Make dogfood review gate handle terminal attention states

Author: andyrewlee

internal/cli/assistant_dogfood_script_test.go

@@ -74,6 +74,7 @@ esac
 	fakeDX := filepath.Join(fakeBinDir, "assistant-dx.sh")
 	writeExecutable(t, fakeDX, `#!/usr/bin/env bash
 set -euo pipefail
+printf '%s\n' "$*" >> "${AMUX_DX_CALL_LOG:?missing AMUX_DX_CALL_LOG}"
 cmd="${1:-}"
 sub="${2:-}"
 case "$cmd $sub" in
@@ -88,6 +89,12 @@ case "$cmd $sub" in
     fi
     jq -cn --arg ws "$ws_id" '{ok:true,command:"workspace.create",status:"ok",summary:"Workspace created.",next_action:"",suggested_command:"",data:{id:$ws,assistant:"codex"},quick_actions:[]}'
     ;;
+  "status --workspace")
+    jq -cn '{ok:true,command:"status",status:"attention",summary:"Implement pass reached bounded terminal state.",next_action:"Proceed to review.",suggested_command:"",data:{task:{overall_status:"partial"}},quick_actions:[]}'
+    ;;
+  "review --workspace")
+    jq -cn '{ok:true,command:"review",status:"ok",summary:"Review completed.",next_action:"Continue.",suggested_command:"",data:{},quick_actions:[]}'
+    ;;
   *)
     jq -cn '{ok:true,command:"noop",status:"ok",summary:"ok",next_action:"",suggested_command:"",data:{},quick_actions:[]}'
     ;;
@@ -106,9 +113,12 @@ esac
 	env := os.Environ()
 	env = withEnv(env, "PATH", fakeBinDir+":"+os.Getenv("PATH"))
 	env = withEnv(env, "AMUX_ASSISTANT_DOGFOOD_DX_SCRIPT", fakeDX)
+	env = withEnv(env, "AMUX_DX_CALL_LOG", filepath.Join(fakeBinDir, "dx-calls.log"))
 	env = withEnv(env, "AMUX_ASSISTANT_DOGFOOD_CHANNEL_EPHEMERAL_AGENT", "false")
 	env = withEnv(env, "AMUX_ASSISTANT_DOGFOOD_CHANNEL_REQUIRE_PROOF", "false")
 	env = withEnv(env, "AMUX_ASSISTANT_DOGFOOD_REQUIRE_CHANNEL_EXECUTION", "false")
+	env = withEnv(env, "AMUX_ASSISTANT_DOGFOOD_REVIEW_GATE_TIMEOUT_SECONDS", "1")
+	env = withEnv(env, "AMUX_ASSISTANT_DOGFOOD_REVIEW_GATE_POLL_SECONDS", "0")
 	cmd.Env = env
 	out, err := cmd.CombinedOutput()
 	if err != nil {
@@ -127,4 +137,27 @@ esac
 	if strings.Contains(string(out), "failed to resolve ws1 id from project_add") {
 		t.Fatalf("script still relied on project_add workspace id:\n%s", string(out))
 	}
+
+	callLogRaw, err := os.ReadFile(filepath.Join(fakeBinDir, "dx-calls.log"))
+	if err != nil {
+		t.Fatalf("read dx call log: %v", err)
+	}
+	callLog := string(callLogRaw)
+	if strings.Contains(callLog, "workflow dual") {
+		t.Fatalf("dogfood should not invoke removed workflow command, got:\n%s", callLog)
+	}
+	if !strings.Contains(callLog, "start --workspace ws-primary") || !strings.Contains(callLog, "--allow-new-run") {
+		t.Fatalf("dogfood should run implement step with --allow-new-run, got:\n%s", callLog)
+	}
+	if !strings.Contains(callLog, "status --workspace ws-primary --assistant codex") {
+		t.Fatalf("dogfood should gate review on workspace status checks, got:\n%s", callLog)
+	}
+	if !strings.Contains(callLog, "review --workspace ws-primary") {
+		t.Fatalf("dogfood should run review step for primary workspace, got:\n%s", callLog)
+	}
+	for _, line := range strings.Split(callLog, "\n") {
+		if strings.Contains(line, "review --workspace ws-primary") && strings.Contains(line, "--no-monitor") {
+			t.Fatalf("dogfood review step should monitor to terminal status, got:\n%s", callLog)
+		}
+	}
 }

