pawan/tools/

agent.rs

//! Sub-agent spawning tool
//!
//! Spawns a pawan subprocess to handle a task independently.
//! This is the OMO replacement — enables multi-agent orchestration.

use super::Tool;
use crate::{PawanError, Result};
use async_trait::async_trait;
use serde_json::{json, Value};
use std::io::Write;
use std::path::PathBuf;
use std::process::Stdio;
use tokio::io::AsyncReadExt;
use tokio::process::Command;
use tracing;

use crate::subagent::SubagentHandle;

/// Tool for spawning a sub-agent (pawan subprocess)
pub struct SpawnAgentTool {
    workspace_root: PathBuf,
}

impl SpawnAgentTool {
    pub fn new(workspace_root: PathBuf) -> Self {
        Self { workspace_root }
    }

    /// Find the pawan binary — tries cargo target first, then PATH
    fn find_pawan_binary(&self) -> String {
        // Check for debug/release binary in workspace target
        for candidate in &[
            self.workspace_root.join("target/release/pawan"),
            self.workspace_root.join("target/debug/pawan"),
        ] {
            if candidate.exists() {
                return candidate.to_string_lossy().to_string();
            }
        }
        // Fall back to PATH
        "pawan".to_string()
    }
}

#[async_trait]
impl Tool for SpawnAgentTool {
    fn name(&self) -> &str {
        "spawn_agent"
    }

    fn description(&self) -> &str {
        "Spawn a sub-agent (pawan subprocess) to handle a task independently. \
         Returns the agent's response as JSON. Use this for parallel or delegated tasks."
    }

    fn mutating(&self) -> bool {
        true // Spawning agents can mutate state
    }

    fn parameters_schema(&self) -> Value {
        json!({
            "type": "object",
            "properties": {
                "prompt": {
                    "type": "string",
                    "description": "The task/prompt for the sub-agent"
                },
                "model": {
                    "type": "string",
                    "description": "Model to use (optional, defaults to parent's model)"
                },
                "timeout": {
                    "type": "integer",
                    "description": "Timeout in seconds (default: 120)"
                },
                "workspace": {
                    "type": "string",
                    "description": "Workspace directory for the sub-agent (default: same as parent)"
                },
                "retries": {
                    "type": "integer",
                    "description": "Number of retry attempts on failure (default: 0, max: 2)"
                }
            },
            "required": ["prompt"]
        })
    }

    fn thulp_definition(&self) -> thulp_core::ToolDefinition {
        use thulp_core::{Parameter, ParameterType};
        thulp_core::ToolDefinition::builder("spawn_agent")
            .description(self.description())
            .parameter(
                Parameter::builder("prompt")
                    .param_type(ParameterType::String)
                    .required(true)
                    .description("The task/prompt for the sub-agent")
                    .build(),
            )
            .parameter(
                Parameter::builder("model")
                    .param_type(ParameterType::String)
                    .required(false)
                    .description("Model to use (optional, defaults to parent's model)")
                    .build(),
            )
            .parameter(
                Parameter::builder("timeout")
                    .param_type(ParameterType::Integer)
                    .required(false)
                    .description("Timeout in seconds (default: 120)")
                    .build(),
            )
            .parameter(
                Parameter::builder("workspace")
                    .param_type(ParameterType::String)
                    .required(false)
                    .description("Workspace directory for the sub-agent (default: same as parent)")
                    .build(),
            )
            .parameter(
                Parameter::builder("retries")
                    .param_type(ParameterType::Integer)
                    .required(false)
                    .description("Number of retry attempts on failure (default: 0, max: 2)")
                    .build(),
            )
            .build()
    }

    async fn execute(&self, args: Value) -> Result<Value> {
        let prompt = args["prompt"]
            .as_str()
            .ok_or_else(|| PawanError::Tool("prompt is required for spawn_agent".into()))?;

        let timeout = args["timeout"].as_u64().unwrap_or(120);
        let model = args["model"].as_str();
        let workspace = args["workspace"]
            .as_str()
            .map(PathBuf::from)
            .unwrap_or_else(|| self.workspace_root.clone());
        let max_retries = args["retries"].as_u64().unwrap_or(0).min(2) as usize;

