Security

Approval-Gated Safe Execution Model

ForgeAI Level 0 uses explicit gating and bounded execution controls. This page avoids unsupported compliance claims.

Current Controls

  • Planning-only by default
  • Default-deny execution
  • Explicit human approval
  • Explicit authorization
  • Proposal-bound approval
  • Approved operation IDs
  • Repository-root containment
  • Path traversal rejection
  • Symlink rejection
  • SHA256 and byte-size preconditions
  • Atomic write
  • Rollback on post-write failure
  • Command verification remains NOT EXECUTED
  • No source content or secret exposure in summaries