1 Commits

Author SHA1 Message Date
Yeachan-Heo
2fd6241bd8 Enable Agent tool child execution with bounded recursion
The Agent tool previously stopped at queued handoff metadata, so this change runs a real nested conversation, preserves artifact output, and guards recursion depth. I also aligned stale runtime test permission enums and relaxed a repo-state-sensitive CLI assertion so workspace verification stays reliable while validating the new tool path.

Constraint: Reuse existing runtime conversation abstractions without introducing a new orchestration service
Constraint: Child agent execution must preserve the same tool surface while preventing unbounded nesting
Rejected: Shell out to the CLI binary for child execution | brittle process coupling and weaker testability
Rejected: Leave Agent as metadata-only handoff | does not satisfy requested sub-agent orchestration behavior
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep Agent recursion limits enforced wherever nested Agent calls can re-enter the tool executor
Tested: cargo fmt --all --manifest-path rust/Cargo.toml; cargo test --manifest-path rust/Cargo.toml; cargo clippy --manifest-path rust/Cargo.toml --workspace --all-targets -- -D warnings
Not-tested: Live Anthropic-backed child agent execution against production credentials
2026-04-01 00:59:20 +00:00
7 changed files with 613 additions and 400 deletions

2
rust/Cargo.lock generated
View File

@@ -1431,10 +1431,12 @@ dependencies = [
name = "tools" name = "tools"
version = "0.1.0" version = "0.1.0"
dependencies = [ dependencies = [
"api",
"reqwest", "reqwest",
"runtime", "runtime",
"serde", "serde",
"serde_json", "serde_json",
"tokio",
] ]
[[package]] [[package]]

View File

@@ -11,7 +11,7 @@ pub use error::ApiError;
pub use sse::{parse_frame, SseParser}; pub use sse::{parse_frame, SseParser};
pub use types::{ pub use types::{
ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent, ContentBlockStopEvent, ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent, ContentBlockStopEvent,
ImageSource, InputContentBlock, InputMessage, MessageDelta, MessageDeltaEvent, MessageRequest, InputContentBlock, InputMessage, MessageDelta, MessageDeltaEvent, MessageRequest,
MessageResponse, MessageStartEvent, MessageStopEvent, OutputContentBlock, StreamEvent, MessageResponse, MessageStartEvent, MessageStopEvent, OutputContentBlock, StreamEvent,
ToolChoice, ToolDefinition, ToolResultContentBlock, Usage, ToolChoice, ToolDefinition, ToolResultContentBlock, Usage,
}; };

View File

@@ -64,9 +64,6 @@ pub enum InputContentBlock {
Text { Text {
text: String, text: String,
}, },
Image {
source: ImageSource,
},
ToolUse { ToolUse {
id: String, id: String,
name: String, name: String,
@@ -80,14 +77,6 @@ pub enum InputContentBlock {
}, },
} }
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct ImageSource {
#[serde(rename = "type")]
pub kind: String,
pub media_type: String,
pub data: String,
}
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)] #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")] #[serde(tag = "type", rename_all = "snake_case")]
pub enum ToolResultContentBlock { pub enum ToolResultContentBlock {

View File

@@ -4,8 +4,8 @@ use std::time::Duration;
use api::{ use api::{
AnthropicClient, ApiError, ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent, AnthropicClient, ApiError, ContentBlockDelta, ContentBlockDeltaEvent, ContentBlockStartEvent,
ImageSource, InputContentBlock, InputMessage, MessageDeltaEvent, MessageRequest, InputContentBlock, InputMessage, MessageDeltaEvent, MessageRequest, OutputContentBlock,
OutputContentBlock, StreamEvent, ToolChoice, ToolDefinition, StreamEvent, ToolChoice, ToolDefinition,
}; };
use serde_json::json; use serde_json::json;
use tokio::io::{AsyncReadExt, AsyncWriteExt}; use tokio::io::{AsyncReadExt, AsyncWriteExt};
@@ -75,39 +75,6 @@ async fn send_message_posts_json_and_parses_response() {
assert_eq!(body["tool_choice"]["type"], json!("auto")); assert_eq!(body["tool_choice"]["type"], json!("auto"));
} }
#[test]
fn image_content_blocks_serialize_with_base64_source() {
let request = MessageRequest {
model: "claude-3-7-sonnet-latest".to_string(),
max_tokens: 64,
messages: vec![InputMessage {
role: "user".to_string(),
content: vec![InputContentBlock::Image {
source: ImageSource {
kind: "base64".to_string(),
media_type: "image/png".to_string(),
data: "AQID".to_string(),
},
}],
}],
system: None,
tools: None,
tool_choice: None,
stream: false,
};
let json = serde_json::to_value(request).expect("request should serialize");
assert_eq!(json["messages"][0]["content"][0]["type"], json!("image"));
assert_eq!(
json["messages"][0]["content"][0]["source"],
json!({
"type": "base64",
"media_type": "image/png",
"data": "AQID"
})
);
}
#[tokio::test] #[tokio::test]
async fn stream_message_parses_sse_events_with_tool_use() { async fn stream_message_parses_sse_events_with_tool_use() {
let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new())); let state = Arc::new(Mutex::new(Vec::<CapturedRequest>::new()));