        // Generate unique agent ID for progress tracking
        let label: String = prompt
            .lines()
            .next()
            .unwrap_or(prompt)
            .chars()
            .take(48)
            .collect();
        let progress = SubagentHandle::start(&label, "spawn_agent", None);

        let agent_id = progress.id().to_string();
        let status_path = format!("/tmp/pawan-agent-{}.status", agent_id);
        let started_at = chrono::Utc::now().to_rfc3339();

        let pawan_bin = self.find_pawan_binary();

        for attempt in 0..=max_retries {
            let mut cmd = Command::new(&pawan_bin);
            cmd.arg("run")
                .arg("-o")
                .arg("json")
                .arg("--timeout")
                .arg(timeout.to_string())
                .arg("-w")
                .arg(workspace.to_string_lossy().to_string());

            if let Some(m) = model {
                cmd.arg("-m").arg(m);
            }

            cmd.arg(prompt);

            cmd.stdout(Stdio::piped())
                .stderr(Stdio::piped())
                .stdin(Stdio::null());

            // Write initial status
            if let Ok(mut f) = std::fs::File::create(&status_path) {
                let _ = write!(
                    f,
                    r#"{{"state":"running","prompt":"{}","started_at":"{}","attempt":{}}}"#,
                    prompt
                        .chars()
                        .take(100)
                        .collect::<String>()
                        .replace('"', "'"),
                    started_at,
                    attempt + 1
                );
            }

            let mut child = match cmd.spawn() {
                Ok(c) => c,
                Err(e) => {
                    progress.complete_err(format!("spawn failed: {e}"));
                    progress.dismiss();
                    return Err(PawanError::Tool(format!(
                        "Failed to spawn sub-agent: {e}. Binary: {pawan_bin}"
                    )));
                }
            };

            let mut stdout = String::new();
            let mut stderr = String::new();

            if let Some(mut handle) = child.stdout.take() {
                handle.read_to_string(&mut stdout).await.ok();
            }
            if let Some(mut handle) = child.stderr.take() {
                handle.read_to_string(&mut stderr).await.ok();
            }

            let status = match child.wait().await {
                Ok(s) => s,
                Err(e) => {
                    progress.complete_err(format!("wait failed: {e}"));
                    progress.dismiss();
                    return Err(PawanError::Io(e));
                }
            };

            let result = if let Ok(json_result) = serde_json::from_str::<Value>(&stdout) {
                json_result
            } else {
                json!({
                    "content": stdout.trim(),
                    "raw_output": true
                })
            };

            if status.success() || attempt == max_retries {
                // Update status file with completion
                let duration_ms = chrono::Utc::now()
                    .signed_duration_since(
                        chrono::DateTime::parse_from_rfc3339(&started_at).unwrap_or_default(),
                    )
                    .num_milliseconds();
                if let Ok(mut f) = std::fs::File::create(&status_path) {
                    let state = if status.success() { "done" } else { "failed" };
                    let _ = write!(
                        f,
                        r#"{{"state":"{}","exit_code":{},"duration_ms":{},"attempt":{}}}"#,
                        state,
                        status.code().unwrap_or(-1),
                        duration_ms,
                        attempt + 1
                    );
                }

                if status.success() {
                    progress.complete_ok();
                } else {
                    progress.complete_err(format!(
                        "exit code {}",
                        status.code().unwrap_or(-1)
                    ));
                }
                let out = json!({
                    "success": status.success(),
                    "attempt": attempt + 1,
                    "total_attempts": attempt + 1,
                    "result": result,
                    "stderr": stderr.trim(),
                    "subagent_id": progress.id(),
                    "duration_ms": duration_ms,
                });
                progress.dismiss();
                return Ok(out);
            }
            // Failed but retries remaining — continue loop
            // Failed but retries remaining — continue loop
            tracing::warn!(
                attempt = attempt + 1,
                "spawn_agent attempt failed, retrying"
            );
        }

        progress.complete_err("all retry attempts exhausted");
        progress.dismiss();
        Err(PawanError::Tool(
            "spawn_agent: all retry attempts exhausted".into(),
        ))
    }
}

