ambiguity-preemptlisted

# Ambiguity Preempt ## Purpose Audit past conversations for two failure shapes: clarify-loops (the assistant had to ask "which one?") and wrong-guess-then-corrected sequences (the assistant picked an instance, the user corrected it). Cluster the incidents by ambiguity axis (machine, account, repo, path, shell, surface, other), then draft standing disambiguation rules — default + inline-statement + ask-only-if-destructive — condensed into a memory-paste block the user can adopt, with optional gated staging into memory edits. ## Triggers Run this skill when the user says any of: - `run ambiguity audit` - `disambiguation rules` - `stop asking which` - `ambiguity preempt` `stage rules` triggers only the optional staging step (Workflow step 7), and only after an audit in the current conversation has produced a rules table. ## Preconditions - Requires the past-chats tools: `conversation_search`, `recent_chats`, `read_conversation`. If any is unavailable, say so and stop. - Chats inside Projects are invisible from outside Projects (and vice versa). For a global audit, run in a new chat outside any Project; if scope is limited, state it in the report. - Staging (step 7 only) additionally requires the `memory_user_edits` tool, which is capped at 30 total edits — this is why rules are condensed to at most one dense line per axis. ## Lexicon (TUNABLE) `conversation_search` is a literal keyword match, not semantic. Sweep with one search per term, using each short literal phrase