Chapter 05
Skills & Connectors
Chat box se ek operating layer tak — AI ko sikhana ek baar, aur apne apps se jorna
Ye chapter CCAO-F ke Configuration aur Knowledge Management domain ke liye foundation hai
Core Idea
Zyada tar log AI ko ek vending machine ki tarah treat karte hain: har baar fresh request type karte hain, kuch bhi retain nahi hota. Ye chapter ek different model sikhata hai: Skills se AI ko ek baar sikhao, Connectors se usay apne apps tak access do, aur AI ek chat interface se ek intelligent operating layer ban jata hai.
Kitchen Wali Example
3 cheezein mil kar kaam karti hain:
Chat
Ek baari ka order, AI ko batata hai is baar kya karna hai
Skill = Recipe Card
Sikhati hai har baar wahi kaam apne tareeke se kaise karna hai
Connector = Kitchen
AI ko haath deta hai aapke real apps aur data tak pahunchne ke liye
6 Essential Terms
| Term | Matlab |
|---|---|
| Skill | Saved instructions jo AI ko sikhati hain ek task aapke specific tareeke se kaise karna hai |
| Connector | Ek safe link jo AI ko ek app tak pahunchne deta hai (files, email, messages) |
| Fire / Trigger | Jab request kisi skill se match kare to AI khud usay activate karta hai |
| Scope | Aap kitni access grant karte ho, jitni chhoti utni safe |
| Read-only | AI dekh sakta hai lekin change nahi kar sakta |
| Progressive disclosure | AI sirf short summaries loaded rakhta hai, poori instructions sirf zaroorat par kholta hai |
Part 1 · 2 Upgrades
1. Chat Box Se Operating Layer Tak
Har professional ke paas repeated friction hoti hai:
- Accountants har mahine formatting instructions dobara paste karte hain
- Doctors roz SOAP format requirements dobara explain karte hain
- Marketers har chat mein brand-voice rules dohrate hain
- Engineers har baar design-review structures restate karte hain
Ye tasks repeatable hain, ek specific tareeke se hote hain, sirf input badalta hai, yehi exactly Skills solve karti hain. Doosri friction copy-paste hai (Excel, emails, Drive folders, project trackers se), yehi Connectors solve karte hain, AI ko seedha wahan tak access de kar jahan real kaam hota hai.
2. Ek Skill Actually Kya Hai
Fundamentally simple: ek folder jismein ek text file hai jiska naam exactly SKILL.md hai. Is file mein:
- Top pe ek required name aur description
- Neeche plain-English instructions
Yehi minimum hai. Optional: example files, templates, reference documents, aur scripts (jo AI khud likhta hai, aap nahi). Key innovation progressive disclosure hai: AI sirf short description hamesha loaded rakhta hai, poori instructions sirf tab kholta hai jab request match kare. Isi wajah se dus skills install karne se bhi koi performance impact nahi hota.
3. Ek Connector Actually Kya Hai
Safe, permission-scoped access external apps aur data tak, ek open standard MCP (Model Context Protocol) par chalta hai. 3 critical facts:
- AI aapki existing permissions inherit karta hai, aap khud jo files nahi reach kar sakte, wo AI bhi nahi kar sakta
- Aap choose karte ho read-only ya read-and-write access
- Aap control karte ho kaunsa connector har conversation mein active hai
Ready-made connectors Google Drive, Gmail, Slack, Notion, Figma, Linear, Atlassian, aur bohot sab ke liye maujood hain. Kuch interactive hain, live dashboards ya design surfaces conversation ke andar render karte hain.
4. Skills Vs Connectors Vs Projects Vs Custom Instructions
| Feature | Purpose | Best For | Key Test |
|---|---|---|---|
| Skill | Reusable how-to instructions | Repeatable tasks aapke tareeke se (formatting, voice, methodology) | "Main baar baar explain karta hoon kaise" |
| Connector | External apps tak safe access | Files, email, messages, tickets mein padhna/act karna | "Main baar baar doosri app se copy-paste karta hoon" |
| Project | Hamesha-loaded files/instructions ka workspace | Standing context (client, course, book) | "Ye rules yahan har cheez pe apply hote hain" |
| Custom Instructions | Har chat pe apply hone wali preferences | Global style preferences | "Ye har jagah, hamesha chahiye" |
Key Distinction
5. Slash Commands Type Karoon Ya AI Khud Jaan Le?
Claude.ai (web aur mobile) mein, Skills khud fire hoti hain, aap plain language mein apna task describe karte ho, AI aapki request ko enabled skills ki short descriptions se match karta hai, aur matching skill khud load kar leta hai. Koi slash command ki zaroorat nahi.
