claude-shorthand/claude-shorthand.md

Claude Shorthand Cheat-Sheet

Core Operators

Pattern	Meaning	Example
→	then (sequential)	ssh server → check status
+	and (parallel)	check proxy + check dns
X ? Y : Z	if X then Y else Z	port open ? connect : debug
!X	don't / skip	!error handling
w/	with	py script w/ argparse
@	at path/location	new config @ /etc/nginx/
re:	regarding	fix re: pool degraded drive
=	set/assign	port=3000, lang=python
~	approximately / like	~50 lines, ~yesterday
>>	output to / save to	results >> /tmp/report.txt
<-	from / source	data <- rss API

Spec Block (complex tasks)

feat: <short description>
- lang: <language>
- deps: <dependencies>
- in: <input/source>
- out: <output/target>
- !: <things to skip>

Examples

Verbose → Compressed

Service setup:

# Before (~80 tokens):
"Please create a new Python service in the automation directory
that connects to the RSS API, fetches today's articles, formats
them for the printer, and runs as a systemd service"

# After (~20 tokens):
feat: daily print @ automation/
- rss API → format → printer
- systemd service

Infra task:

# Before (~50 tokens):
"SSH into the storage server, check if any drives in the pool are
degraded, and if so tell me which ones need replacement"

# After (~12 tokens):
storage-server → pool degraded drives? → list replacements

Multi-step:

# Before (~60 tokens):
"First read the proxy config, then add a new reverse proxy entry for
wiki.example.com pointing to the backend on port 3000, and also
add it to DNS"

# After (~15 tokens):
add wiki.example.com → backend:3000 to proxy + dns

Conditional:

# Before (~45 tokens):
"Check if the LLM service is running on the inference server. If it is,
test the model. If it's not running, start it first then test."

# After (~15 tokens):
inference: llm running ? test model : start → test

Anchors (use known names)

Define short aliases in your CLAUDE.md, then use them everywhere:

Say	Not
db-server	"the PostgreSQL database server at 10.0.1.50"
pool	"the ZFS storage pool on the NAS"
proxy	"the Caddy/nginx reverse proxy container"
dns	"the Pi-hole DNS server"
workstation	"this machine, my desktop PC"
llm	"the LLM inference service"

Quick Rules

1. Drop articles (a, the, an) — Claude doesn't need them
2. Drop politeness (please, could you, I'd like) — zero effect on output
3. Use aliases from CLAUDE.md — they're shared vocabulary
4. Bullets > paragraphs — less ambiguous, fewer tokens
5. Imply obvious steps — "add X to caddy" implies read/edit/save