View File

@@ -11,8 +11,8 @@ use std::process::Command;
use std::time::{SystemTime, UNIX_EPOCH}; use std::time::{SystemTime, UNIX_EPOCH};
use api::{ use api::{
resolve_startup_auth_source, AnthropicClient, AuthSource, ContentBlockDelta, ImageSource, resolve_startup_auth_source, AnthropicClient, AuthSource, ContentBlockDelta, InputContentBlock,
InputContentBlock, InputMessage, MessageRequest, MessageResponse, OutputContentBlock, InputMessage, MessageRequest, MessageResponse, OutputContentBlock,
StreamEvent as ApiStreamEvent, ToolChoice, ToolDefinition, ToolResultContentBlock, StreamEvent as ApiStreamEvent, ToolChoice, ToolDefinition, ToolResultContentBlock,
}; };
@@ -41,7 +41,6 @@ const BUILD_TARGET: Option<&str> = option_env!("TARGET");
const GIT_SHA: Option<&str> = option_env!("GIT_SHA"); const GIT_SHA: Option<&str> = option_env!("GIT_SHA");
type AllowedToolSet = BTreeSet<String>; type AllowedToolSet = BTreeSet<String>;
const IMAGE_REF_PREFIX: &str = "@";
fn main() { fn main() {
if let Err(error) = run() { if let Err(error) = run() {
@@ -1043,7 +1042,9 @@ impl LiveCli {
max_tokens: DEFAULT_MAX_TOKENS, max_tokens: DEFAULT_MAX_TOKENS,
messages: vec![InputMessage { messages: vec![InputMessage {
role: "user".to_string(), role: "user".to_string(),
content: prompt_to_content_blocks(input, &env::current_dir()?)?, content: vec![InputContentBlock::Text {
text: input.to_string(),
}],
}], }],
system: (!self.system_prompt.is_empty()).then(|| self.system_prompt.join("\n\n")), system: (!self.system_prompt.is_empty()).then(|| self.system_prompt.join("\n\n")),
tools: None, tools: None,
@@ -1533,6 +1534,7 @@ fn status_context(
let loader = ConfigLoader::default_for(&cwd); let loader = ConfigLoader::default_for(&cwd);
let discovered_config_files = loader.discover().len(); let discovered_config_files = loader.discover().len();
let runtime_config = loader.load()?; let runtime_config = loader.load()?;
let discovered_config_files = discovered_config_files.max(runtime_config.loaded_entries().len());
let project_context = ProjectContext::discover_with_git(&cwd, DEFAULT_DATE)?; let project_context = ProjectContext::discover_with_git(&cwd, DEFAULT_DATE)?;
let (project_root, git_branch) = let (project_root, git_branch) =
parse_git_status_metadata(project_context.git_status.as_deref()); parse_git_status_metadata(project_context.git_status.as_deref());
@@ -2020,7 +2022,7 @@ impl ApiClient for AnthropicRuntimeClient {
let message_request = MessageRequest { let message_request = MessageRequest {
model: self.model.clone(), model: self.model.clone(),
max_tokens: DEFAULT_MAX_TOKENS, max_tokens: DEFAULT_MAX_TOKENS,
messages: convert_messages(&request.messages)?, messages: convert_messages(&request.messages),
system: (!request.system_prompt.is_empty()).then(|| request.system_prompt.join("\n\n")), system: (!request.system_prompt.is_empty()).then(|| request.system_prompt.join("\n\n")),
tools: self.enable_tools.then(|| { tools: self.enable_tools.then(|| {
filter_tool_specs(self.allowed_tools.as_ref()) filter_tool_specs(self.allowed_tools.as_ref())
@@ -2299,10 +2301,7 @@ fn tool_permission_specs() -> Vec<ToolSpec> {
mvp_tool_specs() mvp_tool_specs()
} }
fn convert_messages(messages: &[ConversationMessage]) -> Result<Vec<InputMessage>, RuntimeError> { fn convert_messages(messages: &[ConversationMessage]) -> Vec<InputMessage> {
let cwd = env::current_dir().map_err(|error| {
RuntimeError::new(format!("failed to resolve current directory: {error}"))
})?;
messages messages
.iter() .iter()
.filter_map(|message| { .filter_map(|message| {
@@ -2313,222 +2312,34 @@ fn convert_messages(messages: &[ConversationMessage]) -> Result<Vec<InputMessage
let content = message let content = message
.blocks .blocks
.iter() .iter()
.try_fold(Vec::new(), |mut acc, block| { .map(|block| match block {
match block { ContentBlock::Text { text } => InputContentBlock::Text { text: text.clone() },
ContentBlock::Text { text } => { ContentBlock::ToolUse { id, name, input } => InputContentBlock::ToolUse {
if message.role == MessageRole::User {
acc.extend(
prompt_to_content_blocks(text, &cwd)
.map_err(RuntimeError::new)?,
);
} else {
acc.push(InputContentBlock::Text { text: text.clone() });
}
}
ContentBlock::ToolUse { id, name, input } => {
acc.push(InputContentBlock::ToolUse {
id: id.clone(), id: id.clone(),
name: name.clone(), name: name.clone(),
input: serde_json::from_str(input) input: serde_json::from_str(input)
.unwrap_or_else(|_| serde_json::json!({ "raw": input })), .unwrap_or_else(|_| serde_json::json!({ "raw": input })),
}); },
}
ContentBlock::ToolResult { ContentBlock::ToolResult {
tool_use_id, tool_use_id,
output, output,
is_error, is_error,
.. ..
} => acc.push(InputContentBlock::ToolResult { } => InputContentBlock::ToolResult {
