Follow-up: session ownership, portfolio closeouts, and Codex runtime acceptance #101

Recent-session audit follow-ups remain outside PR #100's locked three-task scope. This issue records investigation priorities; it does not authorize ownership migration, session restarts, or new feature implementation.

1. P1 — authoritative owner and supersession across worktrees. Permitted progress evidence shows an owning worktree can abandon a plan while the primary checkout retains an older Locked copy. The anchored-status fix in fix: reject stale plan status and distinguish checkpoint acceptance #100 rejects terminal documents, but cannot decide which checkout owns the plan. Build on #52 and existing session attribution/re-anchor checks; determine an explicit owner binding and supersession contract before any migration. Regressions should model two sessions/checkouts with conflicting current statuses and prove no cross-session claim or mutation.
2. P2 — schema-safe portfolio delta closeouts. The concurrent read-only audit found shared-workspace tracking and stale singleton context cannot establish ownership or customer acceptance. Inventory the existing PROJECTS/PORTFOLIO/FOLLOWUPS schemas and consumers before designing updates. Require preservation of unrelated rows, writer attribution, and separate checkpoint/parent acceptance semantics. No portfolio mutation defect has been reproduced in this patch; this remains an investigation, not an assertion of a verified bug.
3. P2 — semantic version selection in fallback root discovery. hooks/hooks.json uses lexical sort | tail -1 when explicit roots and the working directory fail. Lexical ordering puts 6.9.0 after 6.10.0. Prefer the installed source identity or a validated semantic-version fallback and test multi-digit versions. The current native-root path is valid through Codex's documented CLAUDE_PLUGIN_ROOT compatibility alias; fallback failure is not claimed for observed active sessions.
4. P2 — supported installed-host callback and safe cache-retention acceptance. Native same-identity upgrade/rollback works in an isolated Codex home, but native upgrade deletes prior version caches/custom files. Preserve private backups outside cache and resolve safe active-path retention before an active-user upgrade. A fresh isolated install of PR fix: reject stale plan status and distinguish checkpoint acceptance #100 discovers all 11 hooks without errors, and all 246 tracked files match the candidate. Exact-root wrapper script smoke passes, but discovery/direct execution is not proof of lifecycle callback invocation. Use reviewed/trusted hooks in a new session, never an unsupported trust bypass.

Official Codex hook documentation establishes that unified exec_command matches Bash; apply_patch also matches Edit and Write; plugins receive PLUGIN_ROOT plus CLAUDE_PLUGIN_ROOT compatibility; non-managed hooks require trust. Thus the existing matcher names alone are not a demonstrated compatibility bug.

Verified distribution facts at audit time: installed v6.6.1 matches main ac761fecd2410c85a1dfbc2f2591208b17c5dd64 except mutable progress state, while tag v6.6.1 is 8fb5671f3b2ef99b9190516ee9b9ec0afe146844. Remote marketplace manifest advertises 6.5.11; the existing dirty local external marketplace is 6.1.2 and was not overwritten. Workspace PROJECTS records 6.5.4, which is tracking metadata rather than release truth. The latest GitHub Release is v6.5.0 even though later official tags exist; release-tag workflow creates tags but not a GitHub Release page.

PR #100 already addresses anchored current status and explicit checkpoint-versus-parent acceptance. No unrelated session was stopped or messaged, and no personal transcripts, credentials, or raw private progress records are included here.