The pro toolkit: Claude Code, MCP, skills & plugins
Week 2 ne aapko us shaks ka badla diya jo AI IDE use karta tha, ab aik shaks ban gaye jo ek professional coding agent command karta hai. Chaliye is toolkit ko wapis se jodte hain taake aap dekh sakein ke pieces kaise ek saath fit hote hain — aur finale ke liye extensions set up kar sakein.
Yahan sab kuch aapne pehle hi ek baar kiya hai. Ab ki taraf fluenct hona hai — tool ko bina dekhe pehchanana, aur har ek ki wazahat samajhana.
01Claude Code CLI
Claude Code aapke terminal mein rehta hai aur poora project ko apna workspace maanta hai. Woh commands jo aap rozana use karte hain:
| Command | Kya karta hai |
|---|---|
/login | Aapka account authenticate karta hai. |
/init | Project ko scan karta hai aur shuruaati CLAUDE.md draft karta hai. |
/context | Dekhta hai ke context window kitni bhar gayi hai. |
/compact · /clear | Context bachata hai, ya naye task ke liye use mita deta hai. |
/agents · /hooks · /plugin | Woh pro menus hain jinhein finale mein kholein gaye hain. |
.claude folderShayad har pro feature wapas is folder tak wapas jati hai: .claude/ aapke project mein. Commands, skills, agents, hooks aur settings saare yahan rehte hain — aur is liye ke isay Git mein commit kiya ja sakta hai, aapki poori team inhe apnati hai. Isay saath rakhein; yeh Class 12 ka backbone hai.
02Models & providers
Do alag-alag faislay, jinke aapas mein confuse hona aasan hai: kaun sa tool (Claude Code, Cursor, OpenCode...) aur kaun sa model isay chalata hai. Aur aap kabhi bhi phansay nahi hain — ek provider key usi tool ko saste ya free models par point kar deti hai.
Sab se behtareen quality
Opus-class models khususi taur par mushkil reasoning aur bari refactors ke liye. Slow, mehenga, lekin jaroorat padne par wajihat rakhta hai.
Speed + smarts
Cerebras / fast providers ke sath ghadiyon mein jawab — tight iteration loops ke liye behtareen.
Saste experiments
OpenRouter, AMP credits, GLM, ya apni machine par chalne wala Ollama — practice ke liye bilkul perfect.
Model rankings har mahine badal jate hain — last quarter ke favourite par bharosa na karein, balki kisi source jaise artificialanalysis.ai ko check karein. Skill ek winner ko yaad rakhne ki bajaye,ζζε°Pick karna hai.
03Sessions, Git & Ralph loops
Safety net ke bina autonomy sirf risk hai. Week 2 ne aapko yeh net diya:
Sessions & checkpointsrewind
Project ki history mein piche aur aage badhein; kisi buri direction ko kharab karne ke bina sab kuch khoeyen.
Git as the real safety netcommit often
Risky moves se pehle commit karein. Safar commit ek save point hai jis par aap hamesha wapas wapas kar sakte hain.
YOLO mode--dangerously-skip-permissions
Agent ko bina puchhe kaam karne dein — tez, lekin sirf Git safety net par.
Ralph loopsautonomy
Agent ke gird loop lagayein taake woh ghante bhar build karta rahe. Aapka pehla taste of hands-off autonomy — aur hooks aur swarms ka beej week 3 mein.
04Big three: MCP, skills & plugins
Class 8 ka inflection point. Teenon agent ko extend karte hain, aur teenon ek hi cheez par mabni hain: tools.
05Kaun sa, kab?
MCP jab aapko ek external system se connect karna ho (aik Jira board, GitHub repo, database). Skill jab aap agent ko kuch seekhna chahate hain (Cerebras call karna, aapke house style follow karna). Plugin jab aap poora setup ko bundle aur share karna chahate hain apni team ke sath. Woh sabse hafif cheez chunain jo masla hal kare — aksar ek skill.
✓ Aham baatein
- Claude Code ek terminal agent hai; rozmarrah ki commands aur
.claude/folder unki core hain. - Tool aur model alag-chunain; ek provider key aapko saste ya free models par point kar deti hai.
- Sessions, checkpoints aur Git woh safety net hain jo YOLO aur Ralph loops ko sanakhte hain.
- MCP connect karta hai, skills teach karte hain, plugins bundle karte hain — sab tools par mabni hain.
- Pehle sabse hafif option chunain. Agle: autonomous workflow aur master cheat-sheet.