carlos / platform Public

1. P2 — Minting can permanently override a key whose configuration has not arrived yet. instancekey.go:299 A missing local <host>.config.env is treated as no configured key. However, reconciliation publishes a ready route before the later SyncExec pass; the admission guard checks whether configuration can render, without delivering it. A first wake in that interval mints B even when the bucket already specifies A. Once configuration arrives, B continues winning from <host>.env. Require configuration delivery before deciding to mint, and test first launch with configuration delivery delayed.

1. P2 — A flat env FIFO can block the entire startup backfill. ownenv_unix.go:18 O_NOFOLLOW rejects symlinks but still opens FIFOs in blocking mode. A ready flat row whose <host>.env is a FIFO without a writer hangs here while holding adminTokenMu. Because the new backfill runs synchronously before svc.Start, one malformed file prevents startup from completing. The flat configuration-file read has the same problem. Use nonblocking opens followed by regular-file checks; extend the existing nested FIFO test to flat backfill.

1. P3 — The adoption endpoint accepts malformed trailing JSON. keyadoption.go:96 Decoder.More() does not establish EOF: it returns false for a trailing } or ], and suppresses reader errors. A valid adoption body followed by either delimiter therefore writes and publishes a record instead of returning 400. Require a second decode to return io.EOF, with tests asserting no write or publication on malformed input.

Verdict: not ready to merge.

Reviewed 0901e835..3e32533e. Go formatting and Go-file whitespace checks passed. Focused tests could not start because the read-only sandbox prevented Go from creating its work directory. No files changed.

---

FIRST PROMPT: “Follow the review brief docs/superpowers/plans/CODEX-REVIEW-INSTANCE-KEY-BUILD-BRIEF.md. Review the code diff 0901e835..HEAD on branch instance-key-build.”

WHAT: Review the instance-key build against the brief and binding rulings.

STATUS: Review complete: two P2 findings and one P3.

NEEDED: Nothing.

RECOMMENDED: Fix the findings, add regression coverage, and rerun the gate before merge.

SESSION: Safe to close. Working tree unchanged; no background processes started. Handoff file could not be saved in the read-only sandbox.