Compare commits

..
Author SHA1 Message Date
Claude 1d9821d37a style(agent): fix formatting in approval TOCTOU fix
Apply cargo fmt to resolve CI formatting check failure.

https://claude.ai/code/session_013ZCQWoFHv2hASgHEGHzptg
2026-03-21 18:12:55 +00:00
ZakiandClaude Opus 4.6 52d935d744 fix(agent): eliminate TOCTOU race in approval thread resolution (#1486)
Restructure process_approval() to hold the session lock for the entire
check-take-verify-mutate sequence. Previously, the lock was dropped after
taking the pending approval and re-acquired to verify request_id and set
state, creating a window where concurrent operations could modify thread
state or lose the pending approval.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
2026-03-21 10:07:58 -07:00
21 changed files with 205 additions and 2375 deletions
-3
View File
@@ -55,9 +55,6 @@ RUN npm install -g @anthropic-ai/claude-code@latest
# Copy the binary
COPY --from=builder /build/target/release/ironclaw /usr/local/bin/ironclaw
# Install IronClaw Python SDK for programmatic tool calling (PTC)
COPY sdk/python/ironclaw_tools.py /usr/lib/python3/dist-packages/ironclaw_tools.py
# Create non-root user (UID 1000 matches the orchestrator's container config)
RUN useradd -m -u 1000 -s /bin/bash sandbox \
&& mkdir -p /workspace \
-158
View File
@@ -1,158 +0,0 @@
"""IronClaw Programmatic Tool Calling SDK for container scripts.
Thin wrapper using only Python stdlib. Reads connection details from
environment variables injected by the orchestrator:
IRONCLAW_ORCHESTRATOR_URL - Base URL of the orchestrator API
IRONCLAW_JOB_ID - UUID of the current job
IRONCLAW_WORKER_TOKEN - Bearer token scoped to this job
Usage:
from ironclaw_tools import call_tool, shell, read_file, write_file, http_get
# Call any registered tool by name
result = call_tool("echo", {"message": "hello"})
print(result) # "hello"
# Convenience wrappers
output = shell("ls -la")
content = read_file("/workspace/README.md")
write_file("/workspace/output.txt", "results here")
body = http_get("https://api.example.com/data")
"""
import json
import os
import urllib.request
import urllib.error
def _env(name):
"""Get a required environment variable."""
value = os.environ.get(name)
if not value:
raise RuntimeError(
f"Missing required environment variable: {name}. "
"This SDK must be run inside an IronClaw container."
)
return value
def _base_url():
"""Build the base URL for tool call requests."""
orchestrator = _env("IRONCLAW_ORCHESTRATOR_URL").rstrip("/")
job_id = _env("IRONCLAW_JOB_ID")
return f"{orchestrator}/worker/{job_id}"
def _token():
"""Get the bearer token."""
return _env("IRONCLAW_WORKER_TOKEN")
def call_tool(name, params=None, timeout_secs=60):
"""Call a tool on the orchestrator by name.
Args:
name: Tool name (e.g., "echo", "shell", "read_file").
params: Dictionary of parameters to pass to the tool.
timeout_secs: Timeout in seconds (default 60, max 300).
Returns:
Tool output as a string.
Raises:
RuntimeError: If the tool call fails.
"""
url = f"{_base_url()}/tools/call"
server_timeout = min(int(timeout_secs), 300)
body = {
"tool_name": name,
"parameters": params or {},
"timeout_secs": server_timeout,
}
data = json.dumps(body).encode("utf-8")
req = urllib.request.Request(
url,
data=data,
headers={
"Content-Type": "application/json",
"Authorization": f"Bearer {_token()}",
},
method="POST",
)
try:
# Client-side timeout slightly longer than server-side to account
# for network latency, preventing premature client timeouts.
client_timeout = server_timeout + 5
with urllib.request.urlopen(req, timeout=client_timeout) as resp:
result = json.loads(resp.read().decode("utf-8"))
except urllib.error.HTTPError as e:
body_text = e.read().decode("utf-8", errors="replace") if e.fp else ""
raise RuntimeError(
f"Tool call failed: HTTP {e.code}: {body_text}"
) from None
except urllib.error.URLError as e:
raise RuntimeError(f"Connection to orchestrator failed: {e.reason}") from None
if not result.get("success"):
raise RuntimeError(f"Tool '{name}' failed: {result.get('error', 'unknown error')}")
return result.get("output", "")
def shell(command, timeout_secs=60):
"""Execute a shell command via the orchestrator.
Args:
command: Shell command string to execute.
timeout_secs: Timeout in seconds (default 60).
Returns:
Command output as a string.
"""
return call_tool("shell", {"command": command}, timeout_secs=timeout_secs)
def read_file(path):
"""Read a file via the orchestrator.
Args:
path: Absolute path to the file.
Returns:
File contents as a string.
"""
return call_tool("read_file", {"path": path})
def write_file(path, content):
"""Write a file via the orchestrator.
Args:
path: Absolute path to write to.
content: String content to write.
Returns:
Write confirmation message.
"""
return call_tool("write_file", {"path": path, "content": content})
def http_get(url, headers=None, timeout_secs=30):
"""Make an HTTP GET request via the orchestrator's HTTP tool.
Args:
url: URL to fetch.
headers: Optional dictionary of headers.
timeout_secs: Timeout in seconds (default 30).
Returns:
Response body as a string.
"""
params = {"url": url, "method": "GET"}
if headers:
params["headers"] = headers
return call_tool("http", params, timeout_secs=timeout_secs)
-148
View File
@@ -1,148 +0,0 @@
"""Tests for the IronClaw Programmatic Tool Calling Python SDK."""
import json
import os
import sys
import unittest
from unittest.mock import patch, MagicMock
import urllib.error
# Ensure ironclaw_tools is importable regardless of working directory.
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
class TestEnvMissing(unittest.TestCase):
"""Test that missing env vars produce clear errors."""
def setUp(self):
# Clear all relevant env vars
for var in ["IRONCLAW_ORCHESTRATOR_URL", "IRONCLAW_JOB_ID", "IRONCLAW_WORKER_TOKEN"]:
os.environ.pop(var, None)
def test_env_missing(self):
from ironclaw_tools import call_tool
with self.assertRaises(RuntimeError) as ctx:
call_tool("echo", {"message": "hello"})
# Should mention the missing variable
self.assertIn("IRONCLAW_ORCHESTRATOR_URL", str(ctx.exception))
class TestCallToolRequestFormat(unittest.TestCase):
"""Test that call_tool sends correctly formatted requests."""
def setUp(self):
os.environ["IRONCLAW_ORCHESTRATOR_URL"] = "http://localhost:50051"
os.environ["IRONCLAW_JOB_ID"] = "550e8400-e29b-41d4-a716-446655440000"
os.environ["IRONCLAW_WORKER_TOKEN"] = "test-token-123"
def tearDown(self):
for var in ["IRONCLAW_ORCHESTRATOR_URL", "IRONCLAW_JOB_ID", "IRONCLAW_WORKER_TOKEN"]:
os.environ.pop(var, None)
@patch("ironclaw_tools.urllib.request.urlopen")
def test_call_tool_request_format(self, mock_urlopen):
from ironclaw_tools import call_tool
# Mock successful response
mock_response = MagicMock()
mock_response.read.return_value = json.dumps({
"success": True,
"output": "hello",
"duration_ms": 5,
"was_sanitized": False,
}).encode("utf-8")
mock_response.__enter__ = lambda s: s
mock_response.__exit__ = MagicMock(return_value=False)
mock_urlopen.return_value = mock_response
result = call_tool("echo", {"message": "hello"}, timeout_secs=30)
# Verify the request was made
mock_urlopen.assert_called_once()
call_args = mock_urlopen.call_args
req = call_args[0][0] # First positional arg is the Request object
# Check URL
self.assertIn("/worker/550e8400-e29b-41d4-a716-446655440000/tools/call", req.full_url)
# Check headers
self.assertEqual(req.get_header("Content-type"), "application/json")
self.assertEqual(req.get_header("Authorization"), "Bearer test-token-123")
# Check body
body = json.loads(req.data.decode("utf-8"))
self.assertEqual(body["tool_name"], "echo")
self.assertEqual(body["parameters"], {"message": "hello"})
self.assertEqual(body["timeout_secs"], 30)
# Check return value
self.assertEqual(result, "hello")
class TestCallToolHttpError(unittest.TestCase):
"""Test HTTP error handling."""
def setUp(self):
os.environ["IRONCLAW_ORCHESTRATOR_URL"] = "http://localhost:50051"
os.environ["IRONCLAW_JOB_ID"] = "550e8400-e29b-41d4-a716-446655440000"
os.environ["IRONCLAW_WORKER_TOKEN"] = "test-token-123"
def tearDown(self):
for var in ["IRONCLAW_ORCHESTRATOR_URL", "IRONCLAW_JOB_ID", "IRONCLAW_WORKER_TOKEN"]:
os.environ.pop(var, None)
@patch("ironclaw_tools.urllib.request.urlopen")
def test_call_tool_http_error(self, mock_urlopen):
from ironclaw_tools import call_tool
mock_urlopen.side_effect = urllib.error.HTTPError(
url="http://localhost:50051/worker/test/tools/call",
code=500,
msg="Internal Server Error",
hdrs=None,
fp=None,
)
with self.assertRaises(RuntimeError) as ctx:
call_tool("echo", {"message": "hello"})
self.assertIn("500", str(ctx.exception))
class TestConvenienceWrappers(unittest.TestCase):
"""Test that convenience wrappers call call_tool correctly."""
def setUp(self):
os.environ["IRONCLAW_ORCHESTRATOR_URL"] = "http://localhost:50051"
os.environ["IRONCLAW_JOB_ID"] = "550e8400-e29b-41d4-a716-446655440000"
os.environ["IRONCLAW_WORKER_TOKEN"] = "test-token-123"
def tearDown(self):
for var in ["IRONCLAW_ORCHESTRATOR_URL", "IRONCLAW_JOB_ID", "IRONCLAW_WORKER_TOKEN"]:
os.environ.pop(var, None)
@patch("ironclaw_tools.call_tool")
def test_convenience_wrappers(self, mock_call_tool):
from ironclaw_tools import shell, read_file, write_file, http_get
mock_call_tool.return_value = "output"
# Test shell
shell("ls -la")
mock_call_tool.assert_called_with("shell", {"command": "ls -la"}, timeout_secs=60)
# Test read_file
read_file("/workspace/README.md")
mock_call_tool.assert_called_with("read_file", {"path": "/workspace/README.md"})
# Test write_file
write_file("/workspace/out.txt", "content")
mock_call_tool.assert_called_with("write_file", {"path": "/workspace/out.txt", "content": "content"})
# Test http_get
http_get("https://api.example.com/data")
mock_call_tool.assert_called_with("http", {"url": "https://api.example.com/data", "method": "GET"}, timeout_secs=30)
if __name__ == "__main__":
unittest.main()
+175 -55
View File
@@ -868,7 +868,9 @@ impl Agent {
approved: bool,
always: bool,
) -> Result<SubmissionResult, Error> {
// Get pending approval for this thread
// Get pending approval for this thread.
