elliot35/deterministic-agent-control-protocol

npm i @det-acp/core
npx det-acp init cursor
npx det-acp init codex
npx det-acp init claude-code

Usage

• •Define a policy in agent.policy.yaml (capabilities, limits, gates, forbidden, session)
• •npx det-acp proxy --policy ./policy.yaml
• •npx det-acp exec ./agent.policy.yaml <command>
• •npx det-acp proxy --policy ./policy.yaml --evolve

Key features

• ✓Policy DSL covering capabilities, scopes, budgets, gates, evidence, forbidden patterns, and session limits
• ✓MCP proxy, shell proxy, HTTP session server, and TypeScript library SDK integration modes
• ✓Tamper-evident audit ledger with SHA-256 hash chaining
• ✓Rollback manager with compensation plans for file, git, and network operations
• ✓Policy self-evolution on denials with add-to-policy / allow-once / deny choices
• ✓Ready-built policies for coding agent, devops deploy, video upscaler, data analyst, security audit, infrastructure manager

Best for

Teams wanting to enforce, audit, and human-gate what AI coding agents can do in a workspace

Caveats

• ⚠Does not execute actions itself; you must perform execution after an allow verdict
• ⚠Soft governance modes rely on instructions that a creative prompt could bypass
• ⚠Policy self-evolution suggestions require user approval and are only for permission/scope denials