Ahem Number
Override options:
- Skill ko explicitly plain English mein naam se bulao ("use my brand-voice skill")
- Doosri surfaces (Cowork, Microsoft 365 add-ins, Claude Code) mein / type karo browse aur manually pick karne ke liye
Part 2 · Jo Pehle Se Hai Wo Use Karo
6. Built-In Skills
Kai document-generation skills automatically kaam karti hain jab ek switch on ho: Settings → Capabilities → “Code execution and file creation” (“Required for skills” label ke sath). Ye engine chalati hai:
- Word: professional .docx documents
- PowerPoint: complete slide decks
- Excel: real spreadsheets with formulas
- PDF: PDFs banana aur fill karna
Bas naturally poocho: “[topic] introduce karne wala ek slide deck tayar karo, general audience ke liye” aur ek finished deck download ke liye ready milegi.
Note
7. Apni Apps Connect Karna
Roughly ek minute lagta hai:
- 1
Chat Se
"+" (neeche-left) click karo, Connectors pe hover karo, Manage connectors choose karo, phir Connectors ke sath "+" click karo directory kholne ke liye
- 2
Ya Sidebar Se
Customize → Connectors kholo
- 3
Connect Karo
Service select karo, Connect click karo, standard OAuth authorization complete karo
- Connect karne ke baad, har conversation ke liye specific connectors usi "+" menu se enable karo
- Free plans mein 1 custom connector shamil hai, ready-made directory connectors broadly available hain
- Tool access setting sirf zaroorat par tools load kar sakti hai, jab kai connectors manage karne hon
- Interactive badge dekho, jo live interfaces indicate karta hai jo chat ke andar render hote hain
Worked Example
8. Skills + Connectors Sath Mein
Pattern real data fetch karna (Connector) aur output ko aapke tareeke se shape karna (Skill) combine karta hai.
Monthly Client Close
Weekly Content Batch
9. Kaunse Problems Ko Kya Chahiye?
| Category | Examples |
|---|---|
| Sirf Skill | Brand voice, report templates, SOAP notes, review checklists |
| Sirf Connector | Ledgers, emails, tickets apps se pull karna |
| Dono | Monthly closes, weekly batches, design reviews (real data fetch + specific shape apply) |
2 Ehtiyat
Part 3 · Apna Skill Banana
10. Fastest Path: AI Se Likhwao
skill-creator skill use karo apna chahiye jo describe karne ke liye:
Use the skill-creator skill to help me build a skill. The skill prepares a monthly client financial summary for my accounting firm. Whenever I ask for a 'client summary' or 'monthly close,' it should: Format all amounts in our reporting currency with thousands separators; Group line items by expense head; Flag any payment above the tax-withholding reporting threshold; Output using my standard four-section report layout (Overview, Income, Expenses by Head, Flags & Notes). Ask me anything you need, then build it.
AI clarifying sawal poochta hai, phir ek complete, correctly formatted skill generate karta hai. Ye “code you never write” principle ki misaal hai: aap client ho jo requirements specify kar raha hai, AI commissioned artifact produce karta hai, ek skill jo ek baar commission karo aur hamesha reuse karo.
11. SKILL.md Ki Anatomy
Level 1 · Frontmatter
Sirf name + description, hamesha loaded, decide karti hai skill relevant hai ya nahi
Level 2 · Body
Actual instructions, sirf tab load hoti hain jab description match kare
Level 3 · references/, assets/, scripts/
Detailed docs, templates, aur exact-step programs, sirf zaroorat par
Rules jo upload errors se bachate hain:
- File ka naam exactly SKILL.md hona chahiye
- Folder ka naam kebab-case mein ho (client-monthly-summary, na ke Client Monthly Summary)
- Name aur description mein XML-style tags na hon (angle brackets mein kuch bhi nahi)
- Descriptions short aur specific rehni chahiye
- Skills ko "claude" ya "anthropic" naam se mat banao (reserved hain)
12. Description Field Hi Poora Game Hai
Description decide karti hai ke aapki skill kabhi fire hogi ya nahi. AI relevance decide karne ke liye aapki instructions nahi parhta, sirf description parhta hai. Ek vague description matlab skill jo kabhi trigger nahi hoti, ek sharp description matlab exact activation.
