Skip to content

security: add restricted real-agent scale evidence producer #21

Description

@rblake2320

Problem

The ecosystem scale-readiness gate cannot safely consume the legacy real-agent v3 runner. That runner uses provider bypass/trust-skip flags, broad inherited environment state, timestamp-derived nonces, loose HWND discovery, and model-call claims that are not directly observable.

Scope

Add a separate, versioned restricted scale producer that:

  • requires a disposable fresh checkout verified against canonical remote HEAD;
  • rejects unsafe provider flags;
  • constructs provider-specific environment allowlists;
  • uses cryptographic random nonces;
  • binds each HWND to the spawned process tree and verifies PID, executable, class, and exact title;
  • records CLI invocation accounting only;
  • emits per-rung UTC start/end timestamps and structured guard/cleanup evidence;
  • fails closed for stale/future timing and incomplete cleanup.

Add adversarial unit tests. Do not run live providers, merge, or make readiness claims in this issue.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions