item — don't performatively agree.
requesting-code-review
Use BEFORE handing off / committing. Self-review checklist to run between tasks.
root-cause-tracing
Use when fixing a bug. Trace along the call chain to the source — don't stop at the symptom.
skill-creator
Use when authoring a new skill. Standardized template + quality checklist.
subagent-driven-development
Use for complex multi-task work. One fresh subagent per task + two-stage review.
using-claude-api
Use when writing or debugging code that calls Anthropic SDK. Covers Opus 4.7, prompt caching, tool use, vision, batch, thinking.
using-context7
Use when working with any library/framework/SDK to fetch current docs. Beats training-data memory which is stale.
using-cron
Use when user wants "remind me in N min", "check every hour", "run X daily", or any scheduled work. Pick session-only vs durable correctly.
using-git-worktrees
Use when starting a feature. Isolate work in a worktree to keep main clean.
using-memory
Use when working with the auto-memory system. What goes in / stays out / how to update.
using-playwright
Use when verifying UI changes, scraping web data, or automating browser flows.
using-sequential-thinking
Use for complex multi-step reasoning where you need to think out loud, revise, branch.
using-task-list
Use when deciding whether to TaskCreate. Tasks track multi-step work — don't open one for trivial single-step jobs.
verification-before-completion
Use BEFORE claiming any task complete. "Looks right" is not verification.
writing-skills
Use when creating a new skill. Meta-skill with template + anti-patterns.
condition-based-waiting
Use when waiting for something async. NEVER use Start-Sleep blindly — wait on conditions.
defense-in-depth
Use when a critical invariant matters. Single-layer protection is a tech-debt smell.
document-handling
Use when reading/writing docx/pdf/pptx/xlsx files. Prefer programmatic libs over shelling out.
extracting-from-images
Use when user pastes a screenshot / image. Extract text, layout, intent — don't just describe pixels.
incremental-context-building
Use when entering a large unfamiliar codebase. Build understanding in layers, not all-at-once.
lean-context
Use to keep context lean. Read only what you need. Compress aggressively.
mcp-builder
Use when building a new MCP server. Covers stdio vs HTTP, tool schema, error handling.
prompt-engineering
Use when helping user write an LLM prompt. Cover structure, examples, output format, anti-patterns.
receiving-code-review
Use when receiving review feedback. Reason about each item — don't performatively agree.
requesting-code-review
Use BEFORE handing off / committing. Self-review checklist to run between tasks.
root-cause-tracing
Use when fixing a bug. Trace along the call chain to the source — don't stop at the symptom.
skill-creator
Use when authoring a new skill. Standardized template + quality checklist.
subagent-driven-development
Use for complex multi-task work. One fresh subagent per task + two-stage review.
using-claude-api
Use when writing or debugging code that calls Anthropic SDK. Covers Opus 4.7, prompt caching, tool use, vision, batch, thinking.
using-context7
Use when working with any library/framework/SDK to fetch current docs. Beats training-data memory which is stale.
using-cron
Use when user wants "remind me in N min", "check every hour", "run X daily", or any scheduled work. Pick session-only vs durable correctly.
using-git-worktrees
Use when starting a feature. Isolate work in a worktree to keep main clean.
using-memory
Use when working with the auto-memory system. What goes in / stays out / how to update.
using-playwright
Use when verifying UI changes, scraping web data, or automating browser flows.
using-sequential-thinking
Use for complex multi-step reasoning where you need to think out loud, revise, branch.
using-task-list
Use when deciding whether to TaskCreate. Tasks track multi-step work — don't open one for trivial single-step jobs.
verification-before-completion
Use BEFORE claiming any task complete. "Looks right" is not verification.
writing-skills
Use when creating a new skill. Meta-skill with template + anti-patterns.
brainstorming
Use BEFORE any creative or non-trivial design work. Socratic clarification first, then 2-3 approaches, then a spec.
dispatching-parallel-agents
Use when ≥2 independent tasks exist. Dispatch all in a single message with multiple Agent calls.
executing-plans
Use to execute a written plan. Run tasks in batches with human checkpoints.
writing-plans
Use after brainstorming produces a spec. Break the spec into 2-5 minute independently-verifiable tasks.
brainstorming
Use BEFORE any creative or non-trivial design work. Socratic clarification first, then 2-3 approaches, then a spec.
dispatching-parallel-agents
Use when ≥2 independent tasks exist. Dispatch all in a single message with multiple Agent calls.
executing-plans
Use to execute a written plan. Run tasks in batches with human checkpoints.
writing-plans
Use after brainstorming produces a spec. Break the spec into 2-5 minute independently-verifiable tasks.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.