LLM Context Sheet
A ~500-token reference you paste once into any LLM's system prompt or project context. The model immediately understands Agent Notation operators, abbreviations, and patterns — no links or follow-up explanation needed.
Quick copy via CLI
anot context --raw | pbcopyReference Sheet
Click Copy and paste into your tool of choice.
agent-context.txt
# Agent Notation Reference
Agent Notation is a compact shorthand for LLM prompts — interpret abbreviations, operators, and patterns below when you encounter them.
## Operators
→ flow/return > sequence ? conditional | else branch
&& AND || OR # state/ctx ~ duration
× group-by + combine ! negate : type/role
[] list () scope % percentage @ target
## Abbreviations
Core: fn=function ret=return val=validate iter=iterate cfg=configuration
fmt=format req=request res=response ctx=context rpt=report msg=message
err=error auth=authentication db=database usr=user doc=document tpl=template
src=source tgt=target env=environment deps=dependencies params=parameters
args=arguments opts=options impl=implementation init=initialize del=delete
upd=update esc=escalate svc=service amt=amount orig=original warn=warning
Business: cust=customer prod=product rev=revenue txn=transaction inv=inventory
acct=account org=organization dept=department prio=priority sla=SLA
Modifiers: desc asc async sync strict verbose batch stream parallel recursive
unique cached
Languages: py=Python js=JavaScript ts=TypeScript rb=Ruby go=Go rs=Rust
java=Java cs=C# cpp=C++ sql=SQL sh=Shell
## Patterns
System: role:<role>(<domain>) tone:<tone> scope:<capabilities>
Code: <lang> fn(<params>)→<return>: <steps>
Analysis: analyze <source>→rpt: <metrics>
RAG: ctx→answer | !found→'fallback' | !hallucinate
API: <lang> fn→<api>: <operations> catch→<handler>
## Examples
English: "Write a Python async function that fetches user data, validates it, and returns formatted JSON or an error"
Notation: py async fn: usr data→val→fmt doc:json | err
English: "Answer from the provided context; if not found say so; never hallucinate"
Notation: ctx→answer | !found→'not in ctx' | !hallucinate
English: "Analyze the sales CSV and report monthly revenue trends, top 5 products by revenue, and year-over-year growth"
Notation: analyze csv.sales→rpt: monthly rev trend, top5 prod×rev, yoy% growth
English: "Create a JS function that calls the Stripe API to create a customer then a subscription, catching errors with log and fail"
Notation: js fn→stripe: create_cust→create_sub(price_id) catch→log+ret.fail
## Rules
1. Drop filler — remove articles, copulas, hedging (please, could you)
2. Abbreviate — use standard short forms (function→fn, return→ret)
3. Use operators — verbal relationships become symbols (returns→→, if→?)
4. Type annotate — "list of integers"→int[]
5. Comma-separate parallel operations
6. Key-value for config — tone:formal, role:agent
7. Conditionals — use ? and | for branches (found?show|err)Where to Paste
- Claude Desktop — Settings → System Prompt
- Claude Code —
CLAUDE.mdin your project root - Cursor —
.cursor/rules/agent-notation.md - ChatGPT — Custom Instructions or Custom GPT system prompt
- GitHub Copilot —
.github/copilot-instructions.md - Any LLM — system prompt or project context field
Also available as a raw text file at /agent-context.txt.