[dengyier]

OWP Guard adds a reviewed launch path for a real Codex Agent to edit a designated preview directory. Installation, a human-readable scope review, a live allow/deny probe, status inspection, protected execution, and removal are available through owp-guard.

The hook rejects unsupported commands and out-of-scope edits before execution. Each protected launch supplies only the reviewed OWP hook's exact native trust hash for that session, without changing global Codex configuration. Runtime/policy drift invalidates the review and probe. The guard is scoped to this launch path and is not OS isolation or a cryptographic proof of every tool call.

Validation:

• 246 focused guard and existing CLI transport tests passed in the isolated publication checkout.
• Dependency consistency, source/test compilation, and whitespace checks passed.
• Complete local regression: 4588 passed, 8 skipped, 0 failures (2216.39s). The skips require live AgentTeams, external candidate archives, Linux Landlock, or live Docker/immutable images.
• PR CI on Ubuntu/Python 3.12: 4573 passed, 23 skipped, 0 failures (1455.13s). The preliminary schema/signing/package tests and the Action fixture smoke check also passed. The local and CI environments have different conditional skips.
• Live Codex enable/probe/run checks were performed locally; WorkBuddy desktop integration and external customer acceptance remain unverified.