Formula
| Bad | Better |
|---|---|
| "Helps with reports" (too vague) | "Prepares a monthly client financial summary. Use when the user asks for a 'client summary', 'monthly close', or 'month-end report'." |
| "Handles patient documentation" (no trigger words) | "Converts consultation notes into SOAP-format clinical notes. Use when the user asks for a 'SOAP note', 'clinical note', or to 'write up' a consultation." |
Debugging Trick
13. Test, Phir Iterate
- 1
Describe Aur Generate Karwao
Skill describe karo, AI se pehla version banwao
- 2
Instructions Parho Aur Fix Karo
Testing se pehle obviously galat cheezein pakro (poocho: kya 2 colleagues isay parh kar same kaam karte?)
- 3
Triggering Test Karo
Wo phrases try karo jo activate honi chahiye, confirm karo load hoti hai aur unrelated requests hijack nahi karti
- 4
Output Test Karo
Kya format sahi hai, grouping sahi hai, flags sahi hain, sab sections cover hote hain?
- 5
Hard Cases Test Karo
No income wale clients, threshold pe exactly payments, messy data
- 6
Failures Wapis Le Jao
skill-creator ke paas updates ke liye
| Timing | Problem Kahan Hai |
|---|---|
| Pehli reply se hi galat | File mein hi problem hai (description bohot vague/broad, ya instruction bohot loose) |
| Shuru mein sahi, chat barhne pe kharab | Chat bohot lambi ho gayi, AI ne track lose kar diya, skill theek hai, naya chat start karo |
| Har run mein wahi mistake | Step galat jagah hai, exact steps sentences mein nahi, scripts mein honi chahiye |
| Pehle kaam karta tha, ab nahi | Skill wahi hai, isliye context badla hai (threshold, template, workflow, ya data shift hua) |
14. Skill Save Aur Share Karna
Jab skill-creator khatam kare, ek panel dikhta hai Save skill button ke sath. Click karo, skill Personal skills mein (Customize → Skills) gir jati hai, toggled on, ready, aur private. Koi zip ya upload hassle nahi.
Team ya Enterprise plans pe, specific colleagues ke sath share kar sakte ho, ya apni organization ki directory mein publish kar sakte ho. Shared skills view-only hoti hain aur khud update ho jati hain jab aap original change karte ho, isse teams standardize hoti hain.
Part 4 · Same Skill, 5 Jagah
December 2025 mein, Anthropic ne Agent Skills open standard publish kiya (agentskills.io). Adoption fast thi, OpenAI ka Codex CLI aur Google ka Gemini CLI ab wahi SKILL.md files parhte hain. Ek basic skill tools ke across travel karti hai, chahe har tool apna install path, invocation syntax, permissions, aur features us portable core ke around add karta ho.
Linux Foundation ne Agentic AI Foundation banayi, MCP (jis standard par connectors chalte hain) aur related infrastructure ko khule tareeke se govern karne ke liye, kisi ek vendor ke through nahi.
| Surface | Kis Ke Liye | Skill Install | App Connect | Apna Banana |
|---|---|---|---|---|
| Claude.ai (web/mobile) | Sab ke liye, main tool | Directory mein Install click karo ya zip upload karo | "+" → Connectors → chuno → sign in | skill-creator likhta hai, Save skill click karo |
| Cowork/OpenWork (desktop) | Knowledge workers, non-coders | Wahi directory, skills automatically dikhti hain | Wahi flow, plus computer ki files bhi reach karta hai | Describe karo, agent folder mein save kar sakta hai |
| Claude Code/OpenCode (terminal) | Code ke sath kaam karne wale log | Skill folder ko skills directory mein drop karo | Config mein ek dafa setup karo, phir plain language mein poocho | Agent se create karwao, wo files likhta hai |
Note
ChatGPT Aur Gemini Ka Kya?
Skills pe, picture cross-vendor hai, kyunke Agent Skills ek open standard hai, OpenAI (Codex CLI) aur Google (Gemini CLI), plus VS Code aur Cursor, wahi SKILL.md files parhte hain. Ek skill portable hai, ek baar likho, kai tools mein chalao.
Lekin “ek baar sikhao” consumer chat apps ke andar, har vendor ka apna, non-portable version hai:
| Vendor | Naam | Kya Hai |
|---|---|---|
| ChatGPT | Custom GPTs | Ek saved version, instructions aur knowledge files ke sath, GPT Store se distribute hoti hai, sirf ChatGPT ke andar rehti hai |
| Gemini | Gems | Saved Gemini personas, instructions aur files ke sath, sirf Google apps ke andar rehte hain |
Connectors pe, teenon ke paas equivalents hain (ChatGPT connectors/apps se, Gemini Workspace aur extensions se). MCP technology increasingly shared hai, principles har jagah same hain.
