diff --git a/graph/decisions/live-series-card-update-agent-definition-created-9e1897.md b/graph/decisions/live-series-card-update-agent-definition-created-9e1897.md
new file mode 100644
index 00000000000..8d0c5ad566a
--- /dev/null
+++ b/graph/decisions/live-series-card-update-agent-definition-created-9e1897.md
@@ -0,0 +1,24 @@
+---
+id: 9e1897e8-a5ec-4297-ba85-941dcfe304b9
+type: decision
+title: "Live series card update agent definition created"
+tags: [paper-dynasty, claude-code, agents, architecture]
+importance: 0.7
+confidence: 0.8
+created: "2026-02-16T20:15:49.827270+00:00"
+updated: "2026-02-16T20:15:49.827270+00:00"
+---
+
+Created .claude/agents/live-series-card-update-agent.md for Paper Dynasty card-creation project. Mirrors the retrosheet agent pattern with live-series-specific differences:
+
+1. **7-step pipeline** (vs retrosheet's 8): No separate position validation step — live series handles it inline via --pull-fielding.
+2. **Data prerequisite gate**: Checks for 8 FanGraphs CSV files before starting (vlhp/vrhp/vlhh/vrhh basic+rate).
+3. **PotM variant support**: Separate parameter table for promo cards with --description and --ignore-limits flags.
+4. **Cardset lookup by name**: groundball_b validation looks up cardset ID by name (live series uses names, retrosheet uses IDs).
+5. **Same shared patterns**: Minimal tool set (Bash/Read/Grep/Glob/AskUserQuestion), Sonnet model, dev/prod S3 guard, environment verification, structured report output.
+
+Updated paper-dynasty skill SKILL.md in 3 places to route live series to agent (workflow table, file structure comment, context loading table). Workflow doc retained as reference.
+
+Both card generation workflows (retrosheet + live series) are now agents. Skill retains: queries, gauntlet, packs, bot troubleshooting, reference docs, custom cards.
+
+File: /mnt/NV2/Development/paper-dynasty/card-creation/.claude/agents/live-series-card-update-agent.md