Context adoption use cases
A new project: inspect the actual application manifests and authored policy, preview the bootstrap, then install the authorized layout. Add specialized rules only for its stack. A new directory does not grant permission to install machine-wide services.
An existing project: preserve authored instructions, protocol overrides and selected team records. Resolve ambiguous teams explicitly. A legacy v3 migration needs a reviewed report; malformed managed markers or links require deliberate repair.
A mixed-harness project: keep one documented authority for each rule source and preserve the native definitions each harness already owns. Definitions, hook activation, file discovery and actual delegation have different prerequisites. Follow selected roles sequentially when native delegation is unavailable.
A nested application or vendored subtree: use the nearest applicable rules and actual manifests. Preserve vendor source and licenses. Do not rewrite unrelated nested agent files to make a root context diagnostic green.
Recovery after compaction: recover canonical project/run/task identity, plan revision, owner and outstanding evidence. A waypoint is a compatibility projection, not mutation authority. A fork can retain parent context, so full Codex SessionStart delivers team digest metadata rather than private lesson text.
See quick start, team handoffs and KBD recovery.