ProjectGate Alpha is an AI workflow evidence layer and controlled runtime toolchain. It is not a system sandbox, an enterprise compliance platform, or an operating-system-level enforcement layer.

ProjectGate can enforce rules inside ProjectGate-controlled flows:

• TaskRun creation
• stage gate checks
• delivery checks
• observation gate checks
• TaskRun continuity checks
• pg exec controlled command execution
• candidate review / approval / rejection / merge workflows

ProjectGate cannot yet guarantee interception of every action taken by an external AI client that bypasses ProjectGate.

ProjectGate installers and adapters must not delete arbitrary directories.

Beginning in v0.4.5, managed replacements are restricted:

• overwrite targets must be non-root, non-home, and sufficiently deep paths
• overwrite targets must be inside the intended workspace when applicable
• existing ProjectGate-managed directories receive a backup before replacement
• unmarked non-ProjectGate directories are refused
• copied ProjectGate-managed directories receive .projectgate-managed.json

• Native transcript hooks for all third-party AI clients are not complete.
• ProjectGate-controlled commands are observable; commands run outside ProjectGate are outside this boundary.
• The project is not yet distributed through PyPI or signed release artifacts.
• The test suite is still smoke/selftest oriented and not a full formal verification suite.

Do not use ProjectGate Alpha as a sole security control. Use it as a workflow evidence layer while retaining normal source control, review, sandboxing, backups, and permission controls.