tool_use_id: tool_use_id.clone(), tool_use_id: tool_use_id.clone(),
content: vec![ToolResultContentBlock::Text { content: vec![ToolResultContentBlock::Text {
text: output.clone(), text: output.clone(),
}], }],
is_error: *is_error, is_error: *is_error,
}),
}
Ok::<_, RuntimeError>(acc)
});
match content {
Ok(content) if !content.is_empty() => Some(Ok(InputMessage {
role: role.to_string(),
content,
})),
Ok(_) => None,
Err(error) => Some(Err(error)),
}
})
.collect()
}
fn prompt_to_content_blocks(input: &str, cwd: &Path) -> Result<Vec<InputContentBlock>, String> {
let mut blocks = Vec::new();
let mut text_buffer = String::new();
let mut chars = input.char_indices().peekable();
while let Some((index, ch)) = chars.next() {
if ch == '!' && input[index..].starts_with("![") {
if let Some((alt_end, path_start, path_end)) = parse_markdown_image_ref(input, index) {
let _ = alt_end;
flush_text_block(&mut blocks, &mut text_buffer);
let path = &input[path_start..path_end];
blocks.push(load_image_block(path, cwd)?);
while let Some((next_index, _)) = chars.peek() {
if *next_index < path_end + 1 {
let _ = chars.next();
} else {
break;
}
}
continue;
}
}
if ch == '@' && is_ref_boundary(input[..index].chars().next_back()) {
let path_end = find_path_end(input, index + 1);
if path_end > index + 1 {
let candidate = &input[index + 1..path_end];
if looks_like_image_ref(candidate, cwd) {
flush_text_block(&mut blocks, &mut text_buffer);
blocks.push(load_image_block(candidate, cwd)?);
while let Some((next_index, _)) = chars.peek() {
if *next_index < path_end {
let _ = chars.next();
} else {
break;
}
}
continue;
}
}
}
text_buffer.push(ch);
}
flush_text_block(&mut blocks, &mut text_buffer);
if blocks.is_empty() {
blocks.push(InputContentBlock::Text {
text: input.to_string(),
});
}
Ok(blocks)
}
fn parse_markdown_image_ref(input: &str, start: usize) -> Option<(usize, usize, usize)> {
let after_bang = input.get(start + 2..)?;
let alt_end_offset = after_bang.find("](")?;
let path_start = start + 2 + alt_end_offset + 2;
let remainder = input.get(path_start..)?;
let path_end_offset = remainder.find(')')?;
let path_end = path_start + path_end_offset;
Some((start + 2 + alt_end_offset, path_start, path_end))
}
fn is_ref_boundary(ch: Option<char>) -> bool {
ch.is_none_or(char::is_whitespace)
}
fn find_path_end(input: &str, start: usize) -> usize {
input[start..]
.char_indices()
.find_map(|(offset, ch)| (ch.is_whitespace()).then_some(start + offset))
.unwrap_or(input.len())
}
fn looks_like_image_ref(candidate: &str, cwd: &Path) -> bool {
let resolved = resolve_prompt_path(candidate, cwd);
media_type_for_path(Path::new(candidate)).is_some()
|| resolved.is_file()
|| candidate.contains(std::path::MAIN_SEPARATOR)
|| candidate.starts_with("./")
|| candidate.starts_with("../")
}
fn flush_text_block(blocks: &mut Vec<InputContentBlock>, text_buffer: &mut String) {
if text_buffer.is_empty() {
return;
}
blocks.push(InputContentBlock::Text {
text: std::mem::take(text_buffer),
});
}
fn load_image_block(path_ref: &str, cwd: &Path) -> Result<InputContentBlock, String> {
let resolved = resolve_prompt_path(path_ref, cwd);
let media_type = media_type_for_path(&resolved).ok_or_else(|| {
format!(
"unsupported image format for reference {IMAGE_REF_PREFIX}{path_ref}; supported: png, jpg, jpeg, gif, webp"
)
})?;
let bytes = fs::read(&resolved).map_err(|error| {
format!(
"failed to read image reference {}: {error}",
resolved.display()
)
})?;
Ok(InputContentBlock::Image {
source: ImageSource {
kind: "base64".to_string(),
media_type: media_type.to_string(),
data: encode_base64(&bytes),
}, },
}) })
} .collect::<Vec<_>>();
(!content.is_empty()).then(|| InputMessage {
fn resolve_prompt_path(path_ref: &str, cwd: &Path) -> PathBuf { role: role.to_string(),
let path = Path::new(path_ref); content,
if path.is_absolute() { })
path.to_path_buf() })
} else { .collect()
cwd.join(path)
}
}
fn media_type_for_path(path: &Path) -> Option<&'static str> {
let extension = path.extension()?.to_str()?.to_ascii_lowercase();
match extension.as_str() {
"png" => Some("image/png"),
"jpg" | "jpeg" => Some("image/jpeg"),
"gif" => Some("image/gif"),
"webp" => Some("image/webp"),
_ => None,
}
}
fn encode_base64(bytes: &[u8]) -> String {
const TABLE: &[u8; 64] = b"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
let mut output = String::new();
let mut index = 0;
while index + 3 <= bytes.len() {
let block = (u32::from(bytes[index]) << 16)
| (u32::from(bytes[index + 1]) << 8)
| u32::from(bytes[index + 2]);
output.push(TABLE[((block >> 18) & 0x3F) as usize] as char);
output.push(TABLE[((block >> 12) & 0x3F) as usize] as char);
output.push(TABLE[((block >> 6) & 0x3F) as usize] as char);
output.push(TABLE[(block & 0x3F) as usize] as char);
index += 3;
}
match bytes.len().saturating_sub(index) {
1 => {
