feat: notify_on_complete for background processes (#5779)

* feat: notify_on_complete for background processes

When terminal(background=true, notify_on_complete=true), the system
auto-triggers a new agent turn when the process exits — no polling needed.

Changes:
- ProcessSession: add notify_on_complete field
- ProcessRegistry: add completion_queue, populate on _move_to_finished()
- Terminal tool: add notify_on_complete parameter to schema + handler
- CLI: drain completion_queue after agent turn AND during idle loop
- Gateway: enhanced _run_process_watcher injects synthetic MessageEvent
  on completion, triggering a full agent turn
- Checkpoint persistence includes notify_on_complete for crash recovery
- code_execution_tool: block notify_on_complete in sandbox scripts
- 15 new tests covering queue mechanics, checkpoint round-trip, schema

* docs: update terminal tool descriptions for notify_on_complete

- background: remove 'ONLY for servers' language, describe both patterns
  (long-lived processes AND long-running tasks with notify_on_complete)
- notify_on_complete: more prescriptive about when to use it
- TERMINAL_TOOL_DESCRIPTION: remove 'Do NOT use background for builds'
  guidance that contradicted the new feature
This commit is contained in:
Teknium 2026-04-07 02:40:16 -07:00 committed by GitHub
parent 1c425f219e
commit e120d2afac
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 398 additions and 10 deletions

View file

@ -300,7 +300,7 @@ def _call(tool_name, args):
# ---------------------------------------------------------------------------
# Terminal parameters that must not be used from ephemeral sandbox scripts
_TERMINAL_BLOCKED_PARAMS = {"background", "check_interval", "pty"}
_TERMINAL_BLOCKED_PARAMS = {"background", "check_interval", "pty", "notify_on_complete"}
def _rpc_server_loop(

View file

@ -81,6 +81,7 @@ class ProcessSession:
watcher_chat_id: str = ""
watcher_thread_id: str = ""
watcher_interval: int = 0 # 0 = no watcher configured
notify_on_complete: bool = False # Queue agent notification on exit
_lock: threading.Lock = field(default_factory=threading.Lock)
_reader_thread: Optional[threading.Thread] = field(default=None, repr=False)
_pty: Any = field(default=None, repr=False) # ptyprocess handle (when use_pty=True)
@ -112,6 +113,12 @@ class ProcessRegistry:
# Side-channel for check_interval watchers (gateway reads after agent run)
self.pending_watchers: List[Dict[str, Any]] = []
# Completion notifications — processes with notify_on_complete push here
# on exit. CLI process_loop and gateway drain this after each agent turn
# to auto-trigger a new agent turn with the process results.
import queue as _queue_mod
self.completion_queue: _queue_mod.Queue = _queue_mod.Queue()
@staticmethod
def _clean_shell_noise(text: str) -> str:
"""Strip shell startup warnings from the beginning of output."""
@ -415,6 +422,18 @@ class ProcessRegistry:
self._finished[session.id] = session
self._write_checkpoint()
# If the caller requested agent notification, enqueue the completion
# so the CLI/gateway can auto-trigger a new agent turn.
if session.notify_on_complete:
from tools.ansi_strip import strip_ansi
output_tail = strip_ansi(session.output_buffer[-2000:]) if session.output_buffer else ""
self.completion_queue.put({
"session_id": session.id,
"command": session.command,
"exit_code": session.exit_code,
"output": output_tail,
})
# ----- Query Methods -----
def get(self, session_id: str) -> Optional[ProcessSession]:
@ -721,6 +740,7 @@ class ProcessRegistry:
"watcher_chat_id": s.watcher_chat_id,
"watcher_thread_id": s.watcher_thread_id,
"watcher_interval": s.watcher_interval,
"notify_on_complete": s.notify_on_complete,
})
# Atomic write to avoid corruption on crash
@ -771,6 +791,7 @@ class ProcessRegistry:
watcher_chat_id=entry.get("watcher_chat_id", ""),
watcher_thread_id=entry.get("watcher_thread_id", ""),
watcher_interval=entry.get("watcher_interval", 0),
notify_on_complete=entry.get("notify_on_complete", False),
)
with self._lock:
self._running[session.id] = session

