diff --git a/.claude/skills/campaign-issues/SKILL.md b/.claude/skills/campaign-issues/SKILL.md
index 3ce928b..ea3f222 100644
--- a/.claude/skills/campaign-issues/SKILL.md
+++ b/.claude/skills/campaign-issues/SKILL.md
@@ -17,14 +17,14 @@ All commands use `gh` against `prdlk/leetcode` and fall back to `gh auth token`.
 
 One issue per LeetCode problem, attached as a sub-issue of its topic.
 
-- **Title:** `LC <num> · <Title> · <Difficulty> · <set>`
-  - `<Difficulty>` ∈ `Easy | Medium | Hard` (LeetCode's own rating — do not invent).
-  - `<set>` ∈ `core | optional | deferred` (see set semantics below).
-  - Separators are middle dots `·` (U+00B7), single spaces around each.
-- **Labels** (four, all required):
-  - `problem`
-  - `diff:easy` | `diff:medium` | `diff:hard` — matches `<Difficulty>`.
-  - `set:core` | `set:optional` | `set:deferred` — matches `<set>`.
+- **Title:** `LC <num> · <Title>` — the LC number and the problem's name, nothing
+  else. Separator is a middle dot `·` (U+00B7) with single spaces. Difficulty and
+  set are **labels only**; never repeat them in the title (they used to be a `·
+  Hard · core` suffix — that is gone, and `catalog.ts` no longer parses it).
+- **Labels** (four, all required — these are what the Worker reads):
+  - `problem` — marks the sub-issue as curriculum; without it the reconcile skips it.
+  - `diff:easy` | `diff:medium` | `diff:hard` — LeetCode's own rating, do not invent.
+  - `set:core` | `set:optional` | `set:deferred` — see set semantics below.
   - `phase:N` — the **topic's** phase (same phase label the topic issue carries).
 - **Milestone:**
   - `core` / `optional` → the topic's phase milestone.
@@ -51,7 +51,7 @@ Create it:
 
 ```sh
 gh issue create --repo prdlk/leetcode \
-  --title "LC 42 · Trapping Rain Water · Hard · core" \
+  --title "LC 42 · Trapping Rain Water" \
   --label problem --label set:core --label diff:hard --label phase:1 \
   --milestone "Phase I — Linear Structures" \
   --body "https://leetcode.com/problems/trapping-rain-water/
@@ -146,6 +146,23 @@ gh issue list --repo prdlk/leetcode --label topic --state all --limit 30 \
 
 ## Gotchas
 
+- **Check the LC number is not already in the curriculum BEFORE creating.** D1 keys
+  `problems` on `lc_number` (`ON CONFLICT(lc_number)`), so two issues for one LC number
+  silently fight over a single row — whichever the reconcile walks last wins, and the
+  loser's issue number ends up orphaned in D1. A problem can legitimately move topic or
+  set; that is a relabel/re-parent of the *existing* issue, never a second issue:
+
+  ```sh
+  gh issue list --repo prdlk/leetcode --label problem --state all --limit 300 \
+    --search "LC 42" --json number,title,labels,state
+  ```
+
+  If a row already exists, change the existing issue instead: swap its `set:*` label,
+  move the sub-issue to the new topic, and fix its milestone. Then reconcile the board
+  (`project-fields` skill). If a duplicate did get created and one is deleted, the
+  surviving issue number and set only reach D1 on the next catalog reconcile — SRS
+  columns (`stage`, `next_review`) are preserved, since they are the real attempt history.
+
 - **`-F` vs `-f`:** `sub_issue_id` must be an integer; `-f` sends a string and the API
   rejects it with `is not of type integer`.
 - **id, not number:** the sub-issue payload wants the child's `.id` (a large opaque
diff --git a/AGENTS.md b/AGENTS.md
index 85a139f..96464e9 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -2,7 +2,7 @@
 
 ## Project Overview
 
-LeetCode interview-prep campaign (8 weeks, 2026-08-17 → Oct 11) run as a repo: solution files under `work/`, a Blume docs site under `apps/docs/`, GitHub Actions that reconcile issues and publish docs, and a Cloudflare Worker (`apps/api/`) that runs the spaced-repetition system (SRS) — daily digest email, one-tap logging, weekly review issues, live charts. Issues model the curriculum: ~24 `topic` issues own ~161 `problem` sub-issues (`set:core|optional|deferred`, `diff:*`); milestones are phases; a user-level GitHub Project ("Interview Prep", #2) mirrors review state.
+LeetCode interview-prep campaign (8 weeks, 2026-08-17 → Oct 11) run as a repo: solution files under `work/`, a Blume docs site under `apps/docs/`, GitHub Actions that reconcile issues and publish docs, and a Cloudflare Worker (`apps/api/`) that runs the spaced-repetition system (SRS) — daily digest email, one-tap logging, weekly review issues, live charts. Issues model the curriculum: ~24 `topic` issues (`Topic NN — Name`) own ~161 `problem` sub-issues titled `LC <num> · <Name>`, with difficulty and set carried **only** by labels (`set:core|optional|deferred`, `diff:*`) — never by the title; milestones are phases; a user-level GitHub Project ("Interview Prep", #2) mirrors review state.
 
 ## Architecture & Data Flow