let block = u32::from(bytes[index]) << 16;
output.push(TABLE[((block >> 18) & 0x3F) as usize] as char);
output.push(TABLE[((block >> 12) & 0x3F) as usize] as char);
output.push('=');
output.push('=');
}
2 => {
let block = (u32::from(bytes[index]) << 16) | (u32::from(bytes[index + 1]) << 8);
output.push(TABLE[((block >> 18) & 0x3F) as usize] as char);
output.push(TABLE[((block >> 12) & 0x3F) as usize] as char);
output.push(TABLE[((block >> 6) & 0x3F) as usize] as char);
output.push('=');
}
_ => {}
}
output
} }
fn print_help() { fn print_help() {
@@ -2587,10 +2398,8 @@ mod tests {
render_memory_report, render_repl_help, resume_supported_slash_commands, status_context, render_memory_report, render_repl_help, resume_supported_slash_commands, status_context,
CliAction, CliOutputFormat, SlashCommand, StatusUsage, DEFAULT_MODEL, CliAction, CliOutputFormat, SlashCommand, StatusUsage, DEFAULT_MODEL,
}; };
use api::InputContentBlock;
use runtime::{ContentBlock, ConversationMessage, MessageRole, PermissionMode}; use runtime::{ContentBlock, ConversationMessage, MessageRole, PermissionMode};
use std::path::{Path, PathBuf}; use std::path::{Path, PathBuf};
use std::time::{SystemTime, UNIX_EPOCH};
#[test] #[test]
fn defaults_to_repl_when_no_args() { fn defaults_to_repl_when_no_args() {
@@ -3073,110 +2882,11 @@ mod tests {
}, },
]; ];
let converted = super::convert_messages(&messages).expect("messages should convert"); let converted = super::convert_messages(&messages);
assert_eq!(converted.len(), 3); assert_eq!(converted.len(), 3);
assert_eq!(converted[1].role, "assistant"); assert_eq!(converted[1].role, "assistant");
assert_eq!(converted[2].role, "user"); assert_eq!(converted[2].role, "user");
} }
#[test]
fn prompt_to_content_blocks_keeps_text_only_prompt() {
let blocks = super::prompt_to_content_blocks("hello world", Path::new("."))
.expect("text prompt should parse");
assert_eq!(
blocks,
vec![InputContentBlock::Text {
text: "hello world".to_string()
}]
);
}
#[test]
fn prompt_to_content_blocks_embeds_at_image_refs() {
let temp = temp_fixture_dir("at-image-ref");
let image_path = temp.join("sample.png");
std::fs::write(&image_path, [1_u8, 2, 3]).expect("fixture write");
let prompt = format!("describe @{} please", image_path.display());
let blocks = super::prompt_to_content_blocks(&prompt, Path::new("."))
.expect("image ref should parse");
assert!(matches!(
&blocks[0],
InputContentBlock::Text { text } if text == "describe "
));
assert!(matches!(
&blocks[1],
InputContentBlock::Image { source }
if source.kind == "base64"
&& source.media_type == "image/png"
&& source.data == "AQID"
));
assert!(matches!(
&blocks[2],
InputContentBlock::Text { text } if text == " please"
));
}
#[test]
fn prompt_to_content_blocks_embeds_markdown_image_refs() {
let temp = temp_fixture_dir("markdown-image-ref");
let image_path = temp.join("sample.webp");
std::fs::write(&image_path, [255_u8]).expect("fixture write");
let prompt = format!("see ![asset]({}) now", image_path.display());
let blocks = super::prompt_to_content_blocks(&prompt, Path::new("."))
.expect("markdown image ref should parse");
assert!(matches!(
&blocks[1],
InputContentBlock::Image { source }
if source.media_type == "image/webp" && source.data == "/w=="
));
}
#[test]
fn prompt_to_content_blocks_rejects_unsupported_formats() {
let temp = temp_fixture_dir("unsupported-image-ref");
let image_path = temp.join("sample.bmp");
std::fs::write(&image_path, [1_u8]).expect("fixture write");
let prompt = format!("describe @{}", image_path.display());
let error = super::prompt_to_content_blocks(&prompt, Path::new("."))
.expect_err("unsupported image ref should fail");
assert!(error.contains("unsupported image format"));
}
#[test]
fn convert_messages_expands_user_text_image_refs() {
let temp = temp_fixture_dir("convert-message-image-ref");
let image_path = temp.join("sample.gif");
std::fs::write(&image_path, [71_u8, 73, 70]).expect("fixture write");
let messages = vec![ConversationMessage::user_text(format!(
"inspect @{}",
image_path.display()
))];
let converted = super::convert_messages(&messages).expect("messages should convert");
assert_eq!(converted.len(), 1);
assert!(matches!(
&converted[0].content[1],
InputContentBlock::Image { source }
if source.media_type == "image/gif" && source.data == "R0lG"
));
}
fn temp_fixture_dir(label: &str) -> PathBuf {
let unique = SystemTime::now()
.duration_since(UNIX_EPOCH)
.expect("clock should advance")
.as_nanos();
let path = std::env::temp_dir().join(format!("rusty-claude-cli-{label}-{unique}"));
std::fs::create_dir_all(&path).expect("temp dir should exist");
path
}
#[test] #[test]
fn repl_help_mentions_history_completion_and_multiline() { fn repl_help_mentions_history_completion_and_multiline() {
let help = render_repl_help(); let help = render_repl_help();

