cupcake
cupcake copied to clipboard
A native policy enforcement layer for AI coding agents. Built on OPA/Rego.
Watch out for https://github.com/google-gemini/gemini-cli/issues/2779
Codex hooks do not exist as of September 2025. https://github.com/openai/codex/issues/2109 https://github.com/openai/codex/pull/2904
https://github.com/microsoft/regorus
## Problem `cupcake init --global --harness claude` always wrote hooks to `~/.claude/settings.json`, ignoring Claude Code's `CLAUDE_CONFIG_DIR` environment variable. Users who relocate their Claude config via `CLAUDE_CONFIG_DIR` got the integration installed...
## Problem Several docs and the rulebook template state that builtins are **enabled by default when configured**, and the `enabled` struct field is documented as "defaults to true". The code...
## Problem The `sensitive_data_protection` global builtin matched its keyword set (`"private"`, `"auth"`, `"secret"`, `"token"`, …) with `contains()` against the **entire canonical path**: ```rego is_sensitive_file(path) if { lower_path := lower(path) sensitive_keywords...
## Summary Running a harness from a directory without a local `.cupcake/policies/` folder exited eval with code 1, blocking every tool call (#104). Two related causes: - **Engine**: the project...