diff --git a/.claude/skills/ci/SKILL.md b/.claude/skills/ci/SKILL.md
index 864ae81..5416a5d 100644
--- a/.claude/skills/ci/SKILL.md
+++ b/.claude/skills/ci/SKILL.md
@@ -18,7 +18,7 @@ Monitor GitHub Actions CI across both repos after a push. Always run in backgrou
 - `/ci` — monitor the most recent CI run on both repos for the current branch
 - `/ci pgxntool-test` — monitor pgxntool-test only
 - `/ci pgxntool` — monitor pgxntool only
-- `/ci <branch> <pgxntool-sha> <pgxntool-test-sha>` — monitor specific push SHAs (most reliable)
+- `/ci both <branch> <pgxntool-test-sha> <pgxntool-sha>` — monitor specific push SHAs (most reliable)
 
 ## Workflow
 
@@ -41,9 +41,11 @@ When pushing to both repos, always pass the SHAs to avoid a race condition where
 
 > **Race condition note**: `gh run list --branch` returns the most recent run on
 > that branch — if two pushes happen close together (e.g. two sessions pushing
-> in parallel), it may pick up the wrong run. Passing `--commit SHA` targets the
-> exact push and avoids this. When SHA is unavailable, always verify the
-> `=== BRANCHES: ===` line in the output matches the code you pushed.
+> in parallel), it may pick up the wrong run. Passing the `sha1`/`sha2`
+> positional arguments above lets the script resolve the run via `gh run list
+> --commit` internally, targeting the exact push and avoiding this — the
+> script itself has no `--commit` flag. When SHA is unavailable, always verify
+> the `=== BRANCHES: ===` line in the output matches the code you pushed.
 
 **Always use `run_in_background: true`.**
 
diff --git a/.claude/skills/ci/scripts/monitor-ci.sh b/.claude/skills/ci/scripts/monitor-ci.sh
index da19659..9917d7b 100755
--- a/.claude/skills/ci/scripts/monitor-ci.sh
+++ b/.claude/skills/ci/scripts/monitor-ci.sh
@@ -22,6 +22,17 @@
 
 set -euo pipefail
 
+# Purely positional args, no flags (see the header comment above) - catch a
+# "--commit SHA"-style invocation here instead of letting it silently
+# misparse into REPOS="--commit" (falls through to the "both" default) with
+# the SHA itself landing in BRANCH, which then polls a nonexistent branch
+# until timeout.
+if [[ "${1:-}" == --* ]]; then
+  echo "monitor-ci.sh takes positional args, not flags: [repos] [branch] [sha_pgxntool_test] [sha_pgxntool]" >&2
+  echo "Got '$1' where 'repos' was expected. See the /ci skill's Usage section." >&2
+  exit 64
+fi
+
 REPOS="${1:-both}"
 BRANCH="${2:-$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo "")}"
 SHA_TEST="${3:-}"
diff --git a/CLAUDE.md b/CLAUDE.md
index 335cc48..c32319d 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -25,11 +25,13 @@ convention in `../ai/CLAUDE.md`:
   for each repo — do not monitor them sequentially.
 - Always use the `/ci` skill (`bash .claude/skills/ci/scripts/monitor-ci.sh`)
   rather than raw `gh run`/`gh pr checks` calls — it derives the owner from
-  the current repo and monitors both. Pass the exact push SHA when
-  available — `gh run list --branch` has a race condition: if two pushes
-  land close together on the same branch (e.g., two Claude sessions pushing
-  in parallel), `--branch` may pick up the wrong run. `--commit SHA` targets
-  the exact push and avoids this.
+  the current repo and monitors both. Pass the exact push SHA(s) as
+  positional arguments when available (see the `/ci` skill for exact
+  usage — the script takes `[repos] [branch] [sha_pgxntool_test]
+  [sha_pgxntool]`, not a `--commit` flag) — `gh run list --branch` has a
+  race condition: if two pushes land close together on the same branch
+  (e.g., two Claude sessions pushing in parallel), `--branch` may pick up
+  the wrong run. An exact SHA targets the push directly and avoids this.
 - **After every monitor run, check the `=== BRANCHES: pgxntool=X
   pgxntool-test=Y ===` line** to verify the right code is under test. If the
   branches don't match what you pushed, cancel the run and re-trigger.