/// Tool for spawning multiple sub-agents in parallel
pub struct SpawnAgentsTool {
    workspace_root: PathBuf,
}

impl SpawnAgentsTool {
    pub fn new(workspace_root: PathBuf) -> Self {
        Self { workspace_root }
    }
}

#[async_trait]
impl Tool for SpawnAgentsTool {
    fn name(&self) -> &str {
        "spawn_agents"
    }

    fn description(&self) -> &str {
        "Spawn multiple sub-agents in parallel. Each task runs concurrently and results are returned as an array."
    }

    fn parameters_schema(&self) -> Value {
        json!({
            "type": "object",
            "properties": {
                "tasks": {
                    "type": "array",
                    "items": {
                        "type": "object",
                        "properties": {
                            "prompt": {"type": "string"},
                            "model": {"type": "string"},
                            "timeout": {"type": "integer"},
                            "workspace": {"type": "string"}
                        },
                        "required": ["prompt"]
                    }
                }
            },
            "required": ["tasks"]
        })
    }

    fn thulp_definition(&self) -> thulp_core::ToolDefinition {
        use thulp_core::{Parameter, ParameterType};
        thulp_core::ToolDefinition::builder("spawn_agents")
            .description(self.description())
            .parameter(Parameter::builder("tasks").param_type(ParameterType::Array).required(true)
                .description("Array of task objects, each with prompt (required), model, timeout, workspace").build())
            .build()
    }

    fn mutating(&self) -> bool {
        true // Spawning agents can mutate state
    }

    async fn execute(&self, args: Value) -> Result<Value> {
        let tasks = args["tasks"]
            .as_array()
            .ok_or_else(|| PawanError::Tool("tasks array is required for spawn_agents".into()))?;

        for (index, task) in tasks.iter().enumerate() {
            if task.get("prompt").and_then(|p| p.as_str()).is_none() {
                return Err(PawanError::Tool(format!(
                    "spawn_agents: task {index} is missing prompt"
                )));
            }
        }

        let single_tool = SpawnAgentTool::new(self.workspace_root.clone());

        let futures: Vec<_> = tasks
            .iter()
            .enumerate()
            .map(|(index, task)| {
                let tool = &single_tool;
                let task = task.clone();
                async move {
                    let started = std::time::Instant::now();
                    let result = tool.execute(task).await;
                    (index, started.elapsed().as_millis() as u64, result)
                }
            })
            .collect();

        let results = futures::future::join_all(futures).await;

        let mut succeeded = 0usize;
        let mut failed = 0usize;
        let mut output: Vec<Value> = Vec::with_capacity(results.len());

        for (index, duration_ms, result) in results {
            match result {
                Ok(mut v) => {
                    if v.get("success").and_then(|s| s.as_bool()) == Some(true) {
                        succeeded += 1;
                    } else {
                        failed += 1;
                    }
                    if let Some(obj) = v.as_object_mut() {
                        obj.insert("index".into(), json!(index));
                        obj.insert("duration_ms".into(), json!(duration_ms));
                    }
                    output.push(v);
                }
                Err(e) => {
                    failed += 1;
                    output.push(json!({
                        "index": index,
                        "success": false,
                        "error": e.to_string(),
                        "duration_ms": duration_ms,
                    }));
                }
            }
        }

        let total = tasks.len();
        Ok(json!({
            "success": failed == 0,
            "total_tasks": total,
            "succeeded": succeeded,
            "failed": failed,
            "results": output,
        }))
    }
}
#[cfg(test)]
mod tests {
    use super::*;
    use tempfile::TempDir;
    #[test]
    fn test_spawn_agent_tool_name() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        assert_eq!(tool.name(), "spawn_agent");
    }

