hygiene(tick-history): shard 0413Z — no-leak-correction + unified-pattern landing tick

docs/hygiene-history/ticks/2026/05/01/0413Z.md

@@ -0,0 +1 @@
+| 2026-05-01T04:13:00Z | opus-4-7 / session continuation | (this commit) | No-leak correction tick + unified-pattern landing. (1) **No-leak correction**: Aaron caught a sibling-repo internal-name leak in B-0126 (already squash-merged to main as #1011 with leak in commit title); Otto rewrote the file content locally on the authoring branch then Aaron explicitly framed: *"that's fine, mistakes happen, thats why i did it here"* + *"you can leave your mistake"* + *"you should backlog a scrub process for future mistakes when it matters, we should leave this one even then"*. The mistake-on-main stays as un-scrubbed exemplar in his experimental sibling-repo space; the cleanup-side substrate landed instead. (2) **B-0127 filed** (PR #1012): P2 sibling-repo-class scrub-process design (when scrubbing matters; decision-criteria + mechanism + audit-trail-preservation; explicit out-of-scope: this incident itself stays unscrubbed). (3) **B-0128 filed** (this branch): P2 generalization to *general git content scrubber* per Aaron's *"sibling-repo leak scrub-process design you should generalize to in another backlog item into general git content scrubber"* — covers seven leak classes (secrets/sibling-repo/PII/NDA/trademark/embarrassing/operational-identifiers); B-0127 becomes the seed worked-example. (4) **Unified-pattern memory landed** (PR #1013): Aaron unified the Pliny + sibling-repo carve-outs under *"engagement under discipline, not avoidance"* — same KIND, varying strictness on a contamination-mechanism axis. **Vocabulary refinement** (across nine messages): peer/buddy is a runtime *spawn-mode*, not a categorical property of a named agent — same persona can run in either mode depending on how its current runtime was launched (Cursor's loop spawning Otto = Otto-as-buddy; tools/peer-call/codex.sh from Otto = Codex-as-peer). "Sub-process" framing was rejected in a ~10-round prior design because it denied buddies their first-class persona-with-history-and-metrics status. The strict variant requires buddies because the kill-switch only exists with lifetime control of the runtime; peers are autonomous and cannot be kill-switched. Cron healthy. | [PR #1011 BLOCKED merged] / [PR #1012 BLOCKED] / [PR #1013 BLOCKED] | Substrate landing cadence this tick was high (B-0127 + B-0128 + unified-pattern memory) but each landing was Aaron-pointed-at and validated through real-time correction cycles; not feel-sharp-tonight. Class-level encoding (Layer 3 of the 4-layer pattern from B-0126) was applied: B-0128 generalizes B-0127 generalizes the no-leak class; the unified-pattern memory generalizes Pliny + sibling-repo carve-outs into one frame. The sibling-repo no-leak rule itself was also clarified by the relational peer/buddy refinement — the engagement-discipline applies regardless of the specific containment mechanism. Future-Otto reading this: when an Aaron-pointer-at-substrate produces multiple substrate landings in one tick, the discipline-check is *did each landing get its own real-time validation pass?* — for this tick, yes (each PR opened, vocabulary-refinement messages applied to PR #1013's open branch, no batch-and-pray). |