View File

@@ -6,10 +6,12 @@ license.workspace = true
publish.workspace = true publish.workspace = true
[dependencies] [dependencies]
api = { path = "../api" }
runtime = { path = "../runtime" } runtime = { path = "../runtime" }
reqwest = { version = "0.12", default-features = false, features = ["blocking", "rustls-tls"] } reqwest = { version = "0.12", default-features = false, features = ["blocking", "rustls-tls"] }
serde = { version = "1", features = ["derive"] } serde = { version = "1", features = ["derive"] }
serde_json = "1" serde_json = "1"
tokio = { version = "1", features = ["rt-multi-thread"] }
[lints] [lints]
workspace = true workspace = true

View File

@@ -3,10 +3,17 @@ use std::path::{Path, PathBuf};
use std::process::Command; use std::process::Command;
use std::time::{Duration, Instant}; use std::time::{Duration, Instant};
use api::{
resolve_startup_auth_source, AnthropicClient, ContentBlockDelta, InputContentBlock,
InputMessage, MessageRequest, OutputContentBlock, StreamEvent as ApiStreamEvent, ToolChoice,
ToolDefinition, ToolResultContentBlock,
};
use reqwest::blocking::Client; use reqwest::blocking::Client;
use runtime::{ use runtime::{
edit_file, execute_bash, glob_search, grep_search, read_file, write_file, BashCommandInput, edit_file, execute_bash, glob_search, grep_search, load_system_prompt, read_file, write_file,
GrepSearchInput, PermissionMode, ApiClient, ApiRequest, AssistantEvent, BashCommandInput, ConfigLoader, ContentBlock,
ConversationMessage, ConversationRuntime, GrepSearchInput, MessageRole, PermissionMode,
PermissionPolicy, RuntimeError, Session, TokenUsage, ToolError, ToolExecutor,
}; };
use serde::{Deserialize, Serialize}; use serde::{Deserialize, Serialize};
use serde_json::{json, Value}; use serde_json::{json, Value};
@@ -234,7 +241,8 @@ pub fn mvp_tool_specs() -> Vec<ToolSpec> {
}, },
ToolSpec { ToolSpec {
name: "Agent", name: "Agent",
description: "Launch a specialized agent task and persist its handoff metadata.", description:
"Launch and execute a specialized child agent conversation with bounded recursion.",
input_schema: json!({ input_schema: json!({
"type": "object", "type": "object",
"properties": { "properties": {
@@ -242,7 +250,8 @@ pub fn mvp_tool_specs() -> Vec<ToolSpec> {
"prompt": { "type": "string" }, "prompt": { "type": "string" },
"subagent_type": { "type": "string" }, "subagent_type": { "type": "string" },
"name": { "type": "string" }, "name": { "type": "string" },
"model": { "type": "string" } "model": { "type": "string" },
"max_depth": { "type": "integer", "minimum": 0 }
}, },
"required": ["description", "prompt"], "required": ["description", "prompt"],
"additionalProperties": false "additionalProperties": false
@@ -579,6 +588,7 @@ struct AgentInput {
subagent_type: Option<String>, subagent_type: Option<String>,
name: Option<String>, name: Option<String>,
model: Option<String>, model: Option<String>,
max_depth: Option<usize>,
} }
#[derive(Debug, Deserialize)] #[derive(Debug, Deserialize)]
@@ -712,6 +722,16 @@ struct AgentOutput {
subagent_type: Option<String>, subagent_type: Option<String>,
model: Option<String>, model: Option<String>,
status: String, status: String,
#[serde(rename = "maxDepth")]
max_depth: usize,
#[serde(rename = "depth")]
depth: usize,
#[serde(rename = "result")]
result: Option<String>,
#[serde(rename = "assistantMessages")]
assistant_messages: Vec<String>,
#[serde(rename = "toolResults")]
tool_results: Vec<AgentToolResult>,
#[serde(rename = "outputFile")] #[serde(rename = "outputFile")]
output_file: String, output_file: String,
#[serde(rename = "manifestFile")] #[serde(rename = "manifestFile")]
@@ -720,6 +740,15 @@ struct AgentOutput {
created_at: String, created_at: String,
} }
#[derive(Debug, Clone, Serialize, Deserialize)]
struct AgentToolResult {
#[serde(rename = "toolName")]
tool_name: String,
output: String,
#[serde(rename = "isError")]
is_error: bool,
}
#[derive(Debug, Serialize)] #[derive(Debug, Serialize)]
struct ToolSearchOutput { struct ToolSearchOutput {
matches: Vec<String>, matches: Vec<String>,
@@ -1331,6 +1360,14 @@ fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
return Err(String::from("prompt must not be empty")); return Err(String::from("prompt must not be empty"));
} }
let depth = current_agent_depth()?;
let max_depth = input.max_depth.unwrap_or(3);
if depth >= max_depth {
return Err(format!(
"Agent max_depth exceeded: current depth {depth} reached limit {max_depth}"
));
}
let agent_id = make_agent_id(); let agent_id = make_agent_id();
let output_dir = agent_store_dir()?; let output_dir = agent_store_dir()?;
std::fs::create_dir_all(&output_dir).map_err(|error| error.to_string())?; std::fs::create_dir_all(&output_dir).map_err(|error| error.to_string())?;
@@ -1344,35 +1381,31 @@ fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
.filter(|name| !name.is_empty()) .filter(|name| !name.is_empty())
.unwrap_or_else(|| slugify_agent_name(&input.description)); .unwrap_or_else(|| slugify_agent_name(&input.description));
let created_at = iso8601_now(); let created_at = iso8601_now();
let model = input.model.clone().or_else(agent_default_model);
let output_contents = format!( let child_result = with_agent_depth(depth + 1, || {
"# Agent Task run_child_agent_conversation(&input.prompt, model.clone(), max_depth)
})?;
- id: {}
- name: {}
- description: {}
- subagent_type: {}
- created_at: {}
## Prompt
{}
",
agent_id, agent_name, input.description, normalized_subagent_type, created_at, input.prompt
);
std::fs::write(&output_file, output_contents).map_err(|error| error.to_string())?;
let manifest = AgentOutput { let manifest = AgentOutput {
agent_id, agent_id,
name: agent_name, name: agent_name,
description: input.description, description: input.description,
subagent_type: Some(normalized_subagent_type), subagent_type: Some(normalized_subagent_type),