// Hold the session lock for the entire check-take-verify-mutate sequence
// to eliminate the TOCTOU race window (fixes #1486).
let pending = {
let mut sess = session.lock().await;
let thread = sess
@@ -886,54 +888,46 @@ impl Agent {
return Ok(SubmissionResult::ok_with_message(""));
}
thread.take_pending_approval()
};
let pending = match thread.take_pending_approval() {
Some(p) => p,
None => {
tracing::debug!(
%thread_id,
"Ignoring stale approval: no pending approval found"
);
return Ok(SubmissionResult::ok_with_message(""));
}
};
let pending = match pending {
Some(p) => p,
None => {
tracing::debug!(
%thread_id,
"Ignoring stale approval: no pending approval found"
);
return Ok(SubmissionResult::ok_with_message(""));
}
};
// Verify request ID if provided
if let Some(req_id) = request_id
&& req_id != pending.request_id
{
// Put it back and return error
let mut sess = session.lock().await;
if let Some(thread) = sess.threads.get_mut(&thread_id) {
// Verify request ID while still holding the lock
if let Some(req_id) = request_id
&& req_id != pending.request_id
{
// Restore the pending approval atomically without releasing the lock
thread.await_approval(pending);
return Ok(SubmissionResult::error(
"Request ID mismatch. Use the correct request ID.",
));
}
return Ok(SubmissionResult::error(
"Request ID mismatch. Use the correct request ID.",
));
}
// If approved with "always", set auto-approve while we still have the lock.
// Drop the thread borrow before calling sess methods to satisfy the borrow checker.
if approved && always {
let tool_name = pending.tool_name.clone();
sess.auto_approve_tool(&tool_name);
tracing::info!("Auto-approved tool '{}' for session {}", tool_name, sess.id);
}
// Set thread state to Processing for approved requests.
// Re-borrow the thread since the previous borrow was dropped.
if approved && let Some(thread) = sess.threads.get_mut(&thread_id) {
thread.state = ThreadState::Processing;
}
pending
};
if approved {
// If always, add to auto-approved set
if always {
let mut sess = session.lock().await;
sess.auto_approve_tool(&pending.tool_name);
tracing::info!(
"Auto-approved tool '{}' for session {}",
pending.tool_name,
sess.id
);
}
// Reset thread state to processing
{
let mut sess = session.lock().await;
if let Some(thread) = sess.threads.get_mut(&thread_id) {
thread.state = ThreadState::Processing;
}
}
// Execute the approved tool and continue the loop
let mut job_ctx =
JobContext::with_user(&message.user_id, "chat", "Interactive chat session")
@@ -1460,17 +1454,25 @@ impl Agent {
);
{
let mut sess = session.lock().await;
if let Some(thread) = sess.threads.get_mut(&thread_id) {
thread.clear_pending_approval();
thread.complete_turn(&rejection);
// User message already persisted at turn start; save rejection response
self.persist_assistant_response(
thread_id,
&message.channel,
&message.user_id,
&rejection,
)
.await;
match sess.threads.get_mut(&thread_id) {
Some(thread) => {
thread.clear_pending_approval();
thread.complete_turn(&rejection);
// User message already persisted at turn start; save rejection response
self.persist_assistant_response(
thread_id,
&message.channel,
&message.user_id,
&rejection,
)
.await;
}
None => {
tracing::warn!(
%thread_id,
"Thread disappeared during approval rejection"
);
}
}
}
@@ -2012,6 +2014,124 @@ mod tests {
}
}
/// Regression test for #1486: verify that request_id mismatch atomically
/// restores the pending approval (no TOCTOU window where it could be lost).
#[test]
fn test_request_id_mismatch_preserves_pending_state() {
use crate::agent::session::{PendingApproval, Session, Thread, ThreadState};
use uuid::Uuid;
let session_id = Uuid::new_v4();
let thread_id = Uuid::new_v4();
let correct_request_id = Uuid::new_v4();
let wrong_request_id = Uuid::new_v4();
let mut thread = Thread::with_id(thread_id, session_id);
let pending = PendingApproval {
request_id: correct_request_id,
tool_name: "shell".to_string(),
parameters: serde_json::json!({"command": "rm -rf /"}),
display_parameters: serde_json::json!({"command": "[REDACTED]"}),
description: "Execute dangerous command".to_string(),
tool_call_id: "call_42".to_string(),
context_messages: vec![],
deferred_tool_calls: vec![],
user_timezone: None,
allow_always: false,
};
thread.await_approval(pending);
let mut session = Session::new("test-user");
session.threads.insert(thread_id, thread);
// Verify initial state
assert_eq!(
session.threads[&thread_id].state,
ThreadState::AwaitingApproval
);
assert!(session.threads[&thread_id].pending_approval.is_some());
// Simulate the atomic check-take-verify sequence from the fixed code:
// take the approval, then verify request_id, and restore on mismatch.
let thread = session.threads.get_mut(&thread_id).unwrap();
let taken = thread.take_pending_approval().unwrap();
assert!(thread.pending_approval.is_none(), "take should clear it");
// Wrong request_id -- restore atomically (within same "lock scope" in production)
assert_ne!(wrong_request_id, taken.request_id);
thread.await_approval(taken);
// Verify the pending approval is fully restored
assert_eq!(
session.threads[&thread_id].state,
ThreadState::AwaitingApproval
);
let restored = session.threads[&thread_id]
.pending_approval
.as_ref()
.expect("pending approval must be restored after request_id mismatch");
assert_eq!(restored.request_id, correct_request_id);
assert_eq!(restored.tool_name, "shell");
assert_eq!(restored.tool_call_id, "call_42");
}
/// Regression test for #1486: verify that approval with "always" flag
/// correctly transitions thread state and registers auto-approve in a
/// single lock scope.
#[test]
fn test_approval_with_always_sets_auto_approve_and_processing() {
use crate::agent::session::{PendingApproval, Session, Thread, ThreadState};
use uuid::Uuid;
let session_id = Uuid::new_v4();
let thread_id = Uuid::new_v4();
let request_id = Uuid::new_v4();
let mut thread = Thread::with_id(thread_id, session_id);
let pending = PendingApproval {
request_id,
tool_name: "web_fetch".to_string(),
parameters: serde_json::json!({"url": "https://example.com"}),
display_parameters: serde_json::json!({"url": "https://example.com"}),
description: "Fetch a web page".to_string(),
tool_call_id: "call_99".to_string(),
context_messages: vec![],
deferred_tool_calls: vec![],
user_timezone: None,
allow_always: true,
};
thread.await_approval(pending);
let mut session = Session::new("test-user");
session.threads.insert(thread_id, thread);
// Simulate the combined lock scope: take + verify + auto_approve + state transition
// Simulate the combined lock scope: take + verify + auto_approve + state transition.
// Use separate scopes to mirror the borrow pattern in production code.
let taken = {
let thread = session.threads.get_mut(&thread_id).unwrap();
thread.take_pending_approval().unwrap()
};
// Request ID matches
assert_eq!(taken.request_id, request_id);
// Auto-approve the tool (thread borrow dropped, so session is accessible)
session.auto_approve_tool(&taken.tool_name);
// Re-borrow thread and set state to Processing
{
let thread = session.threads.get_mut(&thread_id).unwrap();
thread.state = ThreadState::Processing;
}
// Verify both mutations happened atomically
assert!(session.is_tool_auto_approved("web_fetch"));
assert_eq!(session.threads[&thread_id].state, ThreadState::Processing);
// Pending approval should be consumed (not restored)
assert!(session.threads[&thread_id].pending_approval.is_none());
}
// Helper function to extract the approval message without needing a full Agent instance
fn extract_approval_message(
session: &crate::agent::session::Session,
-7
View File
@@ -196,12 +196,6 @@ pub struct JobContext {
pub tool_output_stash: Arc<tokio::sync::RwLock<HashMap<String, String>>>,
/// User's preferred timezone (IANA name, e.g. "America/New_York"). Defaults to "UTC".
pub user_timezone: String,
/// Current nesting depth for programmatic tool calling (PTC).
///
/// Tracks how deep we are in a tool-invokes-tool chain so the executor
/// can enforce MAX_NESTING_DEPTH globally, even across WASM→executor→WASM chains.
#[serde(skip)]
pub tool_nesting_depth: u32,
}
impl JobContext {
@@ -243,7 +237,6 @@ impl JobContext {
metadata: serde_json::Value::Null,
tool_output_stash: Arc::new(tokio::sync::RwLock::new(HashMap::new())),
user_timezone: "UTC".to_string(),
tool_nesting_depth: 0,
}
}
-1
View File
@@ -134,7 +134,6 @@ impl JobStore for LibSqlBackend {
// TODO(#661): persist user_timezone in agent_jobs table so
// background/routine jobs retain the session's timezone context.
user_timezone: "UTC".to_string(),
tool_nesting_depth: 0,
}))
}
None => Ok(None),
-1
View File
@@ -258,7 +258,6 @@ impl Store {
// TODO(#661): persist user_timezone in agent_jobs table so
// background/routine jobs retain the session's timezone context.
user_timezone: "UTC".to_string(),
tool_nesting_depth: 0,
}))
}
None => Ok(None),
+2
View File
@@ -1,5 +1,7 @@
//! Shared test helpers for OpenAI Codex provider tests.
#![cfg(test)]
use crate::config::OpenAiCodexConfig;
/// Build a minimal JWT for testing (header.payload.signature).
-2
View File
@@ -306,8 +306,6 @@ async fn async_main() -> anyhow::Result<()> {
&components.llm,
components.db.as_ref(),
components.secrets_store.as_ref(),
&components.tools,
&components.safety,
)
.await;
let container_job_manager = orch.container_job_manager;
+14 -552
View File
@@ -15,18 +15,15 @@ use tokio::sync::{Mutex, broadcast};
use uuid::Uuid;
use crate::channels::web::types::SseEvent;
use crate::context::JobContext;
use crate::db::Database;
use crate::llm::{CompletionRequest, LlmProvider, ToolCompletionRequest};
use crate::orchestrator::auth::{TokenStore, worker_auth_middleware};
use crate::orchestrator::job_manager::ContainerJobManager;
use crate::secrets::SecretsStore;
use crate::tools::ToolExecutor;
use crate::worker::api::JobEventPayload;
use crate::worker::api::{
CompletionReport, CredentialResponse, JobDescription, ProxyCompletionRequest,
ProxyCompletionResponse, ProxyToolCompletionRequest, ProxyToolCompletionResponse, StatusUpdate,
ToolCallRequest, ToolCallResponse,
};
/// A follow-up prompt queued for a Claude Code bridge.
@@ -52,8 +49,6 @@ pub struct OrchestratorState {
pub secrets_store: Option<Arc<dyn SecretsStore + Send + Sync>>,
/// User ID for secret lookups (single-tenant, typically "default").
pub user_id: String,
/// Tool executor for programmatic tool calling (PTC).
pub tool_executor: Option<Arc<ToolExecutor>>,
}
/// The orchestrator's internal API server.
