learn-evallisted

<objective> Evaluate all existing instincts by scoring them against recent event logs, apply confidence decay, execute status transitions, prune stale candidates, merge near-duplicates, and produce a scorecard. </objective> ## Instructions You are evaluating **existing** instincts — not extracting new ones. This is the scoring and lifecycle pass that keeps the instinct system healthy. ### Step 1: Load Instincts Read all `INS-*.md` files from the global scope and optionally from a project scope: ```bash GLOBAL_DIR="$HOME/.claude/instincts/global" ls "$GLOBAL_DIR"/INS-*.md 2>/dev/null | head -50 ``` If `--project <slug>` was passed, also load from: ```bash PROJECT_DIR="$HOME/.claude/instincts/projects/<slug>" ls "$PROJECT_DIR"/INS-*.md 2>/dev/null | head -50 ``` If `--global` was passed, only evaluate global instincts (skip project scopes). For each instinct file, read the full content and parse the YAML frontmatter. Extract these fields: - `id`, `name`, `status`, `confidence` - `score.successes`, `score.failures`, `score.confirmations`, `score.corrections` - `score.last_applied_at`, `score.half_life_days`, `score.decay_floor` - `signals.trigger_patterns` (list of strings) - `evidence` (list of evidence entries) - `created_at`, `updated_at` ### Step 2: Load Recent Events Read event logs from the last 14 days: ```bash EVENTS_DIR="$HOME/.claude/instincts/events" CURRENT_MONTH=$(date -u +%Y-%m) PREV_MONTH=$(date -u -v-1m +%Y-%m 2>/dev/null || date -u -d '1 month ago' +%