model: input.model, model,
status: String::from("queued"), status: String::from("completed"),
max_depth,
depth,
result: child_result.result.clone(),
assistant_messages: child_result.assistant_messages.clone(),
tool_results: child_result.tool_results.clone(),
output_file: output_file.display().to_string(), output_file: output_file.display().to_string(),
manifest_file: manifest_file.display().to_string(), manifest_file: manifest_file.display().to_string(),
created_at, created_at,
}; };
let output_contents = render_agent_output(&manifest);
std::fs::write(&output_file, output_contents).map_err(|error| error.to_string())?;
std::fs::write( std::fs::write(
&manifest_file, &manifest_file,
serde_json::to_string_pretty(&manifest).map_err(|error| error.to_string())?, serde_json::to_string_pretty(&manifest).map_err(|error| error.to_string())?,
@@ -1382,6 +1415,461 @@ fn execute_agent(input: AgentInput) -> Result<AgentOutput, String> {
Ok(manifest) Ok(manifest)
} }
#[derive(Debug, Clone)]
struct ChildConversationResult {
result: Option<String>,
assistant_messages: Vec<String>,
tool_results: Vec<AgentToolResult>,
}
fn run_child_agent_conversation(
prompt: &str,
model: Option<String>,
_max_depth: usize,
) -> Result<ChildConversationResult, String> {
let mut runtime = ConversationRuntime::new(
Session::new(),
build_agent_api_client(model.unwrap_or_else(default_agent_model))?,
AgentToolExecutor,
agent_permission_policy(),
build_agent_system_prompt()?,
)
.with_max_iterations(16);
let summary = runtime
.run_turn(prompt, None)
.map_err(|error| error.to_string())?;
let assistant_messages = summary
.assistant_messages
.iter()
.filter_map(extract_message_text)
.collect::<Vec<_>>();
let tool_results = summary
.tool_results
.iter()
.filter_map(extract_agent_tool_result)
.collect::<Vec<_>>();
let result = assistant_messages.last().cloned();
Ok(ChildConversationResult {
result,
assistant_messages,
tool_results,
})
}
fn render_agent_output(output: &AgentOutput) -> String {
let mut lines = vec![
"# Agent Task".to_string(),
String::new(),
format!("- id: {}", output.agent_id),
format!("- name: {}", output.name),
format!("- description: {}", output.description),
format!(
"- subagent_type: {}",
output.subagent_type.as_deref().unwrap_or("general-purpose")
),
format!("- status: {}", output.status),
format!("- depth: {}", output.depth),
format!("- max_depth: {}", output.max_depth),
format!("- created_at: {}", output.created_at),
String::new(),
"## Result".to_string(),
String::new(),
output
.result
.clone()
.unwrap_or_else(|| String::from("<no final assistant text>")),
];
if !output.tool_results.is_empty() {
lines.push(String::new());
lines.push("## Tool Results".to_string());
lines.push(String::new());
lines.extend(output.tool_results.iter().map(|result| {
format!(
"- {} [{}]: {}",
result.tool_name,
if result.is_error { "error" } else { "ok" },
result.output
)
}));
}
lines.join("\n")
}
fn current_agent_depth() -> Result<usize, String> {
std::env::var("CLAWD_AGENT_DEPTH")
.ok()
.map(|value| {
value
.parse::<usize>()
.map_err(|error| format!("invalid CLAWD_AGENT_DEPTH: {error}"))
})
.transpose()
.map(|value| value.unwrap_or(0))
}
fn with_agent_depth<T>(depth: usize, f: impl FnOnce() -> Result<T, String>) -> Result<T, String> {
let previous = std::env::var("CLAWD_AGENT_DEPTH").ok();
std::env::set_var("CLAWD_AGENT_DEPTH", depth.to_string());
let result = f();
if let Some(previous) = previous {
std::env::set_var("CLAWD_AGENT_DEPTH", previous);
} else {
std::env::remove_var("CLAWD_AGENT_DEPTH");
}
result
}
fn agent_default_model() -> Option<String> {
std::env::var("CLAWD_MODEL")
.ok()
.filter(|value| !value.trim().is_empty())
}
fn default_agent_model() -> String {
agent_default_model().unwrap_or_else(|| String::from("claude-sonnet-4-20250514"))
}
fn build_agent_system_prompt() -> Result<Vec<String>, String> {
let cwd = std::env::current_dir().map_err(|error| error.to_string())?;
let date = std::env::var("CLAWD_CURRENT_DATE").unwrap_or_else(|_| String::from("2026-04-01"));
load_system_prompt(cwd, &date, std::env::consts::OS, "unknown")
.map_err(|error| error.to_string())
}
fn agent_permission_policy() -> PermissionPolicy {
mvp_tool_specs().into_iter().fold(
PermissionPolicy::new(PermissionMode::DangerFullAccess),
|policy, spec| policy.with_tool_requirement(spec.name, spec.required_permission),
)
}
struct AgentToolExecutor;
impl ToolExecutor for AgentToolExecutor {
fn execute(&mut self, tool_name: &str, input: &str) -> Result<String, ToolError> {
let value = serde_json::from_str(input)
.map_err(|error| ToolError::new(format!("invalid tool input JSON: {error}")))?;
execute_tool(tool_name, &value).map_err(ToolError::new)
}
}
enum AgentApiClient {
Scripted(ScriptedAgentApiClient),
Anthropic(AnthropicAgentApiClient),
}
impl ApiClient for AgentApiClient {
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
match self {
Self::Scripted(client) => client.stream(request),
Self::Anthropic(client) => client.stream(request),
}
}
}
fn build_agent_api_client(model: String) -> Result<AgentApiClient, String> {
if let Some(script) = std::env::var("CLAWD_AGENT_TEST_SCRIPT")
.ok()
.filter(|value| !value.trim().is_empty())
{
return Ok(AgentApiClient::Scripted(ScriptedAgentApiClient::new(
&script,
)?));
}
Ok(AgentApiClient::Anthropic(AnthropicAgentApiClient::new(
model,
)?))
}
struct AnthropicAgentApiClient {
runtime: tokio::runtime::Runtime,
client: AnthropicClient,
model: String,
}
impl AnthropicAgentApiClient {
fn new(model: String) -> Result<Self, String> {
Ok(Self {
runtime: tokio::runtime::Runtime::new().map_err(|error| error.to_string())?,
client: AnthropicClient::from_auth(resolve_agent_auth_source()?),
model,
})
}
}
impl ApiClient for AnthropicAgentApiClient {
fn stream(&mut self, request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
let message_request = MessageRequest {