@@ -75,7 +70,6 @@ impl OrchestratorApi {
.route("/worker/{job_id}/event", post(job_event_handler))
.route("/worker/{job_id}/prompt", get(get_prompt_handler))
.route("/worker/{job_id}/credentials", get(get_credentials_handler))
.route("/worker/{job_id}/tools/call", post(tool_call_handler))
.route_layer(axum::middleware::from_fn_with_state(
state.token_store.clone(),
worker_auth_middleware,
@@ -297,26 +291,20 @@ async fn job_event_handler(
.unwrap_or("")
.to_string(),
},
"tool_use" => {
// Redact raw parameters from worker-reported tool_use events
// before broadcasting via SSE. Workers are untrusted and may
// include sensitive data (API keys, passwords, PII) in the
// input payload. We replace it with a placeholder to prevent
// leaking secrets to the web UI.
let redacted_input = serde_json::json!({
"_note": "parameters redacted for security"
});
SseEvent::JobToolUse {
job_id: job_id_str,
tool_name: payload
.data
.get("tool_name")
.and_then(|v| v.as_str())
.unwrap_or("unknown")
.to_string(),
input: redacted_input,
}
}
"tool_use" => SseEvent::JobToolUse {
job_id: job_id_str,
tool_name: payload
.data
.get("tool_name")
.and_then(|v| v.as_str())
.unwrap_or("unknown")
.to_string(),
input: payload
.data
.get("input")
.cloned()
.unwrap_or(serde_json::Value::Null),
},
"tool_result" => SseEvent::JobToolResult {
job_id: job_id_str,
tool_name: payload
@@ -455,106 +443,6 @@ async fn get_credentials_handler(
))
}
/// Execute a tool programmatically on behalf of a container worker (PTC).
///
/// Builds a minimal `JobContext` from the job metadata and delegates to
/// `ToolExecutor::execute`. Emits SSE events for tool_use/tool_result so
/// the web UI can observe PTC calls.
async fn tool_call_handler(
State(state): State<OrchestratorState>,
Path(job_id): Path<Uuid>,
Json(req): Json<ToolCallRequest>,
) -> Result<Json<ToolCallResponse>, StatusCode> {
let executor = state
.tool_executor
.as_ref()
.ok_or(StatusCode::SERVICE_UNAVAILABLE)?;
tracing::debug!(
job_id = %job_id,
tool = %req.tool_name,
"PTC tool call request"
);
// Build a minimal JobContext for the tool execution
let mut ctx = JobContext::with_user(
state.user_id.clone(),
format!("PTC call: {}", req.tool_name),
format!("Programmatic tool call from job {}", job_id),
);
// Do not trust client-provided nesting_depth — a malicious worker
// could send any value to bypass the limit. The orchestrator must
// increment the depth server-side: each hop through the orchestrator
// adds 1. This way even if a worker always sends 0, the depth still
// increases with each real nesting level.
ctx.tool_nesting_depth = req.nesting_depth.saturating_add(1);
// Emit tool_use SSE event with redacted parameters to avoid leaking
// sensitive data (API keys, passwords, PII) to the web UI.
if let Some(ref tx) = state.job_event_tx {
let redacted_params = serde_json::json!({
"_note": "parameters redacted for security"
});
let _ = tx.send((
job_id,
SseEvent::JobToolUse {
job_id: job_id.to_string(),
tool_name: req.tool_name.clone(),
input: redacted_params,
},
));
}
// Determine timeout override
let timeout_override = req
.timeout_secs
.map(|s| std::time::Duration::from_secs(s.min(300)));
// Execute the tool
match executor
.execute(&req.tool_name, req.parameters, &ctx, timeout_override)
.await
{
Ok(result) => {
// Emit tool_result SSE event
if let Some(ref tx) = state.job_event_tx {
let _ = tx.send((
job_id,
SseEvent::JobToolResult {
job_id: job_id.to_string(),
tool_name: req.tool_name.clone(),
output: result.output.clone(),
},
));
}
Ok(Json(ToolCallResponse {
success: true,
output: Some(result.output),
error: None,
duration_ms: result.duration.as_millis() as u64,
was_sanitized: result.was_sanitized,
}))
}
Err(e) => {
tracing::warn!(
job_id = %job_id,
tool = %req.tool_name,
error = %e,
"PTC tool call failed"
);
Ok(Json(ToolCallResponse {
success: false,
output: None,
error: Some(e.to_string()),
duration_ms: 0,
was_sanitized: false,
}))
}
}
}
fn format_finish_reason(reason: crate::llm::FinishReason) -> String {
match reason {
crate::llm::FinishReason::Stop => "stop".to_string(),
@@ -592,7 +480,6 @@ mod tests {
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: None,
}
}
@@ -822,7 +709,6 @@ mod tests {
store: None,
secrets_store: Some(secrets_store),
user_id: "default".to_string(),
tool_executor: None,
};
let router = OrchestratorApi::router(state);
@@ -858,7 +744,6 @@ mod tests {
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: None,
};
let job_id = Uuid::new_v4();
@@ -914,7 +799,6 @@ mod tests {
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: None,
};
let job_id = Uuid::new_v4();
@@ -963,7 +847,6 @@ mod tests {
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: None,
};
let job_id = Uuid::new_v4();
@@ -1043,425 +926,4 @@ mod tests {
assert_eq!(handle.worker_iteration, 5);
assert_eq!(handle.last_worker_status.as_deref(), Some("Iteration 5"));
}
// -- Programmatic tool calling (PTC) tests --
use std::time::Duration;
use crate::config::SafetyConfig;
use crate::context::JobContext;
use crate::safety::SafetyLayer;
use crate::tools::{Tool, ToolError, ToolExecutor, ToolOutput, ToolRegistry};
/// A tool that sleeps for 10 seconds (used to test timeout enforcement).
struct SlowTool;
#[async_trait::async_trait]
impl Tool for SlowTool {
fn name(&self) -> &str {
"slow_tool"
}
fn description(&self) -> &str {
"A tool that sleeps"
}
fn parameters_schema(&self) -> serde_json::Value {
serde_json::json!({"type": "object"})
}
async fn execute(
&self,
_params: serde_json::Value,
_ctx: &JobContext,
) -> Result<ToolOutput, ToolError> {
tokio::time::sleep(Duration::from_secs(10)).await;
Ok(ToolOutput::text("done", Duration::from_secs(10)))
}
fn requires_sanitization(&self) -> bool {
false
}
}
/// Build an `OrchestratorState` with a real `ToolExecutor` wired in.
///
/// Also returns the broadcast receiver when `with_broadcast` is true,
/// so SSE-related tests can observe emitted events.
fn test_state_with_executor(
with_broadcast: bool,
) -> (
OrchestratorState,
Option<broadcast::Receiver<(Uuid, SseEvent)>>,
) {
let tools = Arc::new(ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(SafetyLayer::new(&SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
}));
let executor = ToolExecutor::new(Arc::clone(&tools), safety, Duration::from_secs(60));
let token_store = TokenStore::new();
let jm = ContainerJobManager::new(ContainerJobConfig::default(), token_store.clone());
let (tx, rx) = if with_broadcast {
let (tx, rx) = broadcast::channel(16);
(Some(tx), Some(rx))
} else {
(None, None)
};
let state = OrchestratorState {
llm: Arc::new(StubLlm::default()),
job_manager: Arc::new(jm),
token_store,
job_event_tx: tx,
prompt_queue: Arc::new(Mutex::new(HashMap::new())),
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: Some(Arc::new(executor)),
};
(state, rx)
}
#[tokio::test]
async fn tool_call_echo_success() {
let (state, _) = test_state_with_executor(false);
let job_id = Uuid::new_v4();
let token = state.token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "echo",
"parameters": {"message": "hello"},
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::OK);
let body = axum::body::to_bytes(resp.into_body(), 4096).await.unwrap();
let json: serde_json::Value = serde_json::from_slice(&body).unwrap();
assert_eq!(json["success"], true);
assert!(
json["output"]
.as_str()
.map(|s| s.contains("hello"))
.unwrap_or(false),
"output should contain 'hello', got: {:?}",
json["output"]
);
assert!(
json["duration_ms"].is_u64(),
"duration_ms should be present as a number"
);
}
#[tokio::test]
async fn tool_call_not_found() {
let (state, _) = test_state_with_executor(false);
let job_id = Uuid::new_v4();
let token = state.token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "nonexistent_tool",
"parameters": {},
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
// Handler returns Ok(Json(...)) even on tool failure
assert_eq!(resp.status(), StatusCode::OK);
let body = axum::body::to_bytes(resp.into_body(), 4096).await.unwrap();
let json: serde_json::Value = serde_json::from_slice(&body).unwrap();
assert_eq!(json["success"], false);
assert!(
json["error"]
.as_str()
.map(|s| s.to_lowercase().contains("not found"))
.unwrap_or(false),
"error should mention 'not found', got: {:?}",
json["error"]
);
}
#[tokio::test]
async fn tool_call_no_executor() {
// Use regular test_state() which has tool_executor: None
let state = test_state();
let job_id = Uuid::new_v4();
let token = state.token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "echo",
"parameters": {"message": "hello"},
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::SERVICE_UNAVAILABLE);
}
#[tokio::test]
async fn tool_call_with_sse_events() {
let (state, rx) = test_state_with_executor(true);
let mut rx = rx.expect("broadcast receiver should be present");
let job_id = Uuid::new_v4();
let token = state.token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "echo",
"parameters": {"message": "hello"},
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::OK);
// Collect events from broadcast channel
let mut saw_tool_use = false;
let mut saw_tool_result = false;
while let Ok((recv_id, event)) = rx.try_recv() {
assert_eq!(recv_id, job_id);
match event {
SseEvent::JobToolUse { tool_name, .. } => {
assert_eq!(tool_name, "echo");
saw_tool_use = true;
}
SseEvent::JobToolResult { tool_name, .. } => {
assert_eq!(tool_name, "echo");
saw_tool_result = true;
}
_ => {}
}
}
assert!(saw_tool_use, "should have emitted JobToolUse event");
assert!(saw_tool_result, "should have emitted JobToolResult event");
}
#[tokio::test]
async fn tool_call_with_timeout() {
// Build a registry that includes our SlowTool
let tools = Arc::new(ToolRegistry::new());
tools.register(Arc::new(SlowTool)).await;
let safety = Arc::new(SafetyLayer::new(&SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
}));
let executor = ToolExecutor::new(Arc::clone(&tools), safety, Duration::from_secs(60));
let token_store = TokenStore::new();
let jm = ContainerJobManager::new(ContainerJobConfig::default(), token_store.clone());
let state = OrchestratorState {
llm: Arc::new(StubLlm::default()),
job_manager: Arc::new(jm),
token_store: token_store.clone(),
job_event_tx: None,
prompt_queue: Arc::new(Mutex::new(HashMap::new())),
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: Some(Arc::new(executor)),
};
let job_id = Uuid::new_v4();
let token = token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "slow_tool",
"parameters": {},
"timeout_secs": 1,
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::OK);
let body = axum::body::to_bytes(resp.into_body(), 4096).await.unwrap();
let json: serde_json::Value = serde_json::from_slice(&body).unwrap();
assert_eq!(json["success"], false);
assert!(
json["error"]
.as_str()
.map(|s| {
let lower = s.to_lowercase();
lower.contains("timed out") || lower.contains("timeout")
})
.unwrap_or(false),
"error should mention timeout, got: {:?}",
json["error"]
);
}
#[tokio::test]
async fn tool_call_auth_required() {
let (state, _) = test_state_with_executor(false);
let job_id = Uuid::new_v4();
// Do NOT create a token -- request should be rejected
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "echo",
"parameters": {"message": "hello"},
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
// No Authorization header
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::UNAUTHORIZED);
}
#[tokio::test]
async fn tool_call_nesting_depth_incremented_server_side() {
// A worker sending nesting_depth=4 should get depth=5 after the
// orchestrator increments it. With MAX_NESTING_DEPTH=5, this
// should be rejected (depth >= max).
