hyperhive/hive-ag3nt/src/bin/hive-ag3nt.rs

464 lines
18 KiB
Rust

use std::path::{Path, PathBuf};
use std::sync::{Arc, Mutex};
use std::time::Duration;
use hive_ag3nt::web_ui::TurnLock;
use anyhow::Result;
use clap::{Parser, Subcommand};
use hive_ag3nt::events::{Bus, LiveEvent, TurnState};
use hive_ag3nt::login::{self, LoginState};
use hive_ag3nt::turn_stats::TurnStats;
use hive_ag3nt::{DEFAULT_SOCKET, DEFAULT_WEB_PORT, client, mcp, plugins, serve_common, turn, web_ui};
use hive_sh4re::{AgentRequest, AgentResponse};
#[derive(Parser)]
#[command(name = "hive-ag3nt", about = "hyperhive sub-agent harness")]
struct Cli {
/// Path to the per-agent MCP socket (bind-mounted from the host).
#[arg(long, global = true, default_value = DEFAULT_SOCKET)]
socket: PathBuf,
#[command(subcommand)]
cmd: Cmd,
}
#[derive(Subcommand)]
enum Cmd {
/// Run the long-lived harness loop. Polls inbox; replies via `claude --print`
/// when available, falling back to a simple echo otherwise.
Serve {
/// Inbox poll interval in milliseconds.
#[arg(long, default_value_t = 1000)]
poll_ms: u64,
},
/// Run the agent's MCP server on stdio. Spawned by `claude` via
/// `--mcp-config`; tools dispatch through `/run/hive/mcp.sock` back into
/// the hyperhive broker.
Mcp,
/// Inject a wake-up event into this agent's inbox so the next turn
/// fires with the given body. Intended for extra MCP servers /
/// helpers running inside the container (matrix bridge, scraper,
/// webhook listener) that need to nudge claude on external events.
/// `from` is the sender label that appears in the wake prompt
/// (claude sees "from: matrix" etc.).
Wake {
#[arg(long)]
from: String,
/// Body of the wake message. Pass `-` to read from stdin.
#[arg(long)]
body: String,
},
}
#[tokio::main]
async fn main() -> Result<()> {
tracing_subscriber::fmt()
.with_env_filter(
tracing_subscriber::EnvFilter::try_from_default_env()
.unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("info")),
)
.init();
let cli = Cli::parse();
match cli.cmd {
Cmd::Serve { poll_ms } => {
let port = std::env::var("HIVE_PORT")
.ok()
.and_then(|s| s.parse::<u16>().ok())
.unwrap_or(DEFAULT_WEB_PORT);
let label = std::env::var("HIVE_LABEL").unwrap_or_else(|_| "hive-ag3nt".into());
let claude_dir = login::default_dir();
let initial = LoginState::from_dir(&claude_dir);
tracing::info!(state = ?initial, claude_dir = %claude_dir.display(), "harness boot");
let login_state = Arc::new(Mutex::new(initial));
let bus = Bus::new();
let stats = TurnStats::open_default();
if let Some(s) = &stats {
let (ctx, cost) = s.last_usage();
if ctx.is_some() || cost.is_some() {
bus.seed_usage(ctx, cost);
}
}
let files = turn::TurnFiles::prepare(&cli.socket, &label, mcp::Flavor::Agent).await?;
let turn_lock: TurnLock = Arc::new(tokio::sync::Mutex::new(()));
plugins::install_configured(&cli.socket, Some("manager")).await;
tokio::spawn(hive_ag3nt::forge_notify::run(cli.socket.clone(), false));
tokio::spawn(web_ui::serve(
label.clone(),
port,
login_state.clone(),
bus.clone(),
cli.socket.clone(),
files.clone(),
turn_lock.clone(),
));
match initial {
LoginState::Online => {
serve(
&cli.socket,
Duration::from_millis(poll_ms),
login_state,
claude_dir,
bus,
stats,
&files,
turn_lock,
&label,
)
.await
}
LoginState::NeedsLogin => {
// Partial-run mode: keep the harness alive (so the web UI
// stays bound) but don't drive the turn loop. Poll the
// claude dir; once a session lands we enter `serve`.
