praise-minerlisted

# praise-miner ## Purpose Mine the user's chat history for assistant outputs that drew praise and needed zero corrections, open those threads to find the structural antecedents that preceded first-try success, and codify repeated antecedents into do-more rules. Praise is evidence about what to repeat — never a record of a user decision. An optional final step stages the rules into memory behind an explicit confirm gate. ## Triggers Run this skill when the user says `run praise miner`, `praise miner`, `what works well`, or `success patterns`. Staging (workflow step 7) runs only on a further explicit `stage rules`. ## Preconditions - Past-chats tools must be available: `conversation_search`, `recent_chats`, `read_conversation`. If any is missing, say so and stop. - Run in a chat outside any Project; project chats search only project-scoped history. - `conversation_search` is a literal keyword match — one search per term, no operators. In-thread probes use `within_conversation_id`. - `recent_chats` returns at most 20 chats per call. If ≤5 paginated calls do not cover the window, label the report `SAMPLED`. - Snippets truncate. Antecedents come only from opened threads, never from snippets. - Memory staging edits `memory_user_edits` (30-entry cap): rules must condense to ≤5 dense lines and execute only behind the confirm gate. ## Lexicon | Positive terms | Strength | |---|---| | perfect | strong | | exactly what | strong | | ship it | strong | | nailed | strong | | first