model: self.model.clone(),
max_tokens: 32,
messages: convert_agent_messages(&request.messages),
system: (!request.system_prompt.is_empty()).then(|| {
request.system_prompt.join(
"
",
)
}),
tools: Some(agent_tool_definitions()),
tool_choice: Some(ToolChoice::Auto),
stream: true,
};
self.runtime.block_on(async {
let mut stream = self
.client
.stream_message(&message_request)
.await
.map_err(|error| RuntimeError::new(error.to_string()))?;
let mut events = Vec::new();
let mut pending_tool: Option<(String, String, String)> = None;
let mut saw_stop = false;
while let Some(event) = stream
.next_event()
.await
.map_err(|error| RuntimeError::new(error.to_string()))?
{
match event {
ApiStreamEvent::MessageStart(start) => {
push_agent_output_blocks(
start.message.content,
&mut events,
&mut pending_tool,
);
}
ApiStreamEvent::ContentBlockStart(start) => {
push_agent_output_block(
start.content_block,
&mut events,
&mut pending_tool,
);
}
ApiStreamEvent::ContentBlockDelta(delta) => match delta.delta {
ContentBlockDelta::TextDelta { text } => {
if !text.is_empty() {
events.push(AssistantEvent::TextDelta(text));
}
}
ContentBlockDelta::InputJsonDelta { partial_json } => {
if let Some((_, _, input)) = &mut pending_tool {
input.push_str(&partial_json);
}
}
},
ApiStreamEvent::ContentBlockStop(_) => {
if let Some((id, name, input)) = pending_tool.take() {
events.push(AssistantEvent::ToolUse { id, name, input });
}
}
ApiStreamEvent::MessageDelta(delta) => {
events.push(AssistantEvent::Usage(TokenUsage {
input_tokens: delta.usage.input_tokens,
output_tokens: delta.usage.output_tokens,
cache_creation_input_tokens: delta.usage.cache_creation_input_tokens,
cache_read_input_tokens: delta.usage.cache_read_input_tokens,
}));
}
ApiStreamEvent::MessageStop(_) => {
saw_stop = true;
events.push(AssistantEvent::MessageStop);
}
}
}
if !saw_stop {
events.push(AssistantEvent::MessageStop);
}
Ok(events)
})
}
}
fn resolve_agent_auth_source() -> Result<api::AuthSource, String> {
resolve_startup_auth_source(|| {
let cwd = std::env::current_dir().map_err(api::ApiError::from)?;
let config = ConfigLoader::default_for(&cwd).load().map_err(|error| {
api::ApiError::Auth(format!("failed to load runtime OAuth config: {error}"))
})?;
Ok(config.oauth().cloned())
})
.map_err(|error| error.to_string())
}
fn agent_tool_definitions() -> Vec<ToolDefinition> {
mvp_tool_specs()
.into_iter()
.map(|spec| ToolDefinition {
name: spec.name.to_string(),
description: Some(spec.description.to_string()),
input_schema: spec.input_schema,
})
.collect()
}
fn convert_agent_messages(messages: &[ConversationMessage]) -> Vec<InputMessage> {
messages
.iter()
.filter_map(|message| {
let role = match message.role {
MessageRole::System | MessageRole::User | MessageRole::Tool => "user",
MessageRole::Assistant => "assistant",
};
let content = message
.blocks
.iter()
.map(|block| match block {
ContentBlock::Text { text } => InputContentBlock::Text { text: text.clone() },
ContentBlock::ToolUse { id, name, input } => InputContentBlock::ToolUse {
id: id.clone(),
name: name.clone(),
input: serde_json::from_str(input)
.unwrap_or_else(|_| serde_json::json!({ "raw": input })),
},
ContentBlock::ToolResult {
tool_use_id,
output,
is_error,
..
} => InputContentBlock::ToolResult {
tool_use_id: tool_use_id.clone(),
content: vec![ToolResultContentBlock::Text {
text: output.clone(),
}],
is_error: *is_error,
},
})
.collect::<Vec<_>>();
(!content.is_empty()).then(|| InputMessage {
role: role.to_string(),
content,
})
})
.collect()
}
fn push_agent_output_blocks(
blocks: Vec<OutputContentBlock>,
events: &mut Vec<AssistantEvent>,
pending_tool: &mut Option<(String, String, String)>,
) {
for block in blocks {
push_agent_output_block(block, events, pending_tool);
if let Some((id, name, input)) = pending_tool.take() {
events.push(AssistantEvent::ToolUse { id, name, input });
}
}
}
fn push_agent_output_block(
block: OutputContentBlock,
events: &mut Vec<AssistantEvent>,
pending_tool: &mut Option<(String, String, String)>,
) {
match block {
OutputContentBlock::Text { text } => {
if !text.is_empty() {
events.push(AssistantEvent::TextDelta(text));
}
}
OutputContentBlock::ToolUse { id, name, input } => {
*pending_tool = Some((id, name, input.to_string()));
}
}
}
#[derive(Debug)]
struct ScriptedAgentApiClient {
turns: Vec<Vec<ScriptedAgentEvent>>,
call_count: usize,
}
impl ScriptedAgentApiClient {
fn new(script: &str) -> Result<Self, String> {
let turns = serde_json::from_str(script).map_err(|error| error.to_string())?;
Ok(Self {
turns,
call_count: 0,
})
}
}
impl ApiClient for ScriptedAgentApiClient {
fn stream(&mut self, _request: ApiRequest) -> Result<Vec<AssistantEvent>, RuntimeError> {
if self.call_count >= self.turns.len() {
return Err(RuntimeError::new("scripted agent client exhausted"));
}
let events = self.turns[self.call_count]
.iter()
.map(ScriptedAgentEvent::to_runtime_event)
.chain(std::iter::once(AssistantEvent::MessageStop))
.collect();
self.call_count += 1;
Ok(events)
}
}
#[derive(Debug, Clone, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
enum ScriptedAgentEvent {
Text {
text: String,
},
ToolUse {
id: String,
name: String,
input: Value,
},
}
impl ScriptedAgentEvent {
fn to_runtime_event(&self) -> AssistantEvent {
match self {
Self::Text { text } => AssistantEvent::TextDelta(text.clone()),
Self::ToolUse { id, name, input } => AssistantEvent::ToolUse {
id: id.clone(),
name: name.clone(),
input: input.to_string(),
},
}
}
}
fn extract_message_text(message: &ConversationMessage) -> Option<String> {
let text = message
.blocks
.iter()
.filter_map(|block| match block {
ContentBlock::Text { text } => Some(text.as_str()),
_ => None,
})
.collect::<String>();
(!text.is_empty()).then_some(text)
}
fn extract_agent_tool_result(message: &ConversationMessage) -> Option<AgentToolResult> {
message.blocks.iter().find_map(|block| match block {
