10 CLAUDE CODE TIPS NO ONE KNOWS ABOUT:
1\ use /effort high then add "ultrathink" in your prompt
forces full extended thinking with a 31,999 token budget.
2\ end every important session with a "summary of learnings" paragraph
the Dream system automatically extracts it and injects it into future sessions
3\ /fork creates a fully isolated conversation branch with its own plan files
/rewind undoes the last turn including code changes.
4\ tell Claude to "start an Explore agent" or "enter Plan mode"
the state manager instantly injects the correct role instructions.
5\ always use absolute paths
sub-agents and worktrees enforce this strictly. relative paths cause friction and errors.
6\ set up custom hooks in .claude/settings.json
use exit code 2 in a PostToolUse hook to silently block actions and force a rewind.
7\ /fast does NOT change the model
it runs the same Opus with faster output. pair it with /effort medium for the best speed/quality balance.
8\ keep ~/.claude/debug/latest open with tail -f
it shows every trigger, whisper injection, and state manager decision in real time.
9\ run your own local MCP servers
they let you expose custom tools and pause the agent mid-task for structured user input.
10\ prefix important instructions with <system-reminder> tags
the model treats them with the same priority as internal whispers.
what did i miss?
Gm ☀️
I’m still building, still shipping, and continuing to improve Pmpr.
A lot of progress has been made in the last week, and there’s more features actively being built
PMPR remains an open experiment focused on real tooling and real usage
Just focused on continuing to build and making Pmpr the best CT tool there is 🫡
$PMPR Don't forget that the price might be dipping.. but the project itself is at ATH. And growing. The product doesn't change.. people just have short attention spans. They will roundtrip in the next update.
Be smarter.