Preserve completed command assessments

2026-08-13 16:41:19 -05:00
parent 934cddc063
commit 5737f8342c
29 changed files with 864 additions and 94 deletions
@@ -110,18 +110,29 @@ fn prepare_rig_turn_for_provider(
    let mode = request_mode(&input);
    let available_tools = match mode {
        RigRequestMode::Cli => supported_cli_agent_tools,
+        RigRequestMode::CompletedCommandAssessment => Vec::new(),
        RigRequestMode::Normal | RigRequestMode::Plan | RigRequestMode::Orchestrate => {
            supported_tools
        }
    };
-    let (mut tools, mcp_tool_aliases) = tool_definitions(&available_tools, mcp_context.as_ref());
-    if matches!(mode, RigRequestMode::Cli) {
-        // History recall cannot advance a running command and is handled inline by the Rig
-        // adapter (without producing a client action that can trigger another turn). Keeping it
-        // in the CLI tool list lets the model spend its entire monitor turn recalling the prior
-        // snapshot instead of scheduling `read_shell_command_output`, so make polling the only
-        // way to inspect the active command here.
-        tools.retain(|tool| tool.name != "recall_tool_history");
+    let (mut tools, mut mcp_tool_aliases) =
+        tool_definitions(&available_tools, mcp_context.as_ref());
+    match mode {
+        RigRequestMode::Cli => {
+            // History recall cannot advance a running command and is handled inline by the Rig
+            // adapter (without producing a client action that can trigger another turn). Keeping it
+            // in the CLI tool list lets the model spend its entire monitor turn recalling the prior
+            // snapshot instead of scheduling `read_shell_command_output`, so make polling the only
+            // way to inspect the active command here.
+            tools.retain(|tool| tool.name != "recall_tool_history");
+        }
+        RigRequestMode::CompletedCommandAssessment => {
+            // The caller deliberately disables tools for the final assessment. The inline history
+            // tool is added independently of supported tool types, so remove it explicitly too.
+            tools.clear();
+            mcp_tool_aliases.clear();
+        }
+        RigRequestMode::Normal | RigRequestMode::Plan | RigRequestMode::Orchestrate => {}
    }
    let system_prompt = build_system_prompt(&input, &tools, &global_rules, mode);

@@ -243,6 +254,20 @@ fn input_message(input: AIAgentInput) -> Option<ConversationMessage> {
            };
            (text, image_parts(&context))
        }
+        AIAgentInput::CommandCompletionAssessment {
+            prompt,
+            context,
+            completed_command,
+        } => (
+            format!(
+                "[Completed command: {}]\n[Command ID: {}]\n[Final terminal output:\n{}\n]\n{}",
+                completed_command.command,
+                completed_command.block_id,
+                completed_command.grid_contents,
+                prompt
+            ),
+            image_parts(&context),
+        ),
        AIAgentInput::ActionResult { .. } => return None,
        AIAgentInput::AutoCodeDiffQuery { query, .. } => (query, Vec::new()),
        AIAgentInput::ResumeConversation { .. } => (
@@ -389,7 +414,8 @@ fn input_user_query(input: &AIAgentInput) -> Option<String> {
    match input {
        AIAgentInput::UserQuery { query, .. } => Some(query.clone()),
        AIAgentInput::InvokeSkill { skill, .. } => Some(format!("/{}", skill.name)),
-        AIAgentInput::AutoCodeDiffQuery { .. }
+        AIAgentInput::CommandCompletionAssessment { .. }
+        | AIAgentInput::AutoCodeDiffQuery { .. }
        | AIAgentInput::ResumeConversation { .. }
        | AIAgentInput::InitProjectRules { .. }
        | AIAgentInput::CreateEnvironment { .. }
@@ -414,9 +440,17 @@ enum RigRequestMode {
    Plan,
    Orchestrate,
    Cli,
+    CompletedCommandAssessment,
 }