View file

@ -421,9 +421,11 @@ Do NOT use sed/awk to edit files — use patch instead.
Do NOT use echo/cat heredoc to create files use write_file instead.
Reserve terminal for: builds, installs, git, processes, scripts, network, package managers, and anything that needs a shell.
Foreground (default): Commands return INSTANTLY when done, even if the timeout is high. Set timeout=300 for long builds/scripts you'll still get the result in seconds if it's fast. Prefer foreground for everything that finishes.
Background: ONLY for long-running servers, watchers, or processes that never exit. Set background=true to get a session_id, then use process(action="wait") to block until done it returns instantly on completion, same as foreground. Use process(action="poll") only when you need a progress check without blocking.
Do NOT use background for scripts, builds, or installs foreground with a generous timeout is always better (fewer tool calls, instant results).
Foreground (default): Commands return INSTANTLY when done, even if the timeout is high. Set timeout=300 for long builds/scripts you'll still get the result in seconds if it's fast. Prefer foreground for short commands.
Background: Set background=true to get a session_id. Two patterns:
(1) Long-lived processes that never exit (servers, watchers).
(2) Long-running tasks with notify_on_complete=true you can keep working on other things and the system auto-notifies you when the task finishes. Great for test suites, builds, deployments, or anything that takes more than a minute.
Use process(action="poll") for progress checks, process(action="wait") to block until done.
Working directory: Use 'workdir' for per-command cwd.
PTY mode: Set pty=true for interactive CLI tools (Codex, Claude Code, Python REPL).
@ -1009,6 +1011,7 @@ def terminal_tool(
workdir: Optional[str] = None,
check_interval: Optional[int] = None,
pty: bool = False,
notify_on_complete: bool = False,
) -> str:
"""
Execute a command in the configured terminal environment.
@ -1022,6 +1025,7 @@ def terminal_tool(
workdir: Working directory for this command (optional, uses session cwd if not set)
check_interval: Seconds between auto-checks for background processes (gateway only, min 30)
pty: If True, use pseudo-terminal for interactive CLI tools (local backend only)
notify_on_complete: If True and background=True, auto-notify the agent when the process exits
Returns:
str: JSON string with output, exit_code, and error fields
@ -1254,6 +1258,32 @@ def terminal_tool(
f"configured limit of {max_timeout}s"
)
# Mark for agent notification on completion
if notify_on_complete and background:
proc_session.notify_on_complete = True
result_data["notify_on_complete"] = True
# In gateway mode, auto-register a fast watcher so the
# gateway can detect completion and trigger a new agent
# turn. CLI mode uses the completion_queue directly.
_gw_platform = os.getenv("HERMES_SESSION_PLATFORM", "")
if _gw_platform and not check_interval:
_gw_chat_id = os.getenv("HERMES_SESSION_CHAT_ID", "")
_gw_thread_id = os.getenv("HERMES_SESSION_THREAD_ID", "")
proc_session.watcher_platform = _gw_platform
proc_session.watcher_chat_id = _gw_chat_id
proc_session.watcher_thread_id = _gw_thread_id
proc_session.watcher_interval = 5
process_registry.pending_watchers.append({
"session_id": proc_session.id,
"check_interval": 5,
"session_key": session_key,
"platform": _gw_platform,
"chat_id": _gw_chat_id,
"thread_id": _gw_thread_id,
"notify_on_complete": True,
})
# Register check_interval watcher (gateway picks this up after agent run)
if check_interval and background:
effective_interval = max(30, check_interval)
@ -1550,7 +1580,7 @@ TERMINAL_SCHEMA = {
},
"background": {
"type": "boolean",
"description": "ONLY for servers/watchers that never exit. For scripts, builds, installs — use foreground with timeout instead (it returns instantly when done).",
"description": "Run the command in the background. Two patterns: (1) Long-lived processes that never exit (servers, watchers). (2) Long-running tasks paired with notify_on_complete=true — you can keep working and get notified when the task finishes. For short commands, prefer foreground with a generous timeout instead.",
"default": False
},
"timeout": {
@ -1571,6 +1601,11 @@ TERMINAL_SCHEMA = {
"type": "boolean",
"description": "Run in pseudo-terminal (PTY) mode for interactive CLI tools like Codex, Claude Code, or Python REPL. Only works with local and SSH backends. Default: false.",
"default": False
},
"notify_on_complete": {
"type": "boolean",
"description": "When true (and background=true), you'll be automatically notified when the process finishes — no polling needed. Use this for tasks that take a while (tests, builds, deployments) so you can keep working on other things in the meantime.",
"default": False
}
},
"required": ["command"]
@ -1587,6 +1622,7 @@ def _handle_terminal(args, **kw):
workdir=args.get("workdir"),
check_interval=args.get("check_interval"),
pty=args.get("pty", False),
notify_on_complete=args.get("notify_on_complete", False),
)