Mearman/agent-permissions

agent-perms is a vendor-neutral JSON policy format and tooling for controlling permissions of AI coding agents (Claude Code, OpenAI Codex, Crush, Gemini CLI, Cursor, etc.). It’s intended for teams or projects that run multiple agent harnesses or migrate between them. What it does: - Single policy file (.agents/permissions.json) expressing rules as tool + optional pattern + tier (deny/ask/allow) with optional when-conditions; deny-first evaluation and a defaultMode fallback. - Claude Code compatibility and zero-translation migration. - Supports profiles, delegation, sandbox/network settings, env, and per-agent overrides; local overrides via .agents/permissions.local.json. How to use: - Install as an MCP server (syncs native agent configs) or as a library. - Programmatic API: convert, validate, check, detectFormat, plus codecs and loaders. - Provides JSON Schema for IDE autocomplete and validation.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.