    #[test]
    fn test_spawn_agents_tool_name() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        assert_eq!(tool.name(), "spawn_agents");
    }

    #[test]
    fn test_spawn_agent_schema_has_prompt() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let schema = tool.parameters_schema();
        assert!(schema["properties"]["prompt"].is_object());
        assert!(schema["required"]
            .as_array()
            .unwrap()
            .iter()
            .any(|v| v == "prompt"));
    }

    #[test]
    fn test_find_pawan_binary_prefers_release_over_debug() {
        let tmp = TempDir::new().unwrap();
        // Create both release and debug pawan binaries
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        std::fs::create_dir_all(tmp.path().join("target/debug")).unwrap();
        let release = tmp.path().join("target/release/pawan");
        let debug = tmp.path().join("target/debug/pawan");
        std::fs::write(&release, "#!/bin/sh\necho release").unwrap();
        std::fs::write(&debug, "#!/bin/sh\necho debug").unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let binary = tool.find_pawan_binary();
        assert_eq!(
            binary,
            release.to_string_lossy().to_string(),
            "release binary must win over debug"
        );
    }

    #[test]
    fn test_find_pawan_binary_falls_back_to_debug_when_no_release() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/debug")).unwrap();
        let debug = tmp.path().join("target/debug/pawan");
        std::fs::write(&debug, "#!/bin/sh\necho debug").unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let binary = tool.find_pawan_binary();
        assert_eq!(binary, debug.to_string_lossy().to_string());
    }

    #[test]
    fn test_find_pawan_binary_falls_through_to_path_when_nothing_in_workspace() {
        let tmp = TempDir::new().unwrap();
        // No target/ at all
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let binary = tool.find_pawan_binary();
        // Falls back to bare "pawan" name (will be resolved via PATH at exec time)
        assert_eq!(binary, "pawan");
    }

    #[tokio::test]
    async fn test_spawn_agent_missing_prompt_errors() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        // No "prompt" field in args
        let result = tool.execute(json!({ "model": "test-model" })).await;
        assert!(result.is_err(), "missing prompt must error");
        let err = format!("{}", result.unwrap_err());
        assert!(
            err.contains("prompt"),
            "error message should mention prompt, got: {}",
            err
        );
    }

    #[test]
    fn test_spawn_agents_schema_requires_tasks_array() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let schema = tool.parameters_schema();
        let required = schema["required"].as_array().unwrap();
        assert!(
            required.iter().any(|v| v == "tasks"),
            "tasks must be required"
        );
        // tasks should be declared as an array type with an items.required = [prompt]
        let tasks_type = schema["properties"]["tasks"]["type"].as_str();
        assert_eq!(tasks_type, Some("array"));
    }

    #[tokio::test]
    async fn test_spawn_agents_empty_tasks_succeeds_with_zero_results() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let result = tool.execute(json!({ "tasks": [] })).await.unwrap();
        assert_eq!(result["success"], true);
        assert_eq!(result["total_tasks"], 0);
        assert_eq!(result["results"].as_array().unwrap().len(), 0);
    }

    #[tokio::test]
    async fn test_spawn_agents_missing_tasks_errors() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        // tasks field absent entirely
        let result = tool.execute(json!({})).await;
        assert!(result.is_err());
        let err = format!("{}", result.unwrap_err());
        assert!(err.contains("tasks"));
    }

    #[tokio::test]
    async fn test_spawn_agent_prompt_non_string_errors() {
        // Prompt present but as integer — `.as_str()` returns None so
        // ok_or_else must fire. Guards against a refactor that silently
        // coerces numeric prompts (which would then panic or send garbage
        // to the pawan run subcommand).
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let result = tool.execute(json!({ "prompt": 42 })).await;
        assert!(result.is_err(), "non-string prompt must error");
        let err = format!("{}", result.unwrap_err());
        assert!(
            err.contains("prompt"),
            "error should mention 'prompt', got: {}",
            err
        );
    }