let (state, _) = test_state_with_executor(false);
let job_id = Uuid::new_v4();
let token = state.token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
let payload = serde_json::json!({
"tool_name": "echo",
"parameters": {"message": "hello"},
"nesting_depth": 4,
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/tools/call", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::OK);
let body = axum::body::to_bytes(resp.into_body(), 4096).await.unwrap();
let json: serde_json::Value = serde_json::from_slice(&body).unwrap();
assert_eq!(json["success"], false);
assert!(
json["error"]
.as_str()
.map(|s| s.to_lowercase().contains("nesting"))
.unwrap_or(false),
"error should mention nesting depth, got: {:?}",
json["error"]
);
}
#[tokio::test]
async fn job_event_tool_use_redacts_input() {
// Worker-reported tool_use events must have their input redacted
// before SSE broadcast to prevent leaking sensitive parameters.
let (tx, mut rx) = broadcast::channel(16);
let token_store = TokenStore::new();
let jm = ContainerJobManager::new(ContainerJobConfig::default(), token_store.clone());
let state = OrchestratorState {
llm: Arc::new(StubLlm::default()),
job_manager: Arc::new(jm),
token_store: token_store.clone(),
job_event_tx: Some(tx),
prompt_queue: Arc::new(Mutex::new(HashMap::new())),
store: None,
secrets_store: None,
user_id: "default".to_string(),
tool_executor: None,
};
let job_id = Uuid::new_v4();
let token = token_store.create_token(job_id).await;
let router = OrchestratorApi::router(state);
// Worker sends a tool_use event with sensitive data in input
let payload = serde_json::json!({
"event_type": "tool_use",
"data": {
"tool_name": "shell",
"input": {"command": "curl -H 'Authorization: Bearer sk-secret-key' https://api.example.com"}
}
});
let req = Request::builder()
.method("POST")
.uri(format!("/worker/{}/event", job_id))
.header("Authorization", format!("Bearer {}", token))
.header("Content-Type", "application/json")
.body(Body::from(serde_json::to_vec(&payload).unwrap()))
.unwrap();
let resp = router.oneshot(req).await.unwrap();
assert_eq!(resp.status(), StatusCode::OK);
let (_recv_id, event) = rx.recv().await.unwrap();
match event {
SseEvent::JobToolUse {
tool_name, input, ..
} => {
assert_eq!(tool_name, "shell");
// The input must be redacted, not the raw worker payload
assert!(
input.get("_note").is_some(),
"input should be redacted placeholder, got: {}",
input
);
assert!(
!input.to_string().contains("sk-secret-key"),
"input must not contain sensitive data"
);
}
other => panic!("Expected JobToolUse, got {:?}", other),
}
}
}
-16
View File
@@ -49,9 +49,7 @@ use uuid::Uuid;
use crate::channels::web::types::SseEvent;
use crate::db::Database;
use crate::llm::LlmProvider;
use crate::safety::SafetyLayer;
use crate::secrets::SecretsStore;
use crate::tools::{ToolExecutor, ToolRegistry};
/// Resolve the orchestrator port from the `ORCHESTRATOR_PORT` environment
/// variable, falling back to 50051.
@@ -77,8 +75,6 @@ pub async fn setup_orchestrator(
llm: &Arc<dyn LlmProvider>,
db: Option<&Arc<dyn Database>>,
secrets_store: Option<&Arc<dyn SecretsStore + Send + Sync>>,
tools: &Arc<ToolRegistry>,
safety: &Arc<SafetyLayer>,
) -> OrchestratorSetup {
let prompt_queue = Arc::new(Mutex::new(
HashMap::<Uuid, VecDeque<api::PendingPrompt>>::new(),
@@ -129,17 +125,6 @@ pub async fn setup_orchestrator(
};
let jm = Arc::new(ContainerJobManager::new(job_config, token_store.clone()));
// Build ToolExecutor for programmatic tool calling (PTC)
let tool_executor = Arc::new(ToolExecutor::new(
Arc::clone(tools),
Arc::clone(safety),
std::time::Duration::from_secs(60),
));
// Wire the executor into the shared slot so WASM tools registered
// during build_all() can resolve it lazily at execution time.
tools.set_tool_executor(Arc::clone(&tool_executor));
let orchestrator_state = api::OrchestratorState {
llm: Arc::clone(llm),
job_manager: Arc::clone(&jm),
@@ -149,7 +134,6 @@ pub async fn setup_orchestrator(
store: db.cloned(),
secrets_store: secrets_store.cloned(),
user_id: "default".to_string(),
tool_executor: Some(tool_executor),
};
tokio::spawn(async move {
-2
View File
@@ -9,7 +9,6 @@ mod json;
mod memory;
mod message;
pub mod path_utils;
pub mod ptc_script;
mod restart;
pub mod routine;
pub mod secrets_tools;
@@ -32,7 +31,6 @@ pub use job::{
pub use json::JsonTool;
pub use memory::{MemoryReadTool, MemorySearchTool, MemoryTreeTool, MemoryWriteTool};
pub use message::MessageTool;
pub use ptc_script::PtcScriptTool;
pub use restart::RestartTool;
pub use routine::{
EventEmitTool, RoutineCreateTool, RoutineDeleteTool, RoutineFireTool, RoutineHistoryTool,
-371
View File
@@ -1,371 +0,0 @@
//! PTC script tool for running multi-step Python programs that call tools.
//!
//! Wraps user-provided Python code in a preamble that imports the IronClaw
//! SDK (`ironclaw_tools`), then executes it via `python3 -c`. The script
//! runs in the same environment as the worker container and can call any
//! registered tool through the SDK's `call_tool()` function.
use std::process::Stdio;
use std::time::Duration;
use async_trait::async_trait;
use tokio::io::AsyncReadExt;
use tokio::process::Command;
use crate::context::JobContext;
use crate::tools::tool::{
ApprovalRequirement, Tool, ToolDomain, ToolError, ToolOutput, require_str,
};
/// Maximum output size before truncation (64KB).
const MAX_OUTPUT_SIZE: usize = 64 * 1024;
/// Default script timeout.
const DEFAULT_TIMEOUT_SECS: u64 = 120;
/// Maximum allowed timeout.
const MAX_TIMEOUT_SECS: u64 = 300;
/// Environment variables safe to forward to the Python subprocess.
const SAFE_ENV_VARS: &[&str] = &[
"PATH",
"HOME",
"USER",
"LOGNAME",
"SHELL",
"TERM",
"LANG",
"LC_ALL",
"LC_CTYPE",
"PWD",
"TMPDIR",
"TMP",
"TEMP",
"CARGO_HOME",
"RUSTUP_HOME",
"NODE_PATH",
"NPM_CONFIG_PREFIX",
];
/// PTC environment variables required by the ironclaw_tools SDK.
const PTC_ENV_VARS: &[&str] = &[
"IRONCLAW_ORCHESTRATOR_URL",
"IRONCLAW_JOB_ID",
"IRONCLAW_WORKER_TOKEN",
];
/// Python preamble injected before the user's script.
const PREAMBLE: &str = r#"
import json, sys, os
# Import IronClaw SDK
from ironclaw_tools import call_tool, shell, read_file, write_file, http_get
# Structured output collector
_ptc_outputs = {}
def ptc_output(key, value):
"""Register a named output value for structured results."""
_ptc_outputs[key] = value
try:
"#;
/// Python postamble appended after the user's script.
const POSTAMBLE: &str = r#"
except Exception as _ptc_err:
print(f"SCRIPT_ERROR: {type(_ptc_err).__name__}: {_ptc_err}", file=sys.stderr)
sys.exit(1)
# Print structured outputs if any were registered
if _ptc_outputs:
print("\n__PTC_OUTPUTS__")
print(json.dumps(_ptc_outputs))
"#;
pub struct PtcScriptTool;
impl Default for PtcScriptTool {
fn default() -> Self {
Self
}
}
impl PtcScriptTool {
pub fn new() -> Self {
Self
}
/// Build the full Python program from user script + preamble/postamble.
fn build_program(script: &str) -> String {
let mut program =
String::with_capacity(PREAMBLE.len() + script.len() + POSTAMBLE.len() + 256);
program.push_str(PREAMBLE);
// Indent user script into the try: block
for line in script.lines() {
program.push_str(" ");
program.push_str(line);
program.push('\n');
}
program.push_str(POSTAMBLE);
program
}
/// Truncate output to MAX_OUTPUT_SIZE with a truncation notice.
fn truncate_output(output: &str) -> String {
if output.len() <= MAX_OUTPUT_SIZE {
output.to_string()
} else {
let mut i = MAX_OUTPUT_SIZE;
while i > 0 && !output.is_char_boundary(i) {
i -= 1;
}
format!(
"{}\n\n[Output truncated at {} bytes]",
&output[..i],
MAX_OUTPUT_SIZE
)
}
}
}
#[async_trait]
impl Tool for PtcScriptTool {
fn name(&self) -> &str {
"ptc_script"
}
fn description(&self) -> &str {
"Execute a Python script that can call IronClaw tools programmatically. \
The script has access to call_tool(), shell(), read_file(), write_file(), \
and http_get() from the ironclaw_tools SDK. Use ptc_output(key, value) \
to return structured results."
}
fn parameters_schema(&self) -> serde_json::Value {
serde_json::json!({
"type": "object",
"properties": {
"script": {
"type": "string",
"description": "Python script to execute. Has access to call_tool(), shell(), read_file(), write_file(), http_get(), and ptc_output()."