 fn request_mode(inputs: &[AIAgentInput]) -> RigRequestMode {
+    if inputs
+        .iter()
+        .any(|input| matches!(input, AIAgentInput::CommandCompletionAssessment { .. }))
+    {
+        return RigRequestMode::CompletedCommandAssessment;
+    }
+
    for input in inputs {
        // A direct-provider follow-up carries an LRC snapshot as an action result rather than
        // as a user query with `running_command`. Treat that result as a CLI-monitor turn so the
@@ -747,6 +781,9 @@ fn build_system_prompt(
        RigRequestMode::Cli => prompt.push_str(
            "## Running Command Monitor\nThis turn concerns a running or just-finished shell command. Act as its dedicated monitor while still following the user's steering messages. Use the command ID from the tool result for every read/write operation. If the result says the command finished, report its outcome and stop polling. If it says the command is still running, the next assistant output MUST be a tool call: use `read_shell_command_output` with a short delay, or use `interrupt_shell_command` immediately when the user's explicit stop condition is met. Do not end a still-running monitor turn with prose, a status message, or a request for the user to say continue. Never choose a poll interval that crosses a user-specified deadline or stop condition. After an interrupt, poll briefly to verify the outcome. Never start a duplicate command merely to check its state, and never report completion while a result says it is still running.\n\n",
        ),
+        RigRequestMode::CompletedCommandAssessment => prompt.push_str(
+            "## Completed Command Assessment\nThe monitored command has finished. Use its command, command ID, final terminal output, and the assessment instruction in the latest hidden input to provide the final user-facing outcome. Do not continue polling, request more terminal output, or call tools.\n\n",
+        ),
    }
    prompt.push_str("## Available Tools\n");
    if tools.is_empty() {
@@ -248,6 +248,77 @@ fn rig_prompt_requires_follow_through_without_manual_continue_prompts() {
    assert!(prompt.contains("After each tool result, choose and perform the next necessary step"));
 }

+#[test]
+#[allow(deprecated)]
+fn completed_command_assessment_uses_root_history_without_tools_or_monitor_instructions() {
+    let block_id: galaxy_terminal::model::BlockId = "completed-lrc-test".to_string().into();
+    let mcp_tool = serde_json::from_value(serde_json::json!({
+        "name": "echo",
+        "description": "Echo input",
+        "inputSchema": {"type": "object"}
+    }))
+    .unwrap();
+    let mut params = RequestParams::new_for_test();
+    params.root_task_id = Some("root-task".to_string());
+    params.message_history = vec![galaxy_agent_core::ConversationMessage {
+        role: MessageRole::User,
+        content: MessageContent::Text("Prior root conversation".to_string()),
+    }];
+    params.mcp_context = Some(MCPContext {
+        resources: Vec::new(),
+        tools: Vec::new(),
+        servers: vec![MCPServer {
+            id: "11111111-1111-4111-8111-111111111111".to_string(),
+            name: "Echo".to_string(),
+            description: String::new(),
+            resources: Vec::new(),
+            tools: vec![mcp_tool],
+        }],
+    });
+    params.input = vec![AIAgentInput::CommandCompletionAssessment {
+        prompt: "Report the final result to the user.".to_string(),
+        context: Arc::from([]),
+        completed_command: crate::ai::agent::RunningCommand {
+            command: "bash loop.sh".to_string(),
+            block_id,
+            grid_contents: "All 42 checks passed.".to_string(),
+            cursor: String::new(),
+            requested_command_id: None,
+            is_alt_screen_active: false,
+        },
+    }];
+
+    let prepared = prepare_rig_turn(
+        &config(),
+        params,
+        vec![ToolType::RunShellCommand, ToolType::CallMcpTool],
+        vec![ToolType::ReadShellCommandOutput],
+    );
+    let prompt = prepared.request.system_prompt.expect("system prompt");
+
+    assert_eq!(prepared.task_id, "root-task");
+    assert_eq!(prepared.user_query, None);
+    assert!(prepared.request.tools.is_empty());
+    assert!(prepared.mcp_tool_aliases.is_empty());
+    assert!(prompt.contains("## Completed Command Assessment"));
+    assert!(prompt.contains("No tools are available"));
+    assert!(!prompt.contains("## Running Command Monitor"));
+    assert!(!prompt.contains("next assistant output MUST be a tool call"));
+    assert!(prepared.persistent_messages.iter().any(|message| matches!(
+        &message.content,
+        MessageContent::Text(text) if text == "Prior root conversation"
+    )));
+    assert!(prepared.persistent_messages.iter().any(|message| matches!(
+        &message.content,
+        MessageContent::Text(text)
+            if text.contains("[Completed command: bash loop.sh]")
+                && text.contains("[Command ID: completed-lrc-test]")
+                && text.contains("[Final terminal output:\nAll 42 checks passed.")
+                && text.contains("Report the final result to the user.")
+    )));
+    assert_eq!(prepared.request.messages, prepared.persistent_messages);
+}
+
 #[test]
 fn lrc_snapshot_follow_up_uses_the_cli_monitor_prompt_and_tools() {
    let block_id: galaxy_terminal::model::BlockId = "precmd-lrc-test".to_string().into();