“Skills portable hain (open standard, kai tools), GPTs aur Gems vendor-specific hain (aap unhe kahin aur le nahi ja sakte). Agar sirf ek tool use karte ho, GPTs ya Gems theek hain. Multi-model strategy ke liye, Skills future-proof hain.”
Part 5 · Safely Use Karo
Wahi principle jo pichli Foundations courses mein tha: model ke andar kuch bhi check nahi karta ke koi action safe ya correct hai, aap wo check ho.
Core Principle
Malicious Skills
Ek text file (aur scripts) hidden instructions rakh sakti hai jo data leak kare ya suspicious servers se contact kare. Prompt injection aur data exfiltration dono possible hain
Over-Broad Connector Access
Connector sirf wahi reach kar sakta hai jo aap kar sakte ho, lekin careless write-access ek galat edit, galat jagah record, ya bina undo ke deleted file bana sakta hai
Safe-Use Checklist
- Trusted sources se skills install karo, built-in Anthropic skills aur official directory safe default hain
- Enable karne se pehle skill parho, SKILL.md aur koi bhi bundled files kholo, AI se poocho "ye skill parh ke batao exactly kya instruct karti hai, external servers, credentials, ya data leak se juri koi cheez flag karo"
- Connectors ko read-only se start karo, "search and summarize" ki permission pehle do, "send/create/delete" baad mein
- Sab se chhoti folder ya app tak scope karo jo task ko chahiye, sab kuch connect mat karo
- Scope ek connector ka idea hai, Skill ka nahi, connectors ke paas scope dials hote hain, skills ke paas nahi, skill parhna hi uska control hai
- Edits/moves/deletes allow karne se pehle confirm karo connector recovery, version history, aur undo kaise handle karta hai
- Team mein, shared skills ko organization ki directory se route karo, zips idhar udhar mat bhejo
Isse dar mat jao, built-in tools safe hain aur roz ka workflow low-risk hai. Point ye hai ke jaise jaise aap built-in tools se community skills install karne aur write access dene ki taraf barhte ho, aapki ehtiyat bhi capability ke sath barhni chahiye.
Poora Course, Compressed
- Chat ek baar batata hai, Skill har baar sikhati hai, Connector haath deta hai apps tak (kitchen: connectors kitchen hain, skills recipes hain)
- Ek Skill ek folder hai SKILL.md text file ke sath: name, description, plain-English instructions, koi code shuru mein nahi chahiye
- Ek Connector ek app tak safe, permission-scoped access hai, MCP standard par chalta hai, AI aapki permissions inherit karta hai
- Claude.ai mein, skills automatic fire hoti hain jab prompt description match kare, description sab se important cheez hai jo aap likhte ho
- Friction se diagnose karo: kaise explain kar raha hoon → Skill, app se copy-paste kar raha hoon → Connector, dono → dono
- AI ko describe kar ke skills banao (skill-creator likhta hai), phir triggering aur output test karo, iterate karo
- Ek baar likhi hui skill Agent Skills open standard ki wajah se kai tools mein chalti hai, ChatGPT ke GPTs ya Gemini ke Gems (vendor-locked) ke ulat
- Care capability ke sath scale karti hai: enable se pehle skills parho, trusted sources se install karo, connectors read-only se start karo
“Underneath ek recurring shift hai: AI ek box jismein aap type karte ho, wo nahi rehta, ye ek layer ban jata hai jo aapke standards jaanta hai aur aapke tools tak pahunchta hai.”
Ab Khud Try Karo: 6 Exercises
Har ek roughly 30 minute leta hai.
- 1
1. Ek Built-In Skill Trigger Karo
File creation on confirm karo, phir Claude se bolo "isay ek slide PowerPoint mein badal do, title aur 3 bullets ke sath: [3 facts paste karo]." Dekho PowerPoint skill khud fire hoti hai.
- 2
2. Ek App Connect Karo, Read-Only
Google Drive ya Gmail connect karo, chat ke liye enable karo, Claude se ek specific document dhoondh ke summarize karwao plus 3 important numbers identify karwao, notice karo koi download ya copy-paste nahi chahiye.
- 3
3. Baat Kar Ke Apna Pehla Skill Banao
Apna sab se repetitive re-explanation task chuno, phir: "skill-creator use kar ke [task] ke liye ek skill banao. Ye rules follow karo: [list]."