},
"timeout_secs": {
"type": "integer",
"description": "Timeout in seconds (default 120, max 300).",
"default": 120,
"minimum": 1,
"maximum": 300
}
},
"required": ["script"]
})
}
async fn execute(
&self,
params: serde_json::Value,
ctx: &JobContext,
) -> Result<ToolOutput, ToolError> {
let start = std::time::Instant::now();
let script = require_str(&params, "script")?;
let timeout_secs = params
.get("timeout_secs")
.and_then(|v| v.as_u64())
.unwrap_or(DEFAULT_TIMEOUT_SECS)
.min(MAX_TIMEOUT_SECS);
let timeout = Duration::from_secs(timeout_secs);
let program = Self::build_program(script);
// Build the subprocess command
let mut command = Command::new("python3");
command.args(["-c", &program]);
// Scrub environment -- only forward safe vars + PTC vars + extra_env
command.env_clear();
for var in SAFE_ENV_VARS {
if let Ok(val) = std::env::var(var) {
command.env(var, val);
}
}
for var in PTC_ENV_VARS {
if let Ok(val) = std::env::var(var) {
command.env(var, val);
}
}
// Forward extra_env from JobContext (credentials fetched by worker runtime)
for (k, v) in ctx.extra_env.iter() {
command.env(k, v);
}
command
.stdin(Stdio::null())
.stdout(Stdio::piped())
.stderr(Stdio::piped());
// Spawn and drain stdout/stderr concurrently
let mut child = command
.spawn()
.map_err(|e| ToolError::ExecutionFailed(format!("Failed to spawn python3: {}", e)))?;
let stdout_handle = child.stdout.take();
let stderr_handle = child.stderr.take();
let result = tokio::time::timeout(timeout, async {
let stdout_fut = async {
if let Some(mut out) = stdout_handle {
let mut buf = Vec::new();
(&mut out)
.take(MAX_OUTPUT_SIZE as u64)
.read_to_end(&mut buf)
.await
.ok();
tokio::io::copy(&mut out, &mut tokio::io::sink()).await.ok();
String::from_utf8_lossy(&buf).to_string()
} else {
String::new()
}
};
let stderr_fut = async {
if let Some(mut err) = stderr_handle {
let mut buf = Vec::new();
(&mut err)
.take(MAX_OUTPUT_SIZE as u64)
.read_to_end(&mut buf)
.await
.ok();
tokio::io::copy(&mut err, &mut tokio::io::sink()).await.ok();
String::from_utf8_lossy(&buf).to_string()
} else {
String::new()
}
};
let (stdout, stderr, wait_result) = tokio::join!(stdout_fut, stderr_fut, child.wait());
let status = wait_result?;
Ok::<_, std::io::Error>((stdout, stderr, status.code().unwrap_or(-1)))
})
.await;
let duration = start.elapsed();
match result {
Ok(Ok((stdout, stderr, exit_code))) => {
if exit_code != 0 {
let error_msg = if stderr.is_empty() {
format!("Script exited with code {}", exit_code)
} else {
format!(
"Script exited with code {}:\n{}",
exit_code,
Self::truncate_output(&stderr)
)
};
return Err(ToolError::ExecutionFailed(error_msg));
}
// Combine output
let output = if stderr.is_empty() {
stdout
} else {
format!("{}\n\n--- stderr ---\n{}", stdout, stderr)
};
Ok(ToolOutput::text(Self::truncate_output(&output), duration))
}
Ok(Err(e)) => Err(ToolError::ExecutionFailed(format!(
"Script execution failed: {}",
e
))),
Err(_) => {
let _ = child.kill().await;
Err(ToolError::Timeout(timeout))
}
}
}
fn requires_sanitization(&self) -> bool {
true
}
fn requires_approval(&self, _params: &serde_json::Value) -> ApprovalRequirement {
ApprovalRequirement::Always
}
fn domain(&self) -> ToolDomain {
ToolDomain::Container
}
fn execution_timeout(&self) -> Duration {
Duration::from_secs(MAX_TIMEOUT_SECS)
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_build_program_indents_script() {
let script = "x = 1\nprint(x)";
let program = PtcScriptTool::build_program(script);
assert!(program.contains(" x = 1\n"));
assert!(program.contains(" print(x)\n"));
assert!(program.contains("from ironclaw_tools import"));
assert!(program.contains("def ptc_output("));
}
#[test]
fn test_build_program_empty_script() {
let program = PtcScriptTool::build_program("");
// Empty script should still have preamble + postamble
assert!(program.contains("try:"));
assert!(program.contains("except Exception"));
}
#[test]
fn test_truncate_output() {
let short = "hello";
assert_eq!(PtcScriptTool::truncate_output(short), "hello");
let long = "x".repeat(MAX_OUTPUT_SIZE + 100);
let truncated = PtcScriptTool::truncate_output(&long);
assert!(truncated.len() < long.len());
assert!(truncated.contains("[Output truncated"));
}
#[test]
fn test_truncate_output_multibyte_boundary() {
// Build a string of multi-byte chars (emoji = 4 bytes each) that crosses MAX_OUTPUT_SIZE
let emoji = "\u{1F600}"; // 4 bytes
let count = MAX_OUTPUT_SIZE / emoji.len() + 10;
let long: String = emoji.repeat(count);
assert!(long.len() > MAX_OUTPUT_SIZE);
let truncated = PtcScriptTool::truncate_output(&long);
// Must not panic and must contain valid UTF-8
assert!(truncated.contains("[Output truncated"));
// The kept portion must end on a char boundary (valid UTF-8 guaranteed by compilation)
let kept = truncated.split("\n\n[Output truncated").next().unwrap();
assert!(kept.len() <= MAX_OUTPUT_SIZE);
// Every char should be complete (no partial emoji)
assert!(kept.chars().all(|c| c == '\u{1F600}'));
}
#[test]
fn test_tool_metadata() {
let tool = PtcScriptTool::new();
assert_eq!(tool.name(), "ptc_script");
assert_eq!(tool.domain(), ToolDomain::Container);
assert_eq!(
tool.requires_approval(&serde_json::json!({})),
ApprovalRequirement::Always
);
assert!(tool.requires_sanitization());
}
}
-490
View File
@@ -1,490 +0,0 @@
//! Tool executor for programmatic tool calling (PTC).
//!
//! Provides a standalone execution engine that can be used by both the
//! Docker HTTP RPC path (orchestrator endpoint) and the WASM host function
//! path (tool_invoke). Extracts the tool dispatch flow into a reusable struct.
use std::sync::Arc;
use std::time::{Duration, Instant};
use crate::context::JobContext;
use crate::safety::SafetyLayer;
use crate::tools::registry::ToolRegistry;
use crate::tools::tool::ToolDomain;
/// Maximum allowed nesting depth for tool-invokes-tool chains.
pub const MAX_NESTING_DEPTH: u32 = 5;
/// Maximum per-call timeout (5 minutes).
const MAX_TIMEOUT_SECS: u64 = 300;
/// Result of a programmatic tool call.
#[derive(Debug, Clone)]
pub struct PtcToolResult {
/// Tool output (potentially sanitized).
pub output: String,
/// Whether the output was modified by the safety layer.
pub was_sanitized: bool,
/// Wall-clock duration of the tool execution.
pub duration: Duration,
}
/// Errors that can occur during programmatic tool execution.
#[derive(Debug, thiserror::Error)]
pub enum PtcError {
#[error("Tool not found: {name}")]
NotFound { name: String },
#[error("Tool execution failed: {name}: {reason}")]
ExecutionFailed { name: String, reason: String },
#[error("Tool execution timed out: {name} (timeout: {timeout:?})")]
Timeout { name: String, timeout: Duration },
#[error("Invalid parameters for tool {name}: {reason}")]
InvalidParameters { name: String, reason: String },
#[error("Tool {name} is rate limited")]
RateLimited { name: String },
#[error("Tool output blocked by safety layer: {reason}")]
SafetyBlocked { reason: String },
#[error("Nesting depth exceeded (max {max})")]
NestingDepthExceeded { max: u32 },
#[error("Tool {name} has domain Container and cannot be executed on the orchestrator")]
DomainBlocked { name: String },
}
/// Standalone tool execution engine for programmatic tool calling.
///
/// Used by:
/// - The orchestrator's `POST /worker/{job_id}/tools/call` endpoint
/// - The WASM `tool_invoke` host function
pub struct ToolExecutor {
tools: Arc<ToolRegistry>,
safety: Arc<SafetyLayer>,
default_timeout: Duration,
}
impl ToolExecutor {
/// Create a new tool executor.
pub fn new(
tools: Arc<ToolRegistry>,
safety: Arc<SafetyLayer>,
default_timeout: Duration,
) -> Self {
Self {
tools,
safety,
default_timeout,
}
}
/// Execute a tool by name with the given parameters.
///
/// Flow: lookup -> execute with timeout -> sanitize output -> return.
pub async fn execute(
&self,
tool_name: &str,
params: serde_json::Value,
ctx: &JobContext,
timeout_override: Option<Duration>,
) -> Result<PtcToolResult, PtcError> {
// Enforce global nesting depth limit
if ctx.tool_nesting_depth >= MAX_NESTING_DEPTH {
return Err(PtcError::NestingDepthExceeded {
max: MAX_NESTING_DEPTH,
});
}
let start = Instant::now();
// Look up the tool
let tool = self
.tools
.get(tool_name)
.await
.ok_or_else(|| PtcError::NotFound {
name: tool_name.to_string(),
})?;
// Reject Container-domain tools — they must run inside a sandbox,
// not on the orchestrator host. Without this check a compromised
// worker could invoke shell/file tools on the host (sandbox escape).
if tool.domain() == ToolDomain::Container {
return Err(PtcError::DomainBlocked {
name: tool_name.to_string(),
});
}
// Determine timeout: caller override -> tool's own timeout -> default,
// capped at MAX_TIMEOUT_SECS.
let timeout = timeout_override
.unwrap_or_else(|| tool.execution_timeout())
.min(Duration::from_secs(MAX_TIMEOUT_SECS));
// Execute with timeout
let tool_result = tokio::time::timeout(timeout, tool.execute(params, ctx))
.await
.map_err(|_| PtcError::Timeout {
name: tool_name.to_string(),
timeout,
})?
.map_err(|e| match e {
crate::tools::ToolError::InvalidParameters(reason) => PtcError::InvalidParameters {
name: tool_name.to_string(),
reason,
},
crate::tools::ToolError::RateLimited(_) => PtcError::RateLimited {
name: tool_name.to_string(),
},
other => PtcError::ExecutionFailed {
name: tool_name.to_string(),
reason: other.to_string(),
},
})?;
// Get output string
let raw_output = tool_result
.raw
.as_deref()
.or_else(|| tool_result.result.as_str())
.unwrap_or("")
.to_string();
let raw_output = if raw_output.is_empty() {
serde_json::to_string(&tool_result.result).unwrap_or_default()
} else {
raw_output
};
// Sanitize output if the tool requires it
let (output, was_sanitized) = if tool.requires_sanitization() {
let sanitized = self.safety.sanitize_tool_output(tool_name, &raw_output);
if sanitized.was_modified && sanitized.content.starts_with("[Output blocked") {
return Err(PtcError::SafetyBlocked {
reason: sanitized.content,
});
}
(sanitized.content, sanitized.was_modified)
} else {
(raw_output, false)
};
Ok(PtcToolResult {
output,
was_sanitized,
duration: start.elapsed(),
})
}
}
impl std::fmt::Debug for ToolExecutor {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("ToolExecutor")
.field("default_timeout", &self.default_timeout)
.finish()
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::safety::SafetyLayer;
use crate::tools::tool::{Tool, ToolDomain, ToolError, ToolOutput};
fn test_safety_config() -> crate::config::SafetyConfig {
crate::config::SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
}
}
struct SlowTool;
#[async_trait::async_trait]
impl Tool for SlowTool {
fn name(&self) -> &str {
"slow_tool"
}
fn description(&self) -> &str {
"A tool that sleeps"
}
fn parameters_schema(&self) -> serde_json::Value {
serde_json::json!({"type": "object"})
}
async fn execute(
&self,
_params: serde_json::Value,
_ctx: &JobContext,
) -> Result<ToolOutput, ToolError> {
tokio::time::sleep(Duration::from_secs(10)).await;
Ok(ToolOutput::text("done", Duration::from_secs(10)))
}
fn requires_sanitization(&self) -> bool {
false
}
}
#[tokio::test]
async fn test_execute_not_found() {
let tools = Arc::new(ToolRegistry::new());
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let result = executor
.execute("nonexistent", serde_json::json!({}), &ctx, None)
.await;
assert!(matches!(result, Err(PtcError::NotFound { .. })));
}
#[tokio::test]
async fn test_execute_echo() {
let tools = Arc::new(ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let result = executor
.execute("echo", serde_json::json!({"message": "hello"}), &ctx, None)
.await;
assert!(result.is_ok());
let ptc_result = result.as_ref().ok();
assert!(ptc_result.is_some());
assert!(
ptc_result
.map(|r| r.output.contains("hello"))
.unwrap_or(false)
);
}
#[tokio::test]
async fn test_execute_timeout() {
let tools = Arc::new(ToolRegistry::new());
tools.register(Arc::new(SlowTool)).await;
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let result = executor
.execute(
"slow_tool",
serde_json::json!({}),
&ctx,
Some(Duration::from_millis(50)),
)
.await;
assert!(matches!(result, Err(PtcError::Timeout { .. })));
}
#[tokio::test]
async fn test_nesting_depth_exceeded() {
let tools = Arc::new(ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let mut ctx = JobContext::new("test", "test");
ctx.tool_nesting_depth = MAX_NESTING_DEPTH; // already at max
let result = executor
.execute("echo", serde_json::json!({"message": "hello"}), &ctx, None)
.await;
assert!(matches!(result, Err(PtcError::NestingDepthExceeded { .. })));
}
#[tokio::test]
async fn test_nesting_depth_within_limit() {
let tools = Arc::new(ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let mut ctx = JobContext::new("test", "test");
ctx.tool_nesting_depth = MAX_NESTING_DEPTH - 1; // one below max
let result = executor
.execute("echo", serde_json::json!({"message": "hello"}), &ctx, None)
.await;
assert!(result.is_ok());
}
struct LeakyTool;
#[async_trait::async_trait]
impl Tool for LeakyTool {
fn name(&self) -> &str {
"leaky_tool"
}
fn description(&self) -> &str {
"Returns output with fake bearer token"
}
fn parameters_schema(&self) -> serde_json::Value {
serde_json::json!({"type": "object"})
}
async fn execute(
&self,
_params: serde_json::Value,
_ctx: &JobContext,
) -> Result<ToolOutput, ToolError> {
// Bearer token pattern triggers LeakAction::Redact (not Block),
// so the safety layer redacts it and returns sanitized output.
