Five repos that stop Claude hitting its limit

Fewer tokens per answer, smaller chats, and rules Claude remembers. So your plan lasts past lunch.

You need: Claude Code (Pro plan or above). Most also work as skills in the Claude app.

Claude's usage limit is mostly eaten by long replies, giant files and you repeating the same corrections. Each of these fixes one of those leaks.

The five repos

Caveman
Makes Claude answer in short caveman-style sentences. Its makers say it cuts about 65% of the output tokens.
Ponytail
Turns Claude into the laziest senior dev in the room: it writes about half the code for the same result.
Headroom
For giant files: Claude reads an outline first and opens only the parts it needs, so your chat doesn't fill up.
Task Observer
Correct Claude once and it saves that as a lasting rule, so you stop paying tokens to repeat yourself. Same repo as Headroom.
Claude Code Setup
Anthropic's plugin: scans your project and recommends which automations (hooks, skills, commands) to add.

1Claude Code (the terminal app or the Code tab in Claude Desktop): type /plugin marketplace add followed by the repo's owner/name, for example /plugin marketplace add anthropics/claude-plugins-official.
2Type /plugin, open the marketplace you just added, pick the plugin and press Install. Restart Claude Code if it asks.
3Claude app instead? Plain skill folders work there too: Settings, Capabilities, turn on Skills, then Upload skill with a ZIP of the skill's folder (the one with SKILL.md inside).
4Then just ask for the job in plain English. Claude loads the right skill by itself.

Ashar's tip: Start with one. Install Caveman, work for a day, and compare how far your limit gets. Then add the next.

follow.txt

Ashar posts a new one daily. Follow so you don't miss the next keyword.

new_packs.txt

Ashar sends a new copy-paste setup most weeks.