← Hub

Skill Shelf

Top-rated AI skills & best practices — one category at a time

> Roundup first, then the deep dives

Skills, plugins, rules and MCP servers that actually exist and are actually used, plus the best practices distilled by the vendors themselves. Each category opens with a roundup — what is out there, who ranks where, by what test, and when not to install any of it — followed by two to four deep dives into the entries worth the trouble.

House rules: nothing gets named unless it can be linked · every number carries its source and month · every issue states the honest cost · the takeaway is a test you can reuse, not a shopping list.

← Super Individual Daily (Day 1–59 archive)
Coding & Codebase— C1
S01Coding Skills RoundupTen you can install today · Install commands · When not to · Three-minute inspection S02Superpowers, Taken ApartThe enforced pipeline · Composable auto-triggering · Cross-harness · Overhead on small tasks S03Code Review SkillsWhy review is the sweet spot · Taming false positives · Division of labor with CI gates S04Git Workflow SkillsCommit / PR / changelog · The diff as sole input · When a hook beats a skill
Meta — Building Skills— C2
S05Meta-Skills RoundupTools that build tools · The official spec and template · Why evals come before docs S06Official Authoring Best PracticesHow name/description trigger · Three disclosure patterns · Degrees of freedom · Where 500 lines comes from S07Writing Evals for a SkillBaseline first · Start with three scenarios · The two-instance loop · Telling a real gain from noise S08AGENTS.md & PortabilityOne instruction file, many agents · Its relation to CLAUDE.md and Cursor rules · What belongs at repo level
MCP & Integrations— C3
S09MCP Server RoundupRanked by stars and real usage · Four main lines · The context tax per server S10Docs-Retrieval MCPCuring the training-cutoff problem · When pasting the doc is simply better S11Browser Automation MCPAccessibility tree vs screenshots · Why structured snapshots win · Testing vs scraping S12The Security Edge of MCPTool output is an injection surface · Least privilege and confirmation gates · Name collisions
Docs & Deliverables— C4
S13Office Deliverable Skills RoundupThe docx/pdf/pptx/xlsx set · Why this class pays off fastest · The shared shape of template skills S14How Doc Skills Avoid CorruptionUnpack-edit-validate-repack · The validator as feedback loop · Why deterministic steps belong in scripts S15Brand & Style SkillsTurning a style guide into enforceable constraints · The hard part is acceptance criteria
Writing & Content— C5
S16Writing Skills RoundupDe-slopping / drafting / summarizing / SEO · Why opinions diverge most here S17De-AI-ing Your ProseWhich constructions it actually deletes · Encoding your own voice · The detector arms race S18Examples Over RulesThree pairs beat ten rules · How to pick examples and how many
Research & Information— C6
S19Research Skills RoundupDeep research / competitive intel / evidence chains · Strong ones leave an audit trail S20Research as a Workflow SkillQuestion → method → sources → evidence → synthesis → caveats · Citation checks as the loop S21Source Grounding & CitationsWhat models almost always fabricate · How tool grounding changes it · Aggregators are not sources
Data & Analysis— C7
S22Data Skills RoundupNL→SQL / charts / anomalies / reports · Why private context matters most here S23Domain-Split Reference StructureOne SKILL.md over many domain files · Where schemas and definitions live S24Plan-Validate-ExecuteA machine-checkable plan before execution · The default for batch and destructive work
Agent Orchestration & Automation— C8
S25Orchestration RoundupWhere subagents / hooks / slash commands / plugins each sit · When one skill suffices S26Subagents vs SkillsContext isolation vs instruction injection · Fan-out payoff and coordination tax S27Hooks & Deterministic AutomationDon't leave to the model what code can guarantee · Pinning down the every-time steps S28Building Effective Agents, Read CloselyWhere workflow ends and agent begins · Five base patterns · Don't reach for an agent
Safety, Review & Governance— C9
S29Safety & Governance RoundupSecret scanning / dependency audit / privilege checks / output validation · Safety must fail closed S30How Risky Is Installing a SkillA skill is executable instruction plus scripts · An intake checklist · Marketplace trust tiers S31Truth-First SkillsMaking the model say I don't know · Uncertainty in the output contract · How to accept-test it
Personal Productivity & PKM— C10
S32Personal Productivity RoundupMeeting notes / inbox / notes / calendar · Why this class should be homemade S33Growing a Skill From RepetitionNotice what you re-explain every time · The shortest path from one real task to one skill S34Skill Library HygieneConsistent naming · dedup · deprecation · retire by usage · when to fold it back into a prompt
Monthly Frontier Refresh— Supplement
S35Agent Plugins 1.0.0A vendor-neutral packaging spec from five vendors · plugin.json + skills/ + mcp.json · When not to package