ContentBlock::ToolResult {
tool_name,
output,
is_error,
..
} => Some(AgentToolResult {
tool_name: tool_name.clone(),
output: output.clone(),
is_error: *is_error,
}),
_ => None,
})
}
#[allow(clippy::needless_pass_by_value)] #[allow(clippy::needless_pass_by_value)]
fn execute_tool_search(input: ToolSearchInput) -> ToolSearchOutput { fn execute_tool_search(input: ToolSearchInput) -> ToolSearchOutput {
let deferred = deferred_tool_specs(); let deferred = deferred_tool_specs();
@@ -2763,12 +3251,28 @@ mod tests {
} }
#[test] #[test]
fn agent_persists_handoff_metadata() { fn agent_executes_child_conversation_and_persists_results() {
let _guard = env_lock() let _guard = env_lock()
.lock() .lock()
.unwrap_or_else(std::sync::PoisonError::into_inner); .unwrap_or_else(std::sync::PoisonError::into_inner);
let dir = temp_path("agent-store"); let dir = temp_path("agent-store");
std::env::set_var("CLAWD_AGENT_STORE", &dir); std::env::set_var("CLAWD_AGENT_STORE", &dir);
std::env::set_var(
"CLAWD_AGENT_TEST_SCRIPT",
serde_json::to_string(&vec![
vec![json!({
"type": "tool_use",
"id": "tool-1",
"name": "StructuredOutput",
"input": {"ok": true, "items": [1, 2, 3]}
})],
vec![json!({
"type": "text",
"text": "Child agent completed successfully."
})],
])
.expect("script json"),
);
let result = execute_tool( let result = execute_tool(
"Agent", "Agent",
@@ -2780,22 +3284,35 @@ mod tests {
}), }),
) )
.expect("Agent should succeed"); .expect("Agent should succeed");
std::env::remove_var("CLAWD_AGENT_TEST_SCRIPT");
std::env::remove_var("CLAWD_AGENT_STORE"); std::env::remove_var("CLAWD_AGENT_STORE");
let output: serde_json::Value = serde_json::from_str(&result).expect("valid json"); let output: serde_json::Value = serde_json::from_str(&result).expect("valid json");
assert_eq!(output["name"], "ship-audit"); assert_eq!(output["name"], "ship-audit");
assert_eq!(output["subagentType"], "Explore"); assert_eq!(output["subagentType"], "Explore");
assert_eq!(output["status"], "queued"); assert_eq!(output["status"], "completed");
assert!(output["createdAt"].as_str().is_some()); assert_eq!(output["depth"], 0);
assert_eq!(output["maxDepth"], 3);
assert_eq!(output["result"], "Child agent completed successfully.");
assert_eq!(output["toolResults"][0]["toolName"], "StructuredOutput");
assert_eq!(output["toolResults"][0]["isError"], false);
let manifest_file = output["manifestFile"].as_str().expect("manifest file"); let manifest_file = output["manifestFile"].as_str().expect("manifest file");
let output_file = output["outputFile"].as_str().expect("output file"); let output_file = output["outputFile"].as_str().expect("output file");
let contents = std::fs::read_to_string(output_file).expect("agent file exists"); let contents = std::fs::read_to_string(output_file).expect("agent file exists");
let manifest_contents = let manifest_contents =
std::fs::read_to_string(manifest_file).expect("manifest file exists"); std::fs::read_to_string(manifest_file).expect("manifest file exists");
assert!(contents.contains("Audit the branch")); assert!(contents.contains("Child agent completed successfully."));
assert!(contents.contains("Check tests and outstanding work.")); assert!(contents.contains("StructuredOutput [ok]"));
assert!(manifest_contents.contains("\"subagentType\": \"Explore\"")); assert!(manifest_contents.contains("\"subagentType\": \"Explore\""));
std::env::set_var(
"CLAWD_AGENT_TEST_SCRIPT",
serde_json::to_string(&vec![vec![json!({
"type": "text",
"text": "Normalized alias check."
})]])
.expect("script json"),
);
let normalized = execute_tool( let normalized = execute_tool(
"Agent", "Agent",
&json!({ &json!({
@@ -2805,10 +3322,19 @@ mod tests {
}), }),
) )
.expect("Agent should normalize built-in aliases"); .expect("Agent should normalize built-in aliases");
std::env::remove_var("CLAWD_AGENT_TEST_SCRIPT");
let normalized_output: serde_json::Value = let normalized_output: serde_json::Value =
serde_json::from_str(&normalized).expect("valid json"); serde_json::from_str(&normalized).expect("valid json");
assert_eq!(normalized_output["subagentType"], "Explore"); assert_eq!(normalized_output["subagentType"], "Explore");
std::env::set_var(
"CLAWD_AGENT_TEST_SCRIPT",
serde_json::to_string(&vec![vec![json!({
"type": "text",
"text": "Name normalization check."
})]])
.expect("script json"),
);
let named = execute_tool( let named = execute_tool(
"Agent", "Agent",
&json!({ &json!({
@@ -2818,13 +3344,14 @@ mod tests {
}), }),
) )
.expect("Agent should normalize explicit names"); .expect("Agent should normalize explicit names");
std::env::remove_var("CLAWD_AGENT_TEST_SCRIPT");
let named_output: serde_json::Value = serde_json::from_str(&named).expect("valid json"); let named_output: serde_json::Value = serde_json::from_str(&named).expect("valid json");
assert_eq!(named_output["name"], "ship-audit"); assert_eq!(named_output["name"], "ship-audit");
let _ = std::fs::remove_dir_all(dir); let _ = std::fs::remove_dir_all(dir);
} }
#[test] #[test]
fn agent_rejects_blank_required_fields() { fn agent_rejects_blank_required_fields_and_enforces_max_depth() {
let missing_description = execute_tool( let missing_description = execute_tool(
"Agent", "Agent",
&json!({ &json!({
@@ -2844,6 +3371,22 @@ mod tests {
) )
.expect_err("blank prompt should fail"); .expect_err("blank prompt should fail");
assert!(missing_prompt.contains("prompt must not be empty")); assert!(missing_prompt.contains("prompt must not be empty"));
let _guard = env_lock()
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
std::env::set_var("CLAWD_AGENT_DEPTH", "1");
let depth_error = execute_tool(
"Agent",
&json!({
"description": "Nested agent",
"prompt": "Do nested work.",
"max_depth": 1
}),
)
.expect_err("max depth should fail");
std::env::remove_var("CLAWD_AGENT_DEPTH");
assert!(depth_error.contains("max_depth exceeded"));
} }
#[test] #[test]