let output =
"Here is some data: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9_longtokenvalue end";
Ok(ToolOutput::text(output, Duration::from_millis(1)))
}
fn requires_sanitization(&self) -> bool {
true
}
}
struct InvalidParamsTool;
#[async_trait::async_trait]
impl Tool for InvalidParamsTool {
fn name(&self) -> &str {
"invalid_params_tool"
}
fn description(&self) -> &str {
"Always fails with InvalidParameters"
}
fn parameters_schema(&self) -> serde_json::Value {
serde_json::json!({"type": "object"})
}
async fn execute(
&self,
_params: serde_json::Value,
_ctx: &JobContext,
) -> Result<ToolOutput, ToolError> {
Err(ToolError::InvalidParameters("bad params".to_string()))
}
fn requires_sanitization(&self) -> bool {
false
}
}
#[tokio::test]
async fn test_execute_safety_sanitization() {
let tools = Arc::new(ToolRegistry::new());
tools.register(Arc::new(LeakyTool)).await;
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let result = executor
.execute("leaky_tool", serde_json::json!({}), &ctx, None)
.await;
// The safety layer should detect the API key pattern and modify the output
assert!(result.is_ok());
let ptc_result = result.unwrap();
assert!(
ptc_result.was_sanitized,
"Output with API key should be sanitized"
);
}
#[tokio::test]
async fn test_execute_invalid_params() {
let tools = Arc::new(ToolRegistry::new());
tools.register(Arc::new(InvalidParamsTool)).await;
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let result = executor
.execute("invalid_params_tool", serde_json::json!({}), &ctx, None)
.await;
match result {
Err(PtcError::InvalidParameters { name, reason }) => {
assert_eq!(name, "invalid_params_tool");
assert!(reason.contains("bad params"));
}
other => panic!("Expected InvalidParameters, got {:?}", other),
}
}
/// A tool that declares Container domain — must be blocked by the executor.
struct ContainerDomainTool;
#[async_trait::async_trait]
impl Tool for ContainerDomainTool {
fn name(&self) -> &str {
"container_tool"
}
fn description(&self) -> &str {
"Simulates a container-domain tool"
}
fn parameters_schema(&self) -> serde_json::Value {
serde_json::json!({"type": "object"})
}
async fn execute(
&self,
_params: serde_json::Value,
_ctx: &JobContext,
) -> Result<ToolOutput, ToolError> {
Ok(ToolOutput::text(
"should not reach here",
Duration::from_millis(1),
))
}
fn domain(&self) -> ToolDomain {
ToolDomain::Container
}
fn requires_sanitization(&self) -> bool {
false
}
}
#[tokio::test]
async fn test_container_domain_blocked() {
let tools = Arc::new(ToolRegistry::new());
tools.register(Arc::new(ContainerDomainTool)).await;
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let result = executor
.execute("container_tool", serde_json::json!({}), &ctx, None)
.await;
assert!(
matches!(result, Err(PtcError::DomainBlocked { .. })),
"Container-domain tools must be rejected: {:?}",
result
);
}
#[tokio::test]
async fn test_execute_sequential_calls() {
let tools = Arc::new(ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(SafetyLayer::new(&test_safety_config()));
let executor = ToolExecutor::new(tools, safety, Duration::from_secs(60));
let ctx = JobContext::new("test", "test");
let messages = ["alpha", "beta", "gamma"];
for msg in &messages {
let result = executor
.execute("echo", serde_json::json!({"message": msg}), &ctx, None)
.await
.expect("echo should succeed");
assert!(
result.output.contains(msg),
"Output should contain '{}'",
msg
);
}
}
}
-2
View File
@@ -18,7 +18,6 @@ pub mod redaction;
pub mod schema_validator;
pub mod wasm;
mod executor;
mod registry;
mod tool;
@@ -32,7 +31,6 @@ pub use builder::{
TestCase, TestHarness, TestResult, TestSuite, ValidationError, ValidationResult, WasmValidator,
};
pub(crate) use coercion::prepare_tool_params;
pub use executor::{PtcError, PtcToolResult, ToolExecutor};
pub use rate_limiter::RateLimiter;
pub use registry::ToolRegistry;
pub use tool::{
+5 -73
View File
@@ -19,12 +19,11 @@ use crate::tools::builder::{
use crate::tools::builtin::{
ApplyPatchTool, CancelJobTool, CreateJobTool, EchoTool, ExtensionInfoTool, HttpTool,
JobEventsTool, JobPromptTool, JobStatusTool, JsonTool, ListDirTool, ListJobsTool,
MemoryReadTool, MemorySearchTool, MemoryTreeTool, MemoryWriteTool, PromptQueue, PtcScriptTool,
ReadFileTool, ShellTool, SkillInstallTool, SkillListTool, SkillRemoveTool, SkillSearchTool,
TimeTool, ToolActivateTool, ToolAuthTool, ToolInstallTool, ToolListTool, ToolRemoveTool,
ToolSearchTool, ToolUpgradeTool, WriteFileTool,
MemoryReadTool, MemorySearchTool, MemoryTreeTool, MemoryWriteTool, PromptQueue, ReadFileTool,
ShellTool, SkillInstallTool, SkillListTool, SkillRemoveTool, SkillSearchTool, TimeTool,
ToolActivateTool, ToolAuthTool, ToolInstallTool, ToolListTool, ToolRemoveTool, ToolSearchTool,
ToolUpgradeTool, WriteFileTool,
};
use crate::tools::executor::ToolExecutor;
use crate::tools::rate_limiter::RateLimiter;
use crate::tools::tool::{ApprovalRequirement, Tool, ToolDomain};
use crate::tools::wasm::{
@@ -79,7 +78,6 @@ const PROTECTED_TOOL_NAMES: &[&str] = &[
"image_edit",
"image_analyze",
"tool_info",
"ptc_script",
];
/// Registry of available tools.
@@ -95,14 +93,6 @@ pub struct ToolRegistry {
rate_limiter: RateLimiter,
/// Reference to the message tool for setting context per-turn.
message_tool: RwLock<Option<Arc<crate::tools::builtin::MessageTool>>>,
/// Shared slot for the tool executor (enables PTC via tool_invoke).
///
/// Uses `std::sync::RwLock` (not tokio) because reads happen inside
/// `spawn_blocking` closures in WASM tool execution. The slot is
/// populated lazily after `AppBuilder::build_all()` completes, so
/// WASM tools registered during startup still get access to the
/// executor when they execute later.
tool_executor_slot: Arc<std::sync::RwLock<Option<Arc<ToolExecutor>>>>,
}
impl ToolRegistry {
@@ -124,7 +114,6 @@ impl ToolRegistry {
secrets_store: None,
rate_limiter: RateLimiter::new(),
message_tool: RwLock::new(None),
tool_executor_slot: Arc::new(std::sync::RwLock::new(None)),
}
}
@@ -149,27 +138,6 @@ impl ToolRegistry {
&self.rate_limiter
}
/// Set the tool executor for programmatic tool calling (PTC).
///
/// Writes the executor into the shared slot so all WASM tools --
/// including those registered before this call -- can resolve it
/// lazily at execution time.
pub fn set_tool_executor(&self, executor: Arc<ToolExecutor>) {
if let Ok(mut guard) = self.tool_executor_slot.write() {
*guard = Some(executor);
} else {
tracing::error!("tool_executor_slot RwLock is poisoned; PTC will be unavailable");
}
}
/// Get a clone of the shared tool executor slot.
///
/// WASM wrappers hold this slot and read from it at execution time,
/// allowing the executor to be set after tool registration.
pub fn tool_executor_slot(&self) -> Arc<std::sync::RwLock<Option<Arc<ToolExecutor>>>> {
Arc::clone(&self.tool_executor_slot)
}
/// Register a tool. Rejects dynamic tools that try to shadow a protected built-in name.
pub async fn register(&self, tool: Arc<dyn Tool>) {
let name = tool.name().to_string();
@@ -362,9 +330,8 @@ impl ToolRegistry {
self.register_sync(Arc::new(WriteFileTool::new()));
self.register_sync(Arc::new(ListDirTool::new()));
self.register_sync(Arc::new(ApplyPatchTool::new()));
self.register_sync(Arc::new(PtcScriptTool::new()));
tracing::debug!("Registered 6 development tools");
tracing::debug!("Registered 5 development tools");
}
/// Register memory tools with a workspace.
@@ -692,11 +659,6 @@ impl ToolRegistry {
wrapper = wrapper.with_oauth_refresh(oauth);
}
// Inject shared tool executor slot for PTC (lazy resolution).
