skill: fold the spool — section 596 gets its number on master

.claude/skills/ci-gates/SKILL.md

@@ -17316,3 +17316,49 @@ One more line, because it cost a rewrite. This section was first written into an
 heredoc, so every backtick in it ran as a command substitution and the section landed with
 its code spans eaten -- `check` became nothing, and one of them printed
 `command not found: check`. Quote the delimiter: `<<'MD'`, never `<<MD`.
+
+## 596. I said master was green after reading two workflows of sixty-three
+
+Twice in one night a report of mine ended with a line like
+
+    master зелёный: перепись PASS, cli-tri и Untrusted Input Gate — success
+
+Both readings were true. The sentence was not. The repository has **63 active
+workflows**, and I had named two — the two I had happened to break and repair
+that evening, which is exactly why they were the ones in front of me.
+
+Measured properly, latest run on master for every active workflow:
+
+    success        40
+    failure        11
+    never ran      12
+
+Among the eleven was `Issue Gate`, which supplies `check-linked-issue` — one of
+the four contexts required to merge.
+
+Two separate errors, and the second is the one worth keeping.
+
+**The sample was the ones I had touched.** A gate you just repaired is the most
+available evidence and the least representative: its greenness is a statement
+about your own afternoon, not about the branch.
+
+**The obvious wider read is still a window, not a population.**
+`gh run list --branch master --limit 100` returns the last hundred RUNS, and a
+hundred runs held only **22 distinct workflows** — a third of them. The other 41
+had not run recently enough to appear, and a workflow that has not run is not a
+workflow that passed. The population lives in
+`/actions/workflows`, and the per-workflow question has to be asked once each:
+
+    gh api 'repos/OWNER/REPO/actions/workflows?per_page=100' | jq '.workflows[] | select(.state=="active")'
+    # then, per id:
+    gh api "repos/OWNER/REPO/actions/workflows/$id/runs?branch=master&per_page=1"
+
+That read also separates a third answer the window cannot express: **12 workflows
+have never run on master at all.** They are neither green nor red, and rolling
+them into either number is the failure this file is full of.
+
+The correction was not self-generated. It came from noticing a neighbour's issue
+titled *"Every workflow red on master"* while looking for something else — a
+title that could not both be true and leave my sentence standing. Two claims that
+cannot both hold are the cheapest instrument there is, and the only reason this
+one fired is that I read a list I did not need.

docs/now/2026-09-06-the-first-fold-a-spooled-lesson-gets-its-number-on-master.md

@@ -0,0 +1,7 @@
+# NOW -- the first fold: a spooled lesson gets its number on master (2026-09-06)
+
+## the first fold: a spooled lesson gets its number on master (Refs #3236)
+
+- tri skill fold run on master for the first time. One spooled lesson became section 596, numbered against the file as it stands, and the spool file was deleted in the same commit.
+- The diff is the point: 46 insertions and ZERO deletions in SKILL.md, plus one deleted spool file. That is the structural signature -- one section added per spool file removed -- which is what makes a direct append distinguishable from a fold in a branch diff.
+- This is the step the whole spool exists for, and it had never been run from master. Numbering happens here, once, on one branch, with the file in front of it -- so two branches cannot choose the same number.