skills/amux/scripts/assistant-dogfood.sh

@@ -17,7 +17,7 @@ Runs a real Assistant/amux dogfood flow end-to-end:
   - continue coding
   - create second workspace + start
   - terminal run + logs
-  - workflow dual
+  - bounded implement + review pass
   - git ship
   - status
 EOF
@@ -219,6 +219,38 @@ run_dx() {
   printf '%s\t%s\n' "$slug" "$(cat "$status_file")"
 }
 
+wait_for_workspace_ready_for_review() {
+  local workspace="$1"
+  local assistant="$2"
+  local timeout_s="${AMUX_ASSISTANT_DOGFOOD_REVIEW_GATE_TIMEOUT_SECONDS:-240}"
+  local poll_s="${AMUX_ASSISTANT_DOGFOOD_REVIEW_GATE_POLL_SECONDS:-5}"
+  local start_ts elapsed attempt status overall json_file
+  start_ts="$(date +%s)"
+  attempt=0
+
+  while true; do
+    attempt=$((attempt + 1))
+    run_dx "dual_gate_status_${attempt}" status --workspace "$workspace" --assistant "$assistant"
+    json_file="$REPORT_DIR/dual_gate_status_${attempt}.json"
+    status="$(jq -r '.status // ""' <"$json_file" 2>/dev/null || true)"
+    overall="$(jq -r '.data.task.overall_status // ""' <"$json_file" 2>/dev/null || true)"
+
+    if [[ "$status" == "ok" || "$overall" == "completed" || "$overall" == "session_exited" || "$overall" == "partial" || "$overall" == "partial_budget" || "$overall" == "timed_out" ]]; then
+      return 0
+    fi
+    if [[ "$status" == "needs_input" ]]; then
+      run_dx "dual_gate_continue_${attempt}" continue --workspace "$workspace" --assistant "$assistant" --text "Continue from current state and finish this run with a concise completion summary." --enter --wait-timeout 70s --idle-threshold 10s
+    fi
+
+    elapsed="$(( $(date +%s) - start_ts ))"
+    if (( elapsed >= timeout_s )); then
+      echo "timed out waiting for workspace $workspace implement run to reach terminal state before review (status=$status overall=$overall)" >&2
+      return 1
+    fi
+    sleep "$poll_s"
+  done
+}
+
 run_assistant_local_ping() {
   local slug="$1"
   local session_id="$2"
@@ -439,7 +471,11 @@ run_dx terminal_run_ws1 terminal run --workspace "$WS1_ID" --text "go run main.g
 sleep 1
 run_dx terminal_logs_ws1 terminal logs --workspace "$WS1_ID" --lines 40
 
-run_dx workflow_dual_ws1 workflow dual --workspace "$WS1_ID" --implement-assistant "$ASSISTANT" --implement-prompt "Append one concise mobile-coding tip to README.md and proceed even if there are unrelated uncommitted changes." --review-assistant "$ASSISTANT" --review-prompt "Review for clarity and correctness." --max-steps 1 --turn-budget 100 --wait-timeout 70s --idle-threshold 10s --auto-continue-impl true
+run_dx dual_impl_ws1 start --workspace "$WS1_ID" --assistant "$ASSISTANT" --prompt "Append one concise mobile-coding tip to README.md and proceed even if there are unrelated uncommitted changes." --max-steps 1 --turn-budget 100 --wait-timeout 70s --idle-threshold 10s --allow-new-run
+if ! wait_for_workspace_ready_for_review "$WS1_ID" "$ASSISTANT"; then
+  exit 1
+fi
+run_dx dual_review_ws1 review --workspace "$WS1_ID" --assistant "$ASSISTANT" --prompt "Review for clarity and correctness." --max-steps 1 --turn-budget 100 --wait-timeout 70s --idle-threshold 10s
 
 run_dx git_ship_ws1 git ship --workspace "$WS1_ID" --message "dogfood: scripted assistant pass"
 run_dx status_ws1 status --workspace "$WS1_ID" --capture-agents 8 --capture-lines 80