turn::wait_for_login(&claude_dir, login_state.clone(), &bus, poll_ms).await;
serve(
&cli.socket,
Duration::from_millis(poll_ms),
login_state,
claude_dir,
bus,
stats,
&files,
turn_lock,
&label,
)
.await
}
}
}
Cmd::Mcp => mcp::serve_agent_stdio(cli.socket).await,
Cmd::Wake { from, body } => {
// Read body from stdin if caller passed `-`. Same convention
// many CLI tools use; keeps multi-line / shell-quoting
// friction out of the body content.
let body = if body == "-" {
let mut buf = String::new();
std::io::Read::read_to_string(&mut std::io::stdin(), &mut buf)?;
buf
} else {
body
};
let resp: AgentResponse =
client::request(&cli.socket, &AgentRequest::Wake { from, body }).await?;
match resp {
AgentResponse::Ok => Ok(()),
AgentResponse::Err { message } => anyhow::bail!("wake: {message}"),
other => anyhow::bail!("wake: unexpected response {other:?}"),
}
}
}
}
#[allow(clippy::too_many_arguments)]
async fn serve(
socket: &Path,
interval: Duration,
login_state: Arc<Mutex<LoginState>>,
claude_dir: std::path::PathBuf,
bus: Bus,
stats: Option<TurnStats>,
files: &turn::TurnFiles,
turn_lock: TurnLock,
label: &str,
) -> Result<()> {
tracing::info!(socket = %socket.display(), "hive-ag3nt serve");
requeue_inflight(socket).await;
loop {
let recv: Result<AgentResponse> =
// Explicit long-poll: park until a message arrives (180s cap).
// `max: None` (= 1) — one turn per wake; claude calls
// recv(max: N) in-turn to drain bursts.
client::request(
socket,
&AgentRequest::Recv {
wait_seconds: Some(180),
max: None,
},
)
.await;
match recv {
Ok(AgentResponse::Messages { messages }) if !messages.is_empty() => {
let first = messages.into_iter().next().expect("checked non-empty");
let auth_failed =
handle_agent_turn(socket, &bus, stats.as_ref(), files, &turn_lock, label, first)
.await;
if auth_failed {
// Park: flip LoginState + wait for the operator's
// re-auth to repopulate claude_dir. wait_for_login
// emits `online` on resume, which clears the
// needs_login sentinel.
*login_state.lock().unwrap() = LoginState::NeedsLogin;
turn::wait_for_login(
&claude_dir,
login_state.clone(),
&bus,
u64::try_from(interval.as_millis()).unwrap_or(2000),
)
.await;
}
}
Ok(AgentResponse::Messages { .. }) => {
// Idle: empty list = nothing pending. Brief sleep
// before next poll so a stretch of empty long-poll
// returns doesn't tight-loop.
tokio::time::sleep(interval).await;
}
Ok(
AgentResponse::Ok
| AgentResponse::Status { .. }
| AgentResponse::Recent { .. }
| AgentResponse::QuestionQueued { .. }
| AgentResponse::LooseEnds { .. }
| AgentResponse::PendingRemindersCount { .. }
| AgentResponse::ReminderRollup { .. }
| AgentResponse::AgentMeta { .. },
) => {
tracing::warn!("recv produced unexpected response kind");
}
Ok(AgentResponse::Err { message }) => {
tracing::warn!(%message, "recv error");
}
Err(e) => {
tracing::warn!(error = ?e, "recv failed; retrying");
}
}
}
}
/// Drive one turn for a received agent-inbox message. Returns `true`
/// when the turn ended with `AuthFailed` so the caller knows to park
/// in `wait_for_login`.
async fn handle_agent_turn(
socket: &Path,
bus: &Bus,
stats: Option<&TurnStats>,
files: &turn::TurnFiles,
turn_lock: &TurnLock,
label: &str,
first: hive_sh4re::DeliveredMessage,
) -> bool {
let from = first.from;
let body = first.body;
let redelivered = first.redelivered;
tracing::info!(%from, %body, %redelivered, "inbox");
let unread = inbox_unread(socket).await;
bus.emit(LiveEvent::TurnStart { from: from.clone(), body: body.clone(), unread });
bus.set_state(TurnState::Thinking);
let started_at = serve_common::now_unix();
let started_instant = std::time::Instant::now();
let model_at_start = bus.model();
let prompt = serve_common::format_wake_prompt(&from, &body, unread, redelivered);
let outcome = {
let _guard = turn_lock.lock().await;
turn::drive_turn(&prompt, files, bus).await
};
turn::emit_turn_end(bus, &outcome);
bus.set_state(TurnState::Idle);
// Ack only on a clean turn-end. `Failed` leaves every message popped
// during the turn in the unacked list; next harness boot requeues them.
