diff --git a/.claude/commands/generate-detail-levels.md b/.claude/commands/generate-detail-levels.md
index 58a6e36..982115b 100644
--- a/.claude/commands/generate-detail-levels.md
+++ b/.claude/commands/generate-detail-levels.md
@@ -16,11 +16,11 @@ independent documents. Run this whenever comprehensive.txt changes.
 <size-targets>
 Character counts (verify with `wc -c prompts/align_to_love/*.txt`):
 
-| Tier | Target | Hard bounds |
-|------|--------|-------------|
+| Tier     | Target                | Hard bounds     |
+| -------- | --------------------- | --------------- |
 | standard | ~45% of comprehensive | 2500-4500 chars |
-| concise | ~20% of comprehensive | 1100-1900 chars |
-| terse | single paragraph | 550-1000 chars |
+| concise  | ~20% of comprehensive | 1100-1900 chars |
+| terse    | single paragraph      | 550-1000 chars  |
 
 Tests in both packages assert strict ordering (terse < concise < standard <
 comprehensive) plus absolute length bounds. Read the current assertions in
@@ -67,15 +67,23 @@ into flowing prose.
 1. Read `prompts/align_to_love/comprehensive.txt` in full. List its themes in order.
 2. Write standard.txt, then concise.txt, then terse.txt, each derived from
    comprehensive (not from each other, and not from the previous versions).
-3. Verify: `pre-commit run forbid-em-dashes --all-files` must pass; `wc -c`
+3. Sync `.cursor/rules/heart-centered-ai.mdc`: the body of that file mirrors
+   `comprehensive.txt` exactly (the first five lines are YAML frontmatter;
+   everything after is the prompt). Replace lines 6 onward with the new
+   comprehensive content so the always-on IDE rule stays in sync:
+   ```
+   head -5 .cursor/rules/heart-centered-ai.mdc > /tmp/mdc_header
+   cat /tmp/mdc_header prompts/align_to_love/comprehensive.txt > .cursor/rules/heart-centered-ai.mdc
+   ```
+4. Verify: `pre-commit run forbid-em-dashes --all-files` must pass; `wc -c`
    must satisfy the bounds above.
-4. Sync and test, both packages must pass:
+5. Sync and test, both packages must pass:
    - `cd python && python3 copy_prompts.py && pytest -q` (use
      `uv run --with pytest pytest -q` if pytest is missing; delete any
      `uv.lock` it creates)
    - `cd typescript && pnpm generate && pnpm test` (run `pnpm install` first
      only if `node_modules/` is missing)
-5. Refresh approximate token counts wherever docs state them (chars / 4 is a
+6. Refresh approximate token counts wherever docs state them (chars / 4 is a
    fine estimate). Find every location with `git grep -n '~[0-9]* tokens'`.
-6. Report the new character counts and what was cut from each tier.
+7. Report the new character counts and what was cut from each tier.
 </process>