tags: [ai, productivity, devtools, tutorial]
understanding how claude code thinks changes everything
it's not a search engine. it's an autonomous agent running in a loop
think → act → observe → correct → repeat
this is why certain prompts work brilliantly while others fall flat
context window basics
every message you send includes the entire conversation history. long conversations consume exponentially more tokens
what fills up your context:
- your prompts
- claude's responses
- tool results (file contents, command outputs)
- system context (claude.md, git status)
- session history
/clear should be your default
most people treat /clear as a last resort. experienced users clear aggressively between tasks
it's not because something went wrong. it's standard practice
plan mode saves hours
shift+tab twice to enter plan mode. claude can read and research but can't modify files
planning is cheap. undoing is expensive
five minutes in plan mode often saves hours of reverting bad implementations
effective prompting patterns
be specific about completion. the loop needs to know when to stop
provide context upfront. help claude find what it needs faster
break down large tasks. one long loop is worse than multiple short loops
part 2 of the mastery series
Read the full guide
Top comments (0)