fp-enforcer

Real-time enforcement of the functional programming paradigm for agent-driven codebases. fp-enforcer ships a canonical rule set, a violation scanner, agent hooks, and skills that let coding agents enforce, validate, repair, and learn functional programming discipline on any project.

Install

Install the runtime once per machine from PyPI:

pipx install fp-enforcer
# or: uv tool install fp-enforcer

Then wire it into a project:

fp-enforcer wire --client cursor --dest /path/to/project
fp-enforcer wire --client all --dest /path/to/project
fp-enforcer unwire --dest /path/to/project

Update is a two-step remote refresh:

pipx upgrade fp-enforcer
fp-enforcer wire --client all --dest /path/to/project

Marketplace installs (agent-facing layer)

• Claude Code: /plugin marketplace add theocarranza/fp-enforcer, then /plugin install fp-enforcer@fp-enforcer.
• Codex: codex plugin marketplace add theocarranza/fp-enforcer, then codex plugin add fp-enforcer@fp-enforcer.
• Cursor: via the Cursor Marketplace (pending listing).

Marketplace plugins deliver the skills and rules; the fp-wire skill then installs the runtime from PyPI and wires the project (hooks included).

What it does

Design

• Generic sources, vendor adapters. Rules (fp_enforcer/rules/), pattern packs (fp_enforcer/rules/patterns/), and the Python core are vendor-neutral and ship inside the PyPI package. wire renders the vendor-specific harness (Cursor, Claude, Codex, Antigravity) into the target project. Nothing is ever written to host-global folders.
• Stdlib only. The runtime is Python 3.10+ standard library, written in the functional style the product enforces: frozen dataclasses, pure functions, pattern matching, effects at the edges.
• Skills follow the Agent Skills open standard; marketplace manifests are optional adapters (Claude Code, Codex, Cursor) on top of the PyPI distribution.

Wiring a project

fp-enforcer wire:

1. Vendors the runtime into <dest>/.fp-enforcer/ (package, rules, launcher).
2. Renders rules for the client (.cursor/rules/*.mdc, .agent/rules/*.md, ...).
3. Merges hook wiring non-destructively into existing hooks.json / settings.json files (existing entries are preserved; ours are tagged by the .fp-enforcer/run_hook.py command and are idempotent).
4. Records everything in <dest>/.fp-enforcer/install_manifest.json so unwire reverses only what was installed.

Validate and learn

# Scan paths (files or directories)
fp-enforcer validate path/to/lib

# Scan only the added lines of a diff
git diff | fp-enforcer validate --diff -

# Generate FP library knowledge profiles for agents
fp-enforcer learn --dest /path/to/project

# Activate curated pattern packs for detected FP libraries/features
fp-enforcer assimilate --dest /path/to/project

Suppressing a justified exception

Tripwire patterns can over-fire on constructs the rules themselves permit (e.g. CompositeSubscription.add, edge throws). Waive a pattern inline with a directive naming the pattern ids — on the violating line or the comment line directly above it:

// fp-allow: no-collection-mutation — disposal binding, fp-rxdart memory 1
composite.add(subscription);

Every suppression is grep-able (grep -r "fp-allow"), keeping exceptions explicit and auditable, as fp-core requires.

Layout

fp_enforcer/           python stdlib core: validate, hook_runtime, adapters, learn, install
fp_enforcer/rules/     canonical rule set (vendor-neutral markdown + manifest.json)
fp_enforcer/rules/patterns/  machine-readable violation patterns per language
fp_enforcer/skills/    fp-validate, fp-repair, fp-learn, fp-assimilate, fp-wire (SKILL.md)
hooks/hooks.json       plugin-level hook wiring template
tests/                 stdlib unittest suite
.github/workflows/     CI + tag-triggered PyPI release (trusted publishing)

Development

Run from a checkout without installing:

python3 -m unittest discover -s tests -v
python3 -m fp_enforcer validate .        # module form works from the repo root

The repository dogfoods itself: its own .cursor/ harness is generated by its own installer.