Plankton

plankton

Write-time code quality enforcement for AI coding agents, built on Claude Code hooks.

Research project under active development. Hooks are tested against
Claude Code >= 2.1.50 (see badge). Newer CC versions usually work
but are not guaranteed. Disable CC auto-updates to prevent
silent breakage (see Quick Start). If you encounter issues, file a
report including the output of claude --version. To pin to a
specific Plankton version: git checkout v0.1.0.

quick start

That's it. Plankton works by being the directory you run Claude Code from.
The hooks in .claude/hooks/ are picked up automatically — no install
command, no plugin, no config. Clone, setup, claude.

If you prefer non-interactive setup, use:

Windows is not supported. Use
WSL2 and
follow the Linux instructions inside the WSL environment.

You can work on any codebase from inside plankton. Just tell Claude:
*"work on /path/to/my-project"* — it will apply the same quality
enforcement to that codebase while the hooks stay active.

Existing codebases: when the agent edits a file, Plankton runs the
configured checks for that file, including pre-existing issues. Some issues
are auto-fixed on first touch; unresolved ones come back as feedback. Scope
down only the Python security scanners here if needed:

Recommended: disable Claude Code auto-updates. Plankton depends on
undocumented CC internals — a silent auto-update can break hooks without
warning.

scripts/setup.sh installs all tools automatically. On macOS it uses
Homebrew; on Linux it downloads prebuilt binaries (no cargo or go
required). See docs/SETUP.md for manual per-language
setup.

Manual install (if you prefer not to use the setup script)

macOS:

Linux:

See docs/SETUP.md for per-language details and download
URLs.

what is plankton

Code…