tests/entrypoint_test.sh

@@ -56,6 +56,26 @@ SH
   chmod +x "$dir/codex"
 }
 
+make_fake_transcript_codex() {
+  local dir="$1"
+
+  cat > "$dir/codex" <<'SH'
+#!/bin/sh
+last_message_file=""
+while [ "$#" -gt 0 ]; do
+  if [ "$1" = "--output-last-message" ]; then
+    shift
+    last_message_file="$1"
+  fi
+  shift
+done
+
+printf '%s\n' 'codex' 'final answer' 'tokens used' '123' 'final answer'
+printf '%s\n' 'final answer' > "$last_message_file"
+SH
+  chmod +x "$dir/codex"
+}
+
 encoded_json() {
   printf '%s' "$1" | base64 | tr -d '\n'
 }
@@ -269,6 +289,20 @@ test_codex_output_truncation() {
   rm -rf "$tmpdir"
 }
 
+test_last_message_output_without_transcript_duplicate() {
+  local tmpdir
+  tmpdir="$(mktemp -d)"
+  make_fake_transcript_codex "$tmpdir"
+
+  run_entrypoint "$tmpdir" OAUTH="$(encoded_json '{}')" MODEL="gpt-test" PROMPT="hello"
+
+  [[ "$(grep -c "final answer" "$tmpdir/stdout")" -eq 1 ]] || die "stdout printed final answer more than once"
+  grep -q "status=completed" "$tmpdir/github-output" || die "last message completed status"
+  [[ "$(grep -c "final answer" "$tmpdir/github-output")" -eq 1 ]] || die "github output included transcript duplicate"
+  ! grep -q "tokens used" "$tmpdir/github-output" || die "github output included codex transcript"
+  rm -rf "$tmpdir"
+}
+
 test_invalid_execution_config_uses_defaults() {
   local tmpdir
   tmpdir="$(mktemp -d)"
@@ -299,6 +333,7 @@ test_missing_codex_command_output
 test_codex_timeout_output
 test_codex_signal_output
 test_codex_output_truncation
+test_last_message_output_without_transcript_duplicate
 test_invalid_execution_config_uses_defaults
 
 echo "entrypoint tests passed"