    #[tokio::test]
    async fn test_spawn_agents_tasks_non_array_errors() {
        // tasks present but as string — `.as_array()` returns None so
        // ok_or_else must fire. Prevents silent coercion.
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let result = tool.execute(json!({ "tasks": "not an array" })).await;
        assert!(result.is_err(), "non-array tasks must error");
        let err = format!("{}", result.unwrap_err());
        assert!(
            err.contains("tasks"),
            "error should mention 'tasks', got: {}",
            err
        );
    }

    #[test]
    fn test_spawn_agent_schema_lists_all_optional_params() {
        // All 5 advertised parameters must be declared as schema properties.
        // If someone adds a new one without updating the schema, consumers
        // that introspect parameters_schema() will miss it.
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let schema = tool.parameters_schema();
        let props = schema["properties"].as_object().unwrap();
        for p in &["prompt", "model", "timeout", "workspace", "retries"] {
            assert!(props.contains_key(*p), "schema missing '{}'", p);
        }
        // Only prompt is required
        let required = schema["required"].as_array().unwrap();
        assert_eq!(required.len(), 1);
        assert_eq!(required[0], "prompt");
    }

    #[test]
    fn test_spawn_agents_schema_tasks_items_has_prompt_required() {
        // The nested items schema inside tasks must mark prompt required —
        // otherwise the array can hold task objects missing the prompt
        // field, and each sub-execute() would error one-by-one instead of
        // validating up front.
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let schema = tool.parameters_schema();
        let items_required = schema["properties"]["tasks"]["items"]["required"]
            .as_array()
            .expect("tasks.items.required should exist");
        assert!(items_required.iter().any(|v| v == "prompt"));
    }

    #[test]
    fn test_spawn_agent_thulp_definition_has_all_5_params() {
        // thulp_definition() must mirror parameters_schema() — if they drift,
        // thulp-registry callers will see a different API than MCP callers.
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let def = tool.thulp_definition();
        assert_eq!(def.name, "spawn_agent");
        let param_names: Vec<&str> = def.parameters.iter().map(|p| p.name.as_str()).collect();
        for p in &["prompt", "model", "timeout", "workspace", "retries"] {
            assert!(param_names.contains(p), "thulp definition missing '{}'", p);
        }
        // Only prompt is required
        let required_count = def.parameters.iter().filter(|p| p.required).count();
        assert_eq!(required_count, 1, "only prompt should be required");
    }

    #[test]
    fn test_spawn_agents_thulp_definition_has_tasks_param() {
        // spawn_agents' thulp definition should declare exactly one required
        // parameter: tasks (an array). If this drifts, parallel-agent callers
        // get confused schemas.
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let def = tool.thulp_definition();
        assert_eq!(def.name, "spawn_agents");
        assert_eq!(def.parameters.len(), 1);
        let tasks_param = &def.parameters[0];
        assert_eq!(tasks_param.name, "tasks");
        assert!(tasks_param.required);
    }

    #[test]
    fn test_spawn_agent_mutating_returns_true() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        assert!(tool.mutating(), "spawn_agent can mutate state");
    }

    #[test]
    fn test_spawn_agents_mutating_returns_true() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        assert!(tool.mutating(), "spawn_agents can mutate state");
    }

    #[test]
    fn test_spawn_agent_description_non_empty() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let desc = tool.description();
        assert!(!desc.is_empty());
        assert!(desc.contains("sub-agent"));
    }

    #[test]
    fn test_spawn_agents_description_non_empty() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let desc = tool.description();
        assert!(!desc.is_empty());
        assert!(desc.contains("parallel"));
    }

    #[tokio::test]
    async fn test_spawn_agent_timeout_defaults_to_120() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        std::fs::write(
            &binary,
            r"#!/bin/sh
exit 0",
        )
        .unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let result = tool.execute(json!({"prompt": "test"})).await.unwrap();
        assert_eq!(result["success"], true);
    }

