pre-pr-gate-hook: .quality-gate-pass marker friction in the sprint flow (per-item rewrite, unreachable env bypass, stale markers)

[lwalden]

Problem

The pre-pr-gate-hook blocks gh pr create unless a fresh .quality-gate-pass marker (within ~60 min, in the Bash-tool CWD) exists — a second artifact separate from the judge verdict the sprint flow already produces (.quality-review-result.json, decision=pass). Three frictions observed:

1. Per-item rewrite required. The sprint TEST flow writes .quality-review-result.json but not the marker; the pipeliner must (re)write .quality-gate-pass for the current item before every gh pr create.
2. Env bypass unreachable. AAM_PR_GATE_BYPASS=1 exported in the Bash-tool shell does not reach the hook — the hook runs as a harness-spawned process that doesn't inherit the tool shell's env.
3. Stale markers falsely unlock. A leftover marker from a prior item (or prior sprint) is present on disk; it must be explicitly removed/overwritten so it can't gate the wrong item.

Evidence (dungeon-game Sprint 13)

Every one of the 5 PRs (#179–#183) required the pipeliner to detect a stale marker (e.g. an S13-001 marker still on disk when creating the S13-004 PR), delete it, and write a fresh per-item marker before gh pr create would pass. The env-var bypass was tried and confirmed non-functional.

Impact

Each PR carries avoidable pipeliner steps + a failure-then-retry on the first gh pr create. Easy to get wrong (wrong-item marker → false pass).

Recommended fix (any one)

1. Hook reads the judge verdict directly — have pre-pr-gate-hook consult .quality-review-result.json (item + decision) instead of requiring a separate freshness marker.
2. Orchestrator stamps the marker during TEST — fold a per-item .quality-gate-pass write (with stale-marker removal) into the documented TEST→REVIEW handoff so the pipeliner never has to.
3. Make the env bypass actually reach the hook, and document it.

Related: #200, #180 (docs-only PR gate handling).