decision-ledgerlisted

# Decision Ledger ## Purpose Decisions accumulate across chat history ("we're going with X", "Y is locked", "Z is rejected"), scattered over dozens of conversations and sometimes reversed or contradicted later. This skill harvests that decision language, applies a provenance gate so only human commitments count, normalizes qualifying entries into one dated ledger with status and source links, conflict-checks within domains, and writes the result to `DECISIONS-GLOBAL.md` as a downloadable file. The ledger records what the user actually decided, never what the assistant merely recommended. That distinction drives everything below. ## Triggers Run this skill when the user says any of: `run decision ledger`, `consolidate decisions`, `harvest decisions`, `decision ledger`, or otherwise asks to gather, compile, or audit the decisions they have made across past chats. The user may add modifiers: - A time window ("last 30 days", "past six months"): overrides the 90-day default in step 1. - An uploaded prior `DECISIONS-GLOBAL.md` in `/mnt/user-data/uploads/`: activates merge mode (see Guardrails). ## Preconditions - The past-chats tools (`conversation_search`, `recent_chats`, `read_conversation`) must be available. If they are missing, tell the user this skill needs the "Search and reference past chats" capability enabled, and stop. - File creation must be available to write the ledger to `/mnt/user-data/outputs/`. - Scope: the past-chats tools only see conversations in the c