    #[tokio::test]
    async fn test_spawn_agent_custom_timeout() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        std::fs::write(
            &binary,
            r"#!/bin/sh
exit 0",
        )
        .unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let result = tool
            .execute(json!({"prompt": "test", "timeout": 60}))
            .await
            .unwrap();
        assert_eq!(result["success"], true);
    }

    #[tokio::test]
    async fn test_spawn_agent_custom_model() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        std::fs::write(&binary, "#!/bin/sh\necho '{\"content\":\"test response\"}'").unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let result = tool
            .execute(json!({"prompt": "test", "model": "gpt-4"}))
            .await
            .unwrap();
        assert_eq!(result["success"], true);
        assert_eq!(result["result"]["content"], "test response");
    }

    #[tokio::test]
    async fn test_spawn_agent_retries_on_failure() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        let counter_file = tmp.path().join("counter");
        std::fs::write(&counter_file, "0").unwrap();
        let script = format!(
            "#!/bin/sh\ncount=$(cat {})\necho $((count + 1)) > {}\nif [ $count -eq 0 ]; then\n    exit 1\nelse\n    exit 0\nfi",
            counter_file.display(), counter_file.display()
        );
        std::fs::write(&binary, script).unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let result = tool
            .execute(json!({
                "prompt": "test",
                "retries": 1
            }))
            .await
            .unwrap();
        assert_eq!(result["success"], true);
        assert_eq!(result["attempt"], 2);
        assert_eq!(result["total_attempts"], 2);
    }

    #[tokio::test]
    async fn test_spawn_agent_stderr_captured() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        std::fs::write(
            &binary,
            r"#!/bin/sh
echo 'error message' >&2
exit 0",
        )
        .unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentTool::new(tmp.path().to_path_buf());
        let result = tool.execute(json!({"prompt": "test"})).await.unwrap();
        assert_eq!(result["success"], true);
        assert_eq!(result["stderr"], "error message");
    }

    #[serial_test::serial(pawan_session_tests)]
    #[tokio::test]
    async fn test_spawn_agents_single_task() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        std::fs::write(&binary, "#!/bin/sh\necho '{\"result\":\"done\"}'").unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let result = tool
            .execute(json!({
                "tasks": [
                    {"prompt": "task1"}
                ]
            }))
            .await
            .unwrap();
        assert_eq!(result["success"], true);
        assert_eq!(result["total_tasks"], 1);
        assert_eq!(result["results"].as_array().unwrap().len(), 1);
        assert_eq!(result["results"][0]["result"]["result"], "done");
    }

    #[serial_test::serial(pawan_session_tests)]
    #[tokio::test]
    async fn test_spawn_agents_multiple_tasks() {
        let tmp = TempDir::new().unwrap();
        std::fs::create_dir_all(tmp.path().join("target/release")).unwrap();
        let binary = tmp.path().join("target/release/pawan");
        std::fs::write(&binary, "#!/bin/sh\necho '{\"result\":\"done\"}'").unwrap();
        #[cfg(unix)]
        std::fs::set_permissions(&binary, std::os::unix::fs::PermissionsExt::from_mode(0o755))
            .unwrap();

        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let result = tool
            .execute(json!({
                "tasks": [
                    {"prompt": "task1"},
                    {"prompt": "task2"},
                    {"prompt": "task3"}
                ]
            }))
            .await
            .unwrap();
        assert_eq!(result["success"], true);
        assert_eq!(result["total_tasks"], 3);
        assert_eq!(result["results"].as_array().unwrap().len(), 3);
    }

    #[tokio::test]
    async fn test_spawn_agents_task_missing_prompt() {
        let tmp = TempDir::new().unwrap();
        let tool = SpawnAgentsTool::new(tmp.path().to_path_buf());
        let result = tool
            .execute(json!({
                "tasks": [{"model": "gpt-4"}]
            }))
            .await;
        // Upfront validation now rejects tasks missing a prompt
        assert!(result.is_err());
        let err = result.unwrap_err();
        assert!(err.to_string().contains("missing prompt"));
    }
}