if matches!(outcome, turn::TurnOutcome::Ok | turn::TurnOutcome::Compacted) {
ack_turn(socket).await;
}
if matches!(outcome, turn::TurnOutcome::RateLimited) {
let secs = turn::rate_limit_sleep_secs();
bus.emit_status("rate_limited");
bus.emit(LiveEvent::Note {
text: format!("API rate-limited — sleeping {secs}s before retry"),
});
tracing::warn!(sleep_secs = secs, "rate-limited; parking");
tokio::time::sleep(Duration::from_secs(secs)).await;
requeue_inflight(socket).await;
bus.emit_status("online");
}
// 401: flip into needs_login + requeue the message that triggered
// the turn so it survives the re-auth. The serve loop's outer
// login-state watcher parks until the operator's `/login` flow
// completes; once it does, the requeued message replays the turn
// (closes #419).
if matches!(outcome, turn::TurnOutcome::AuthFailed) {
bus.emit_status("needs_login_idle");
bus.emit(LiveEvent::Note {
text: "API 401 — waiting for re-login via web UI".into(),
});
tracing::warn!("auth-failed; parking until re-login");
requeue_inflight(socket).await;
}
// Real crash: PromptTooLong is absorbed by compaction inside drive_turn.
if let turn::TurnOutcome::Failed(e) = &outcome {
notify_manager_of_failure(socket, label, e).await;
}
if let Some(stats) = stats {
let ended_at = serve_common::now_unix();
let duration_ms =
i64::try_from(started_instant.elapsed().as_millis()).unwrap_or(i64::MAX);
let (open_threads, open_reminders) = fetch_agent_post_turn_counts(socket).await;
let row = serve_common::build_row(
started_at,
ended_at,
duration_ms,
model_at_start,
from.clone(),
&outcome,
bus,
open_threads,
open_reminders,
);
stats.record(&row);
}
let pending = inbox_unread(socket).await;
if pending > 0 {
tracing::info!(%pending, "pending messages after turn; fetching next");
}
// `request_next_turn` MCP tool: agent wrote a sentinel requesting
// an immediate self-continuation. Clear and inject synthetic wake.
check_and_inject_continue(socket, label).await;
matches!(outcome, turn::TurnOutcome::AuthFailed)
}
// Per-turn user prompt: the role/tools/etc. is in the system prompt
// (`prompts/system.md` filtered to this agent's role-block via
// `hive_ag3nt::prompt::render` → `claude --system-prompt-file`); this
// is just the wake signal claude reacts to. `unread` is the count of *other*
// messages in the inbox right after this one was popped.
// `redelivered` flags messages that were popped in a prior harness
// session, never acked, and resurfaced after a restart — a banner
// at the top of the wake prompt warns that any side-effects of
// previous handling may already have happened.
/// Best-effort: tell the broker every message we popped during the
/// turn is now fully handled (turn-end-OK). Swallows transport
/// errors — the worst case is a redundant requeue on next boot.
async fn ack_turn(socket: &Path) {
match client::request::<_, AgentResponse>(socket, &AgentRequest::AckTurn).await {
Ok(AgentResponse::Ok) => {}
Ok(AgentResponse::Err { message }) => {
tracing::warn!(%message, "ack_turn rejected by broker");
}
Ok(other) => {
tracing::warn!(?other, "ack_turn unexpected response");
}
Err(e) => tracing::warn!(error = ?e, "ack_turn transport error"),
}
}
/// Boot-time recovery: ask the broker to resurface anything we
/// popped in a previous harness session but never acked. The broker
/// resets `delivered_at = NULL` on those rows and remembers their
/// ids so the next `Recv` carries `redelivered: true`. Swallows
/// transport errors — they degrade to "no recovery this boot",
/// which is no worse than the pre-feature behaviour (silent drop).