// The WASM wrapper reads from this slot at execution time, so the
// executor can be set after tool registration.
wrapper = wrapper.with_tool_executor_slot(Arc::clone(&self.tool_executor_slot));
// Register the tool
self.register(Arc::new(wrapper)).await;
@@ -927,36 +889,6 @@ mod tests {
assert!(def.parameters.get("extra").is_none());
}
#[tokio::test]
async fn test_tool_executor_slot_lazy_resolution() {
let registry = ToolRegistry::new();
// Get the slot BEFORE setting the executor (simulates startup order)
let slot = registry.tool_executor_slot();
// Slot should be empty
assert!(slot.read().unwrap().is_none());
// Set the executor (simulates main.rs wiring after build_all)
let tools = Arc::new(ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(crate::safety::SafetyLayer::new(
&crate::config::SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
},
));
let executor = Arc::new(crate::tools::ToolExecutor::new(
tools,
safety,
std::time::Duration::from_secs(60),
));
registry.set_tool_executor(Arc::clone(&executor));
// Slot should now contain the executor
assert!(slot.read().unwrap().is_some());
}
#[tokio::test]
async fn test_builtin_tool_cannot_be_shadowed() {
let registry = ToolRegistry::new();
+9 -383
View File
@@ -19,7 +19,6 @@ use wasmtime_wasi::{ResourceTable, WasiCtx, WasiCtxBuilder, WasiView};
use crate::context::JobContext;
use crate::safety::LeakDetector;
use crate::secrets::SecretsStore;
use crate::tools::ToolExecutor;
use crate::tools::tool::{Tool, ToolError, ToolOutput};
use crate::tools::wasm::capabilities::Capabilities;
use crate::tools::wasm::credential_injector::{
@@ -30,26 +29,6 @@ use crate::tools::wasm::host::{HostState, LogLevel};
use crate::tools::wasm::limits::{ResourceLimits, WasmResourceLimiter};
use crate::tools::wasm::runtime::{EPOCH_TICK_INTERVAL, PreparedModule, WasmToolRuntime};
/// Synchronous tool resolver callable from within a WASM host function.
/// The closure internally creates a tokio runtime to bridge async tool execution.
/// Closure that resolves a tool call by name. The `u32` parameter is the current
/// nesting depth so the executor can enforce the global depth limit across
/// WASM->executor->WASM chains.
pub type ToolResolver =
Arc<dyn Fn(&str, serde_json::Value, u32) -> Result<String, String> + Send + Sync>;
/// RAII guard that decrements the nesting depth counter on drop, ensuring the
/// counter is restored even if the code between increment and decrement panics.
struct NestingGuard<'a> {
depth: &'a mut u32,
}
impl Drop for NestingGuard<'_> {
fn drop(&mut self) {
*self.depth = self.depth.saturating_sub(1);
}
}
// Generate component model bindings from the WIT file.
//
// This creates:
@@ -120,11 +99,6 @@ struct StoreData {
/// Dedicated tokio runtime for HTTP requests, lazily initialized.
/// Reused across multiple `http_request` calls within one execution.
http_runtime: Option<tokio::runtime::Runtime>,
/// Optional tool resolver for programmatic tool calling (PTC).
/// When set, WASM tools can invoke other tools via the `tool_invoke` host function.
tool_resolver: Option<ToolResolver>,
/// Current nesting depth for tool_invoke calls. Prevents infinite recursion.
tool_nesting_depth: u32,
}
impl StoreData {
@@ -133,7 +107,6 @@ impl StoreData {
capabilities: Capabilities,
credentials: HashMap<String, String>,
host_credentials: Vec<ResolvedHostCredential>,
tool_resolver: Option<ToolResolver>,
) -> Self {
// Minimal WASI context: no filesystem, no env vars (security)
let wasi = WasiCtxBuilder::new().build();
@@ -146,8 +119,6 @@ impl StoreData {
credentials,
host_credentials,
http_runtime: None,
tool_resolver,
tool_nesting_depth: 0,
}
}
@@ -467,39 +438,14 @@ impl near::agent::host::Host for StoreData {
result.map_err(|e| self.redact_credentials(&e))
}
fn tool_invoke(&mut self, alias: String, params_json: String) -> Result<String, String> {
use crate::tools::executor::MAX_NESTING_DEPTH;
fn tool_invoke(&mut self, alias: String, _params_json: String) -> Result<String, String> {
// Validate capability and resolve alias
let real_name = self.host_state.check_tool_invoke_allowed(&alias)?;
let _real_name = self.host_state.check_tool_invoke_allowed(&alias)?;
self.host_state.record_tool_invoke()?;
// Check nesting depth
if self.tool_nesting_depth >= MAX_NESTING_DEPTH {
return Err(format!(
"Tool invoke nesting depth exceeded (max {})",
MAX_NESTING_DEPTH
));
}
// Get the resolver
let resolver = self
.tool_resolver
.as_ref()
.ok_or("Tool invocation not available: no tool executor configured")?;
// Parse parameters
let params: serde_json::Value = serde_json::from_str(&params_json)
.map_err(|e| format!("Invalid tool parameters JSON: {}", e))?;
// Increment depth with RAII guard to ensure decrement even on panic
self.tool_nesting_depth += 1;
let current_depth = self.tool_nesting_depth;
let _guard = NestingGuard {
depth: &mut self.tool_nesting_depth,
};
// _guard drops at end of scope (or on panic), decrementing depth
resolver(&real_name, params, current_depth)
// Tool invocation requires async context and access to the tool registry,
// which aren't available inside a synchronous WASM callback.
Err("Tool invocation from WASM tools is not yet supported".to_string())
}
fn secret_exists(&mut self, name: String) -> bool {
@@ -530,11 +476,6 @@ pub struct WasmToolWrapper {
secrets_store: Option<Arc<dyn SecretsStore + Send + Sync>>,
/// OAuth refresh configuration for auto-refreshing expired tokens.
oauth_refresh: Option<OAuthRefreshConfig>,
/// Direct tool executor reference (for tests that wire it explicitly).
tool_executor: Option<Arc<ToolExecutor>>,
/// Shared slot for lazy executor resolution (production path).
/// Reads happen inside `spawn_blocking`, so this uses `std::sync::RwLock`.
tool_executor_slot: Option<Arc<std::sync::RwLock<Option<Arc<ToolExecutor>>>>>,
}
#[derive(Debug, Clone)]
@@ -623,8 +564,6 @@ impl WasmToolWrapper {
credentials: HashMap::new(),
secrets_store: None,
oauth_refresh: None,
tool_executor: None,
tool_executor_slot: None,
}
}
@@ -679,28 +618,6 @@ impl WasmToolWrapper {
self
}
/// Set the tool executor for programmatic tool calling (direct reference).
///
/// When set, the WASM `tool_invoke` host function can call other
/// registered tools synchronously via a bridged resolver closure.
/// Prefer `with_tool_executor_slot()` for production use.
pub fn with_tool_executor(mut self, executor: Arc<ToolExecutor>) -> Self {
self.tool_executor = Some(executor);
self
}
/// Set the shared tool executor slot for lazy resolution.
///
/// The executor is read from this slot at execution time, allowing
/// it to be set after tool registration (production startup order).
pub fn with_tool_executor_slot(
mut self,
slot: Arc<std::sync::RwLock<Option<Arc<ToolExecutor>>>>,
) -> Self {
self.tool_executor_slot = Some(slot);
self
}
/// Get the resource limits for this tool.
pub fn limits(&self) -> &ResourceLimits {
&self.prepared.limits
@@ -729,7 +646,6 @@ impl WasmToolWrapper {
params: serde_json::Value,
context_json: Option<String>,
host_credentials: Vec<ResolvedHostCredential>,
tool_resolver: Option<ToolResolver>,
) -> Result<(String, Vec<crate::tools::wasm::host::LogEntry>), WasmError> {
let engine = self.runtime.engine();
let limits = &self.prepared.limits;
@@ -740,7 +656,6 @@ impl WasmToolWrapper {
self.capabilities.clone(),
self.credentials.clone(),
host_credentials,
tool_resolver,
);
let mut store = Store::new(engine, store_data);
@@ -839,7 +754,6 @@ pub(super) fn extract_wasm_metadata(
Capabilities::default(),
HashMap::new(),
vec![],
None,
);
let mut store = Store::new(engine, store_data);
@@ -939,48 +853,6 @@ impl Tool for WasmToolWrapper {
// Serialize context for WASM
let context_json = serde_json::to_string(ctx).ok();
// Resolve the tool executor: direct reference takes priority, then shared slot.
let resolved_executor: Option<Arc<ToolExecutor>> =
self.tool_executor.as_ref().cloned().or_else(|| {
self.tool_executor_slot
.as_ref()
.and_then(|slot| slot.read().ok())
.and_then(|guard| guard.clone())
});
// Build a tool resolver closure if we have a tool executor.
// The resolver creates a single-threaded tokio runtime (same pattern
// as http_request) to bridge the sync WASM callback to async tool execution.
let tool_resolver: Option<ToolResolver> = resolved_executor.as_ref().map(|executor| {
let executor = Arc::clone(executor);
let user_id = ctx.user_id.clone();
Arc::new(move |name: &str, params: serde_json::Value, depth: u32| {
let executor = Arc::clone(&executor);
let name = name.to_string();
let mut ctx = JobContext::with_user(
user_id.clone(),
format!("WASM PTC: {}", name),
"Programmatic tool call from WASM tool".to_string(),
);
// Propagate depth so the executor enforces the global limit
ctx.tool_nesting_depth = depth;
let rt = tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()
.map_err(|e| format!("Failed to create runtime: {}", e))?;
rt.block_on(async {
executor
.execute(&name, params, &ctx, None)
.await
.map(|r| r.output)
.map_err(|e| e.to_string())
})
})
as Arc<dyn Fn(&str, serde_json::Value, u32) -> Result<String, String> + Send + Sync>
});
// Clone what we need for the blocking task
let runtime = Arc::clone(&self.runtime);
let prepared = Arc::clone(&self.prepared);
@@ -998,14 +870,12 @@ impl Tool for WasmToolWrapper {
description,
schemas,
credentials,
secrets_store: None, // Not needed in blocking task
oauth_refresh: None, // Already used above for pre-refresh
tool_executor: None, // Resolver closure captures the executor
tool_executor_slot: None, // Resolver closure captures the executor
secrets_store: None, // Not needed in blocking task
oauth_refresh: None, // Already used above for pre-refresh
};
tokio::task::spawn_blocking(move || {
wrapper.execute_sync(params, context_json, host_credentials, tool_resolver)
wrapper.execute_sync(params, context_json, host_credentials)
})
.await
.map_err(|e| WasmError::ExecutionPanicked(e.to_string()))?
