Agent Integration¶
Squeez works with any LLM coding agent that can run shell commands.
Claude Code¶
Add this to your project's CLAUDE.md (or ~/.claude/CLAUDE.md for global):
Always when you invoke a shell command, pipe it through `squeez` and tell exactly what you want to know.
Examples:
- `bun test 2>&1 | squeez "did the tests pass?"`
- `git log --oneline -50 | squeez "find the commit that broke CSRF"`
- `cat src/auth/middleware.py | squeez "find the referer validation logic"`
Do NOT use squeez when:
- You need exact, uncompressed output (e.g. writing a patch)
- The command is interactive
Codex CLI¶
Add equivalent instructions to your Codex configuration or system prompt.
OpenCode¶
Add to your OpenCode instructions file.
General pattern¶
The integration pattern is the same for any agent:
- Install squeez and configure a backend (local model or API)
- Instruct the agent to pipe shell output through
squeez "what to look for" - The agent gets compressed output with only the relevant lines
When NOT to use squeez¶
- Writing patches — the agent needs exact file contents
- Interactive commands — squeez reads from stdin
- Short output — no point compressing 5 lines
- Debugging squeez itself — avoid infinite loops