(all install paths will be in the comments) š. šš¤š¢š„š„ šš«ššššØš« this is the factory that builds everything else. describe what you want in plain english ā claude builds the skill, tests it, packages it. you never have to write a skill.md by hand again. drop in an SOP and it becomes a reusable skill instantly. install: /plugin install skill-creator š. šš®š©šš«š©šØš°šš«š¬ fixes the #1 failure in claude code: rushed, sloppy output. forces claude to plan first, work in isolation, write tests before code, then review its own work twice. gets you from 60% ā 80% on the first pass. fewer debugging cycles = less time, lower token costs. 150k+ stars on GitHub btw š. ššš (ššš ššš®šš ššØš§š) fixes context rot. you know when claude starts forgetting requirements halfway through a session and cutting corners? GSD spawns fresh sub-agents for each task so your main session stays clean and every task gets a full context window. also has an autonomous mode if you want to hand it a spec and walk away. š. /š«ššÆš¢šš° šš§š /š®š„šš«š š«ššÆš¢šš° already built into claude code. most people aren't using it. /review = fast structured code review after you build something. free beyond normal tokens. /ultra review = uploads your branch to a cloud sandbox, spins up parallel reviewer agents, each attacking from a different angle. bugs have to be independently verified before they hit your list. no false positives. need claude code 2.1.86+. costs $5ā20/run after 3 free ones on Pro/Max. š. ššØš§ššš±š ššØšš keeps garbage out of your context window. raw tool outputs get routed through a sandbox. a 56kb playwright snapshot becomes 299 bytes. 315kb of session output becomes 5kb total. also tracks every file edit, decision, and error in a local SQL database so when claude compacts the conversation, it doesn't lose anything. sessions that used to fall apart at 30 minutes now run 3 hours clean. š. šš„šš®ššššš¦ carries knowledge across every future session. claude normally starts from scratch every time. ClaudeMem hooks into the session lifecycle, captures what happened, compresses it into semantic summaries stored in a local SQLite db with vector search, then injects the relevant parts back in when you open a new session. also auto-generates and updates your CLAUDE.md files while you work.