@@ -1516,7 +1386,6 @@ fn build_tool_usage_hint(tool_name: &str, schema: &serde_json::Value) -> String
#[cfg(test)]
mod tests {
use std::collections::HashMap;
use std::sync::{Arc, Mutex};
use async_trait::async_trait;
@@ -1533,12 +1402,10 @@ mod tests {
TEST_GOOGLE_OAUTH_TOKEN, TEST_OAUTH_CLIENT_ID, TEST_OAUTH_CLIENT_SECRET,
test_secrets_store,
};
use crate::tools::tool::{Tool, ToolError, ToolOutput};
use crate::tools::tool::Tool;
use crate::tools::wasm::capabilities::Capabilities;
use crate::tools::wasm::runtime::{WasmRuntimeConfig, WasmToolRuntime};
use super::WasmToolWrapper;
struct RecordingSecretsStore {
inner: InMemorySecretsStore,
get_decrypted_lookups: Mutex<Vec<(String, String)>>,
@@ -1766,7 +1633,6 @@ mod tests {
Capabilities::default(),
HashMap::new(),
host_credentials,
None,
);
// Should inject for matching host
@@ -1806,7 +1672,6 @@ mod tests {
Capabilities::default(),
HashMap::new(),
host_credentials,
None,
);
let mut headers = HashMap::new();
@@ -1833,7 +1698,6 @@ mod tests {
Capabilities::default(),
HashMap::new(),
host_credentials,
None,
);
let text = "Error: request to https://api.example.com?key=super-secret-token failed";
@@ -2320,244 +2184,6 @@ mod tests {
assert!(hint.contains("native JSON arrays/objects")); // safety: test-only assertion
}
#[test]
fn test_coerce_params_already_correct_type() {
let schema = serde_json::json!({
"type": "object",
"properties": {
"count": { "type": "number" }
}
});
let params = serde_json::json!({"count": 5});
let result = crate::tools::coercion::prepare_params_for_schema(&params, &schema);
assert_eq!(result["count"], serde_json::json!(5));
}
#[test]
fn test_coerce_params_invalid_string_not_coerced() {
let schema = serde_json::json!({
"type": "object",
"properties": {
"count": { "type": "number" }
}
});
let params = serde_json::json!({"count": "not-a-number"});
let result = crate::tools::coercion::prepare_params_for_schema(&params, &schema);
// Should remain as string since it can't be parsed
assert_eq!(result["count"], serde_json::json!("not-a-number"));
}
// === Programmatic Tool Calling (PTC) integration tests ===
//
// These tests require the test-ptc WASM binary to be pre-built:
// cargo build --target wasm32-wasip2 --release --manifest-path tools-src/test-ptc/Cargo.toml
use crate::config::SafetyConfig;
use crate::tools::executor::ToolExecutor;
fn wasm_binary_path() -> std::path::PathBuf {
let manifest_dir = std::path::Path::new(env!("CARGO_MANIFEST_DIR"));
manifest_dir.join("tools-src/test-ptc/target/wasm32-wasip2/release/test_ptc_tool.wasm")
}
fn load_wasm_binary() -> Option<Vec<u8>> {
let path = wasm_binary_path();
if !path.exists() {
eprintln!(
"WASM test binary not found at {:?}. Build with: \
cargo build --target wasm32-wasip2 --release --manifest-path tools-src/test-ptc/Cargo.toml",
path
);
return None;
}
Some(std::fs::read(&path).expect("failed to read WASM binary"))
}
#[tokio::test]
#[ignore]
async fn test_wasm_tool_invoke_echo() {
let wasm_bytes = match load_wasm_binary() {
Some(b) => b,
None => return, // Skip if binary not built
};
// Set up runtime
let runtime = Arc::new(
WasmToolRuntime::new(WasmRuntimeConfig::default())
.expect("failed to create WASM runtime"),
);
// Set up tool registry with echo
let tools = Arc::new(crate::tools::registry::ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(crate::safety::SafetyLayer::new(&SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
}));
let executor = Arc::new(ToolExecutor::new(
tools,
safety,
std::time::Duration::from_secs(60),
));
// Prepare WASM module
let prepared = runtime
.prepare("test_ptc", &wasm_bytes, None)
.await
.expect("failed to prepare WASM module");
// Build capabilities with echo_alias -> echo
let mut aliases = HashMap::new();
aliases.insert("echo_alias".to_string(), "echo".to_string());
let capabilities = Capabilities::default().with_tool_invoke(aliases);
// Create wrapper with executor
let wrapper =
WasmToolWrapper::new(runtime, prepared, capabilities).with_tool_executor(executor);
// Execute
let ctx = crate::context::JobContext::new("test", "WASM PTC test");
let result: Result<ToolOutput, ToolError> = wrapper
.execute(serde_json::json!({"message": "hello"}), &ctx)
.await;
let result = result.expect("WASM tool execution should succeed");
let output = result.result.as_str().unwrap_or("");
assert!(
output.contains("via_wasm:"),
"Output should contain 'via_wasm:' prefix, got: {}",
output
);
assert!(
output.contains("hello"),
"Output should contain 'hello', got: {}",
output
);
}
#[tokio::test]
#[ignore]
async fn test_wasm_tool_invoke_alias_not_granted() {
let wasm_bytes = match load_wasm_binary() {
Some(b) => b,
None => return,
};
let runtime = Arc::new(
WasmToolRuntime::new(WasmRuntimeConfig::default())
.expect("failed to create WASM runtime"),
);
let tools = Arc::new(crate::tools::registry::ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(crate::safety::SafetyLayer::new(&SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
}));
let executor = Arc::new(ToolExecutor::new(
tools,
safety,
std::time::Duration::from_secs(60),
));
let prepared = runtime
.prepare("test_ptc", &wasm_bytes, None)
.await
.expect("failed to prepare WASM module");
// Only grant a DIFFERENT alias, not "echo_alias"
let mut aliases = HashMap::new();
aliases.insert("other_alias".to_string(), "echo".to_string());
let capabilities = Capabilities::default().with_tool_invoke(aliases);
let wrapper =
WasmToolWrapper::new(runtime, prepared, capabilities).with_tool_executor(executor);
let ctx = crate::context::JobContext::new("test", "WASM PTC test");
let result: Result<ToolOutput, ToolError> = wrapper
.execute(serde_json::json!({"message": "hello"}), &ctx)
.await;
// Should fail because "echo_alias" is not in the aliases
assert!(result.is_err(), "Should fail when alias not granted");
let err_msg = format!("{:?}", result.unwrap_err());
assert!(
err_msg.contains("Unknown tool alias") || err_msg.contains("echo_alias"),
"Error should mention unknown alias, got: {}",
err_msg
);
}
#[tokio::test]
#[ignore]
async fn test_wasm_tool_invoke_no_capability() {
let wasm_bytes = match load_wasm_binary() {
Some(b) => b,
None => return,
};
let runtime = Arc::new(
WasmToolRuntime::new(WasmRuntimeConfig::default())
.expect("failed to create WASM runtime"),
);
let tools = Arc::new(crate::tools::registry::ToolRegistry::new());
tools.register_builtin_tools();
let safety = Arc::new(crate::safety::SafetyLayer::new(&SafetyConfig {
max_output_length: 100_000,
injection_check_enabled: true,
}));
let executor = Arc::new(ToolExecutor::new(
tools,
safety,
std::time::Duration::from_secs(60),
));
let prepared = runtime
.prepare("test_ptc", &wasm_bytes, None)
.await
.expect("failed to prepare WASM module");
// No tool_invoke capability at all
let capabilities = Capabilities::default();
let wrapper =
WasmToolWrapper::new(runtime, prepared, capabilities).with_tool_executor(executor);
let ctx = crate::context::JobContext::new("test", "WASM PTC test");
let result: Result<ToolOutput, ToolError> = wrapper
.execute(serde_json::json!({"message": "hello"}), &ctx)
.await;
assert!(
result.is_err(),
"Should fail when no tool_invoke capability"
);
let err_msg = format!("{:?}", result.unwrap_err());
assert!(
err_msg.contains("not granted") || err_msg.contains("capability"),
"Error should mention capability not granted, got: {}",
err_msg
);
}
/// Regression: permissive fallback schema (empty properties) must NOT coerce.
/// This documents the bug where WASM tools with no sidecar `parameters` field
/// got the permissive fallback, causing coercion to be a no-op and LLM-provided
/// string integers to reach the WASM tool un-coerced.
#[test]
fn test_coerce_noop_with_permissive_schema() {
let permissive = serde_json::json!({
"type": "object",
"properties": {},
"additionalProperties": true
});
let params = serde_json::json!({"query": "test", "count": "10"});
let result = crate::tools::coercion::prepare_params_for_schema(&params, &permissive);
// With empty properties, no coercion happens — string stays string
assert_eq!(result["count"], serde_json::json!("10"));
}
/// Regression test: leak scan must run on raw headers (before credential
/// injection), not after. If it ran post-injection, the host-injected
/// Slack bot token (`xoxb-...`) would trigger a Block and reject the
-38
View File
@@ -114,36 +114,6 @@ pub struct CredentialResponse {
pub value: String,
}
/// Request to call a tool programmatically via the orchestrator.
#[derive(Debug, Serialize, Deserialize)]
pub struct ToolCallRequest {
/// Name of the tool to invoke.
pub tool_name: String,
/// JSON parameters to pass to the tool.
pub parameters: serde_json::Value,
/// Optional timeout in seconds (capped at 300s by the orchestrator).
pub timeout_secs: Option<u64>,
/// Current nesting depth for tool-invokes-tool chains.
/// Defaults to 0 for top-level calls (backward compatible).
#[serde(default)]
pub nesting_depth: u32,
}
/// Response from a programmatic tool call.
#[derive(Debug, Serialize, Deserialize)]
pub struct ToolCallResponse {
/// Whether the tool call succeeded.
pub success: bool,
/// Tool output (present on success).
pub output: Option<String>,
/// Error message (present on failure).
pub error: Option<String>,
/// Execution duration in milliseconds.
pub duration_ms: u64,
/// Whether the output was modified by the safety layer.
pub was_sanitized: bool,
}
impl WorkerHttpClient {
/// Create a new client from environment.
///
@@ -429,14 +399,6 @@ impl WorkerHttpClient {
})
}
/// Call a tool programmatically via the orchestrator (PTC).
///
/// This bypasses the LLM round-trip and invokes a tool directly on the
/// orchestrator side. Useful for scripted multi-step sequences.
pub async fn call_tool(&self, req: &ToolCallRequest) -> Result<ToolCallResponse, WorkerError> {
self.post_json("tools/call", req, "tool call").await
}
/// Signal job completion to the orchestrator.
pub async fn report_complete(&self, report: &CompletionReport) -> Result<(), WorkerError> {
let _: serde_json::Value = self
-1
View File
@@ -65,7 +65,6 @@ async fn core_registration_covers_expected_tools() {
"http",
"json",
"list_dir",
"ptc_script",
"read_file",
"shell",
"time",
-20
View File
@@ -1,20 +0,0 @@
[package]
name = "test-ptc-tool"
version = "0.1.0"
edition = "2021"
publish = false
[dependencies]
wit-bindgen = "0.41.0"
serde_json = "1.0"
[lib]
crate-type = ["cdylib"]
[profile.release]
opt-level = "s"
lto = true
strip = true
codegen-units = 1
[workspace]
-52
View File
@@ -1,52 +0,0 @@
wit_bindgen::generate!({
world: "sandboxed-tool",
path: "../../wit/tool.wit",
});
struct TestPtcTool;
impl exports::near::agent::tool::Guest for TestPtcTool {
fn execute(req: exports::near::agent::tool::Request) -> exports::near::agent::tool::Response {
match execute_inner(&req.params) {
Ok(result) => exports::near::agent::tool::Response {
output: Some(result),
error: None,
},
Err(e) => exports::near::agent::tool::Response {
output: None,
error: Some(e),
},
}
}
fn schema() -> String {
r#"{"type":"object","properties":{"message":{"type":"string"}},"required":["message"]}"#.to_string()
}
fn description() -> String {
"Test tool for PTC: calls echo via tool_invoke".to_string()
}
}
fn execute_inner(params: &str) -> Result<String, String> {
let parsed: serde_json::Value = serde_json::from_str(params)
.map_err(|e| format!("Invalid params: {}", e))?;
let message = parsed.get("message")
.and_then(|v| v.as_str())
.ok_or("Missing 'message' parameter")?;
// Build the parameters for the echo tool
let echo_params = serde_json::json!({"message": message});
// Call tool_invoke with alias "echo_alias" which should resolve to "echo"
let result = near::agent::host::tool_invoke(
"echo_alias",
&echo_params.to_string(),
)?;
// Prefix to prove it went through WASM
Ok(format!("via_wasm:{}", result))
}
export!(TestPtcTool);