Treat README.md as a versioned cascade surface (R13 class) — captured…
Decision
Treat README.md as a versioned cascade surface (R13 class) — captured via feedback_repo_cleanup_includes_readme.md + MEMORY.md pointer cascade. Memory-layer fix shipped this session; hook-layer enforcement deferred as P1 follow-up.
Rationale
AJ has called out stale README “again and again” across multiple repo cleanups — clear recurring failure pattern. This session shipped: (a) v22.0.3 README rewrite to remote master via PR #16 (squash-merged, branch deleted), (b) feedback_repo_cleanup_includes_readme.md with 6-step mandatory checklist + inline session-close gate (one-liner version-match bash check), (c) MEMORY.md Feedback section pointer at line 155 so the rule loads at session bootstrap. ST detect_biases caught my initial framing: “this will definitely prevent recurrence” was overconfidence_bias — disconfirmed. Memory persistence is NECESSARY but NOT SUFFICIENT: I’ve added behavioral feedback rules before for recurring failures and still skipped them when attention was elsewhere. The actual structural fix is a Hookify PreToolUse rule that detects gh pr merge / git push origin master|main / gh repo edit --description and blocks unless either (i) README.md was in the diff, or (ii) explicit README_OK=1 env override. That hook is deferred to P1 because (a) it requires more substantial Hookify rule design than fits this turn’s scope, (b) the memory-layer fix unblocks AJ’s immediate frustration. Logging this as the explicit known gap so the next session picks it up rather than discovering it again the hard way.
Alternatives Rejected
Outcome
Pending