Trust Hardening + Token Cap

v0.2.0 delivers two phases of work: Phase 1 hardened the enforcement boundary and closed known security gaps; Phase 2 updated all public-facing copy to match the new capabilities and repositioned the product around safe flow state.

v0.1.0-Initial release

First public release of GouvernAI — runtime guardrails for Claude Code.

• Dual enforcement: skill-based 4-tier risk classification + deterministic PreToolUse hooks
• /guardrails slash command with strict, relaxed, audit-only, and reset modes
• Mode persistence via guardrails-mode.json (survives context resets)
• Hard blocks: obfuscated commands, credential exfiltration, catastrophic operations, self-modification
• Escalation rules: unfamiliar targets, bulk ops, scope expansion, chained actions
• Append-only audit log
• 67 unit tests on the hook layer
• Zero dependencies, MIT license

Install:
/plugin marketplace add Myr-Aya/GouvernAI-claude-code-plugin
/plugin install gouvernai@mindxo