CLI Enforcement

Model-agnostic, hook-level behavioral enforcement for AI coding agents. The enforcement engine (state manager, points/tiers, anti-hallucination, sandbox, cascade) is platform-neutral; where and how it installs onto a given model is supplied dynamically by cli-wikia — so one engine deploys onto Claude, Gemini, Copilot, DeepSeek, Antigravity, … with no per-platform template.

How it works

cli-enforcement deploy <model>
        │
        ├─ asks cli-wikia:  config_root?  hook_events?  instruction_file?
        │
        ├─ maps enforcement STAGES (pre-tool, post-tool, stop, …)
        │      to that model's REAL event names (PreToolUse / BeforeTool / …)
        │
        └─ copies the engine into <config_root>/mcp/ and writes the hook registry

Install

pip install cli-enforcement      # pulls in cli-wikia

Usage

cli-enforcement deploy claude          # dry-run: show the plan
cli-enforcement deploy claude --write  # actually install into ./.claude/
cli-enforcement deploy gemini --write  # same engine, wired to Gemini's events
cli-enforcement status                 # scorecard: points, tier, streak, KB gates, failures
cli-enforcement status --json          # same data, machine-readable (used by cli-fleet dashboard)
cli-enforcement remove claude --write  # remove the engine

Dry-run by default — nothing is written without --write.

License

PolyForm Noncommercial 1.0.0 — free for noncommercial use. Commercial use requires a paid license: matrixbuilderops@proton.me