PERMISSION/PROTOCOL

Use cases

Authority gates for the places AI agents can hurt you.

Permission Protocol adds explicit approval, consequence-aware policy, and signed receipts to AI-authored production changes, GitHub PRs, and sensitive agent tool calls.

AI agent deploy gate

Deploy gate for AI-authored production changes

Add a deploy gate for AI-authored PRs. Block risky production changes until a named human signs the exact repo, PR, commit, policy, and action.

Example

A coding agent opens a PR that updates a deploy workflow, changes a migration under db/migrations, and removes a rollback step. CI is green, but the production consequence is larger than the code diff looks.

Read guide

GitHub AI code review governance

Approval routing for AI-generated GitHub PRs

Route AI-generated GitHub PRs through consequence analysis, approval gates, and signed receipts before risky code reaches production.

Example

An AI agent opens a polished PR that changes an auth callback, updates a deploy workflow, and includes a confident summary. Reviewers can read the code, but the merge still needs explicit authority because the consequence is production-impacting.

Read guide

MCP tool call authorization

Approval gates for MCP tools and agent actions

Put approval gates around MCP tools and agent actions. Require authority receipts before agents call sensitive tools, write data, deploy code, or spend money.

Example

An agent has an MCP tool that can update Stripe, delete rows, or message customers. The same server also exposes harmless read-only methods, so a broad allowlist cannot distinguish inspection from production mutation.

Read guide

Data Mutation Gate

Data mutation gate for AI agents

Require signed authorization before AI agents delete, update, or mutate production data, backups, volumes, and provider resources.

Example

An agent has a provider token that can delete a production database, backup, or volume. The change does not require a PR, so the gate must sit before the provider API call.

Read guide

What every use case has in common

A sensitive AI action is about to happen.
A named human or policy must authorize it.
A receipt proves what was signed before execution.