async fn requeue_inflight(socket: &Path) {
match client::request::<_, AgentResponse>(socket, &AgentRequest::RequeueInflight).await {
Ok(AgentResponse::Ok) => {}
Ok(AgentResponse::Err { message }) => {
tracing::warn!(%message, "requeue_inflight rejected by broker");
}
Ok(other) => {
tracing::warn!(?other, "requeue_inflight unexpected response");
}
Err(e) => tracing::warn!(error = ?e, "requeue_inflight transport error"),
}
}
/// Best-effort: tell the manager that this agent's last turn crashed
/// (claude exited non-zero, compaction didn't help, etc.). Routed
/// through the normal send path so the manager's inbox surfaces it
/// as a system-style event; `label` is included explicitly in the
/// body so the manager can identify the failing agent without having
/// to look at the `from` field (which is broker-stamped and may
/// differ from what the operator sees in the dashboard). Swallows
/// transport errors — we just logged the failure, the worst case is
/// the manager learns about the crash from the dashboard instead of
/// inbox.
async fn notify_manager_of_failure(socket: &Path, label: &str, err: &anyhow::Error) {
let body = format!("[system] agent `{label}` claude turn failed:\n{err:#}");
let res = client::request::<_, AgentResponse>(
socket,
&AgentRequest::Send {
to: "manager".into(),
body,
in_reply_to: None,
},
)
.await;
if let Err(e) = res {
tracing::warn!(error = ?e, "failed to notify manager of turn failure");
}
}
/// Best-effort: ask our own per-agent socket how many messages are still
/// pending after the wake-up Recv. Returns 0 if anything goes wrong.
async fn inbox_unread(socket: &Path) -> u64 {
match client::request::<_, AgentResponse>(socket, &AgentRequest::Status).await {
Ok(AgentResponse::Status { unread }) => unread,
_ => 0,
}
}
/// Best-effort: ask hive-c0re for this agent's open thread count + pending
/// reminder count, after the turn finishes. Either roundtrip can fail
/// (transport hiccup, race with hive-c0re restart) — in those cases we
/// just drop a `None` into the stats row rather than blocking the loop.
async fn fetch_agent_post_turn_counts(socket: &Path) -> (Option<u64>, Option<u64>) {
let threads = match client::request::<_, AgentResponse>(
socket,
&AgentRequest::GetLooseEnds,
)
.await
{
Ok(AgentResponse::LooseEnds { loose_ends }) => u64::try_from(loose_ends.len()).ok(),
_ => None,
};
let reminders = match client::request::<_, AgentResponse>(
socket,
&AgentRequest::CountPendingReminders,
)
.await
{
Ok(AgentResponse::PendingRemindersCount { count }) => Some(count),
_ => None,
};
(threads, reminders)
}
/// Check for the `request_next_turn` sentinel file. If present, remove it
/// and inject a synthetic `from: "self", body: "continue"` message so the
/// serve loop fires an immediate follow-up turn even when the inbox is empty.
/// Best-effort: any I/O error is logged and ignored (the agent just waits
/// for a real message as normal).
async fn check_and_inject_continue(socket: &Path, label: &str) {
let sentinel = hive_ag3nt::paths::state_dir().join("hyperhive-continue");
if !sentinel.exists() {
return;
}
if let Err(e) = std::fs::remove_file(&sentinel) {
tracing::warn!(error = %e, "check_and_inject_continue: remove sentinel failed");
return;
}
// Sentinel was present: inject a wake so the outer loop fires immediately.
// Route through the `Wake` request which is already wired in agent_server.
let res = client::request::<_, AgentResponse>(
socket,
&AgentRequest::Wake {
from: "self".into(),
body: "continue".into(),
},
)
.await;
match res {
Ok(AgentResponse::Ok) => {
tracing::info!(%label, "request_next_turn: injected self-continue wake");
}
Ok(AgentResponse::Err { message }) => {
tracing::warn!(%message, "check_and_inject_continue: wake rejected");
}
Err(e) => {
tracing::warn!(error = ?e, "check_and_inject_continue: wake transport error");
}
_ => {}
}
}