diff --git a/.claude/skills/running-tend/SKILL.md b/.claude/skills/running-tend/SKILL.md
index d148b36b63b7..22395b1673ff 100644
--- a/.claude/skills/running-tend/SKILL.md
+++ b/.claude/skills/running-tend/SKILL.md
@@ -91,3 +91,49 @@ These tasks run as Step 3 of the bundled weekly skill (only when
   the fix has merged or the upstream problem has been addressed, close the issue
   with a short comment citing the resolution (e.g., "Resolved by #NNNN —
   closing"). Applies to any issue where `author.login == prql-bot`.
+
+## Parked issues — don't open a fix PR
+
+Some issues are open to hold a design discussion, not to hold an unclaimed bug.
+Before opening a PR that fixes an issue, run both checks below — each catches a
+case the other misses.
+
+**Search prior PRs by the symbol the fix touches, not the issue number.** The
+rejected approach is usually a closed PR on the same call site, filed before the
+issue existed and so citing no issue number at all:
+
+```sh
+gh pr list --state all --limit 20 --search "<function or symbol name>" \
+  --json number,title,state --jq '.[] | "\(.state) #\(.number) \(.title)"'
+# the list carries no reason — read a closed PR's own comments before dismissing it:
+gh pr view <pr-number> --json comments,reviews \
+  --jq '[.comments[], .reviews[]] | map({author: .author.login, body: .body})'
+```
+
+Searching `6166` returns #6206 and #6164 and a stray dependency bump, but not
+#6147 — the rejected attempt this section exists to catch, opened two days
+before the issue. Searching `fold_module_def_stmt`, the call site all three PRs
+edit, returns every one of them.
+
+**Read the issue's comment bodies, not the count.** The 2026-08-18 nightly (run
+`32109908394`) projected `COMMENTS: \(.comments|length)`, so a prior run's note
+on #6166 saying not to open another PR was never in context:
+
+```sh
+gh issue view <n> --json comments \
+  --jq '.comments[] | "\(.author.login) (\(.authorAssociation)): \(.body)"'
+```
+
+A PR closed on the _approach_ rather than the code means the semantics are still
+a maintainer's call, and a fresh PR taking that same approach is noise however
+well it is implemented. In that case investigate, reproduce, and add findings to
+the issue thread, but leave the PR until a maintainer specifically asks for one.
+
+Parked today:
+
+- **#6166** — duplicate `module` definitions. Whether re-declaring a module
+  merges or errors is a maintainer decision, and the maintainer has asked
+  directly on that thread that no further PRs be opened on it unless
+  specifically directed. #6147 and #6206 both took the merging approach and were
+  closed on it. Remove this entry once #6166 closes or a maintainer asks for a
+  fix.