friction-auditlisted

# friction-audit ## Purpose Sweep the user's chat history over a bounded window for friction: moments where the user corrected Claude, repeated themselves, or pushed back. Cluster the hits into a behavior taxonomy, rank the patterns by weighted frequency, and propose exactly one remediation per pattern. Optionally — and only behind an explicit confirm gate — stage `MEMORY-EDIT` remediations with `memory_user_edits`. The audit is read-only by default and reports only what was actually observed; it never extrapolates. ## Triggers Run this skill when the user says any of: `run friction audit` · `what annoys me` · `find my complaints` · `frustration audit` · `friction patterns` Modifiers the user may add: `7d` (tighten the window to 7 days; default 30) and `stage fixes` (enable Workflow step 7). ## Preconditions - Tools required: `recent_chats`, `conversation_search`, `read_conversation`. Optional staging additionally requires `memory_user_edits`. - Scope: chats inside a Project are invisible from outside it. Run in a chat outside any Project for a global sweep, or inside a Project to audit only that Project. State the active scope in the report header. - `conversation_search` is a literal keyword match. Query each lexicon term verbatim; never add dates or meta-words ("conversation", "yesterday") to a query. - `recent_chats` returns at most 20 chats per call; treat 5 calls as the practical budget. - Search hits are snippets and may truncate mid-turn. Attribute text to a s