Stop yamllint and markdownlint linting other sessions' worktrees

.markdownlint-cli2.yaml

@@ -19,3 +19,10 @@ globs:
 ignores:
   - node_modules/**
   - .venv/**
+  # Other Claude Code sessions' git worktrees, for the reason spelled out in
+  # .yamllint.yaml. `**/*.md` matches dot directories here, so a second
+  # checkout doubles the file count — measured at 114 files against 57 tracked.
+  # It has passed so far only because a worktree holds a copy of this
+  # repository's own markdown; a branch under review with one lint error in it
+  # would fail `make validate` on main, pointing at a file main does not have.
+  - .claude/worktrees/**

.yamllint.yaml

@@ -15,6 +15,24 @@ ignore: |
   stacks/observability/snmp-exporter/.rendered/
   node_modules/
   .venv/
+  # Other Claude Code sessions' git worktrees. Each one is a whole second
+  # checkout, so without this yamllint lints another branch's copy of every
+  # file here — including the ciphertext the entries above exist to exempt.
+  # Those patterns contain a slash, which anchors them to the repository root
+  # the way .gitignore does, so `secrets/*.sops.yaml` does not match
+  # `.claude/worktrees/<branch>/secrets/observability.sops.yaml`. Measured: 23
+  # findings, none of them in this tree.
+  #
+  # It is a LOCAL false red and CI never reproduces it, which is the worst
+  # shape for a check to have: `make validate` fails on the deploy host for a
+  # clean change, and an operator who learns to dismiss that has stopped
+  # reading the one command that gates a deploy.
+  #
+  # This list has to be maintained by hand because yamllint does not read
+  # .gitignore. editorconfig-checker is immune to the same problem for a
+  # structural reason — scripts/lint.sh hands it a `git ls-files` list rather
+  # than letting it walk the tree. See the comment there.
+  .claude/worktrees/
 
 rules:
   line-length:

scripts/lint.sh

@@ -31,6 +31,32 @@
 # linter it cannot reach, CI may not, because a linter skipped there is one
 # nobody will ever run.
 #
+# WHAT EACH LINTER LOOKS AT, AND WHY THAT IS NOT OBVIOUS
+#
+# Three of these decide their own file list, and none of them reads .gitignore.
+# That only matters because a workstation has trees a CI checkout does not —
+# rendered config, certificates, backups, and another session's git worktree
+# under .claude/worktrees/, which is an entire second copy of this repository.
+# A linter that walks the tree finds all of it and fails locally for a change
+# that is clean, while CI stays green. Green where nothing runs and red where
+# everything does is the wrong way round, and it is how a check stops being
+# read.
+#
+# So, measured rather than assumed:
+#
+#   - yamllint walks `.`, so its exclusions live in .yamllint.yaml
+#   - markdownlint-cli2 globs `**/*.md`, which DOES match dot directories, so
+#     its exclusions live in .markdownlint-cli2.yaml
+#   - shellcheck is handed scripts/*.sh below — one literal glob, cannot wander
+#   - actionlint reads only <repo root>/.github/workflows. Verified with
+#     -verbose: it lints 2 files with a worktree present, not 4
+#   - editorconfig-checker is handed a `git ls-files` list built below, so it
+#     sees tracked files and nothing else
+#
+# The last of those is the shape to copy when this comes up again: a list built
+# from git cannot drift, whereas the first two carry a list of exclusions that
+# has to be extended by hand every time a new ignored directory appears.
+#
 # --skips-file follows seed-validation-env.sh's contract: the caller owns the
 # path and its lifetime. scripts/validate.sh passes an mktemp it removes on exit
 # and adds the line count to its SKIPPED counter. Without it a skip here would