rule-driftlisted

# rule-drift Audit Claude's adherence to the user's standing rules, using the user's own recent chat history as evidence. Every rule leaves the audit with exactly one verdict and, wherever wording changes are proposed, the exact drafted replacement line. ## Purpose Extract the user's standing rules from in-context persistent memory, probe recent chats for violations of each rule, score adherence with evidence, and verdict every rule KEEP / REINFORCE / REWRITE / RETIRE. Optionally stage confirmed changes through a gated `memory_user_edits` sequence. This audits assistant behavior against the user's rules — it never grades the user. ## Triggers Run this skill when the user says any of: `run rule drift` · `rule compliance audit` · `audit my rules` · `rule drift`. ## Preconditions These platform realities shape every step; work with them, never around them: - **Rules live in context, not in a file.** Standing rules sit inside the in-context userMemories block. Parse that block at runtime. If no userMemories block (or no standing-rules content within it) is present in the conversation, say exactly that and stop — do not invent rules and do not mine chat history to reconstruct them. - **`conversation_search` is a literal keyword matcher.** Probes must be short phrases the *user* would actually type when a rule is broken — corrections and complaints in user-voice — never descriptions of the rule itself. - **`recent_chats` caps at 20 chats per call.** Default audit window is