ControlLeast privilege · approvals · sandboxing · audit

Permissions and safety

The trust boundaries that constrain what an agent can read, execute, change, publish, purchase, or send on a user's behalf.

Use it when

  • Always—especially when tools reach files, networks, credentials, production systems, or other people.
  • An agent or subagent can perform side effects without continuous supervision.
  • Downloaded plugins, hooks, or MCP servers introduce new executable code or data access.

Choose another layer when

  • A prompt-only promise where a platform policy, sandbox, or approval boundary is available.
  • Permanent broad access granted to avoid designing a narrower workflow.

Production path

Build it in this order

  1. 01

    Map actions to authority

    Classify each capability by data access, side effect, reversibility, cost, and who or what it affects.

    Quality check: Every meaningful action has an explicit authorization source.

  2. 02

    Enforce boundaries outside the prompt

    Use sandboxing, scoped credentials, allowlists, policy hooks, approval gates, and protected environments.

    Quality check: Ignoring an instruction cannot bypass the critical control.

  3. 03

    Log and rehearse recovery

    Preserve tool calls and approvals without leaking secrets; test denial, timeout, rollback, and credential revocation paths.

    Quality check: Operators can reconstruct what happened and contain a bad run.

Artifacts you may see

  • permission policies
  • sandbox configuration
  • allow and deny rules
  • audit logs

Evidence ledger

Primary documentation behind this guide

Provider-specific behavior is sourced to that provider. Open standards are sourced to their specifications and official maintainers.