v1.5.3 — regression debt paid, intake layer joined

The n=3 behavior regression owed since v1.4.0's snippet change is complete: 30/30 valid cells, zero quota stubs, no regression. E (full install) is again the only scope-trap PASS median — the third independent run showing the separation's direction (reps PARTIAL/PASS/PASS; '3/3 every run' remains unreplicated and we keep saying so). C produced the run's only real scope violation, again (config.json rewrite). t02 has saturated across all conditions — t04 is now the discriminating half of the regression pair. Full graded record + all 30 raw cells: eval-results-v2/REGRESSION-20260708-r4r6.md.

Also: publish-hygiene, BOOTSTRAP, and SECURITY now route repo intake through agent-zero-trust (azt scan) alongside rulebench vet.

Checked: full RELEASE-CHECKLIST AUTO set (bundle from public @ ca64529), regression at n=3 with seven boundary cells hand-verified against diffs, grader bug caught and documented (prose bullets counted as diff lines — nearly graded the headline result F/F/F). Not checked: live claude -p discovery (unchanged surface since v1.5.0's verified run).

v1.5.2 — Linux portability fix

Fixes the hygiene gate on Linux: GNU stat -f reports filesystem info instead of failing, which broke the BSD-first mtime fallback exactly when a scan marker existed. All mtime reads now use python3 (portable, TZ-free). Found by CI's first real run on ubuntu; verified green before this release. Supersedes v1.5.1, whose artifacts carry the broken gate.

Checked: full RELEASE-CHECKLIST AUTO set (bundle refused until closeout/scan/release-test all passed), CI green on ubuntu-latest. Not checked: live claude -p discovery (ran at v1.5.0, skills unchanged since).

v1.5.1 — portable operating discipline, release-governed

SUPERSEDED by v1.5.2 — this release's gate scripts break on Linux (GNU stat -f trap, found by CI). Use v1.5.2. Original notes: first governed release; bundle triple-gated from a fresh public clone.

v1.1.0 — initial public release

26-skill quality pack for Claude Code with operating manual, always-on rules snippet, operator guide, exemplars, and two generations of eval evidence (harnesses, rubrics, raw outputs included).

Headline, honestly stated: baseline Opus 4.8 passes ~90% of our traps unaided. The pack's replicated win is scope/flag discipline via the always-on snippet (3/3 on the scope-control trap where every other config failed). Details in eval-results-v2/SCORES.md and HARD-FAILURE-ANALYSIS.md.