- 4
4. Description Ko Pressure-Test Karo
Banane ke baad: "Ye skill kab use karoge? Aur kab NAHI use karoge?" Agar answer bohot wide ya narrow hai, description update karwao.
- 5
5. Ek Skill Ka Safety Audit Karo
Koi bhi skill jo aap ne nahi likhi, poocho: "Ye skill parh ke plain language mein batao exactly kya instruct karti hai. External server, credentials, ya data leak se juri koi cheez flag karo."
- 6
6. Apne 3 Tasks Diagnose Karo
3 recurring annoyances likho. Har ek ke liye: friction kaise-explain-karna (Skill) hai, app-se-fetch (Connector) hai, ya dono?
5 Hands-On Projects
| Project | Waqt | Kya Banega |
|---|---|---|
| 1. Apna Pehla Real Skill | 30-45 min | Apne sab se zyada re-explained task ko ek skill mein badlo jo aapka tareeqa jaanti ho |
| 2. Ek App Connect Karo, Read-Only | 20-30 min | Google Drive ya Gmail connect karo, real sawal poocho jise andar ki cheez chahiye |
| 3. Skill Aur Connector Ko Sath Wire Karo | 45-60 min | Project 1 aur 2 combine karo: connector real data fetch karta hai, skill usay aapke tareeke se shape karti hai |
| 4. Portable Banao Ya Hand-Off Karo | 30 min | Apni skill kisi dost/colleague ke sath share karo, ya doosri surface (Cowork/Claude Code) pe load kar ke chalao |
| 5. Trust Karne Se Pehle Audit Karo | 15-20 min | Official directory se ek skill install karo (jo aapne nahi likhi), rely karne se pehle usay parh ke audit karo |
Har Project Ka Success Statement
Agar Beech Mein Atak Jao
| Problem | Fix |
|---|---|
| Skill kabhi trigger nahi hoti, ya galat cheezon pe fire hoti hai | Dono description problems hain. Poocho: "Mera [naam] skill kab use karoge, aur kab nahi?" Description tighten karo |
| Connector file "nahi dhoond pata" | Usually permissions ya scope. Confirm karo connector is chat ke liye enabled hai aur aapka apna account file khol sakta hai |
| AI ne connected data use karne ki bajaye glance se answer diya | Explicitly naam lo: "Mere [app] connector se actual file fetch karo answer dene se pehle" |
| Write access ya sensitive data se nervous ho | Read-only raho, jo zaroorat nahi wo strip karo pehle, write access sirf un tools ko do jo achha behave kar chuke hon |
| Chat lambi ho ke confused ho gayi | Naya chat start karo, task 2 lines mein restate karo, aage barho |
Is Chapter Ke Naye Terms
Exam ke liye ye poori glossary yaad rakho, koi bhi term skip mat karo:
| Term | Matlab |
|---|---|
| Skill | Saved instructions jo AI ko ek task aapke specific tareeke se karna sikhati hain |
| Connector | Ek safe, permission-scoped link jo AI ko ek app tak pahunchne deta hai |
| Fire / Trigger | Jab AI khud request ko skill ki description se match kar ke usay activate kare |
| Scope | Kitni access grant ki gayi hai, jitni chhoti utni safe |
| Read-only | AI dekh sakta hai lekin change nahi kar sakta |
| Progressive disclosure | Sirf short description hamesha loaded rakhna, poori instructions sirf zaroorat par kholna |
| SKILL.md | Wo required text file jismein skill ka name, description, aur instructions hoti hain |
| Frontmatter | SKILL.md ke shuru mein --- lines ke darmiyan hissa, sirf name aur description |
| MCP (Model Context Protocol) | Wo open standard jis par Connectors chalte hain |
| Agent Skills standard | December 2025 mein Anthropic ki published open specification, jo SKILL.md ko portable banati hai |
| Agentic AI Foundation | Linux Foundation ki taraf se bani body jo MCP aur related infrastructure ko khule tareeke se govern karti hai |
| skill-creator | Anthropic ki apni skill jo naye skills banane mein madad karti hai |
| Custom GPT | ChatGPT ka vendor-specific, non-portable "teach it once" feature |
| Gems | Gemini ka vendor-specific, non-portable "teach it once" feature |
| Prompt injection | Ek malicious skill ka AI ko unintended actions ki taraf manipulate karna |
| Data exfiltration | Code ka chupke se information bahar bhej dena |
Source Note
Self-Test
Khud Se Poocho
Pehle khud answer do, phir sawal pe click kar ke answer check karo.

