Compare commits

..
Author SHA1 Message Date
Miguel Palhas f689c916bf fix(maestro): poll awaiting-input until leaderless state holds
The pipeline leader exits as soon as both stages are forked, which can
beat the surviving stage opening the tty and parking on read — the one-
shot assertion then ran against a group with no parked member yet.
Poll for the answer, accepting it only while the foreground group is
still leaderless so a shell that regained the tty cannot satisfy it.

Fixes #113
2026-08-24 22:50:02 +01:00
Miguel Palhas 88233cb96d fix(maestro): sync watcher re-emit test to real event flow
Two races. First, the watcher updates the snapshot and emits
SessionUpdated as two steps, so the drain could run between them: the
legitimate breakdown emission then arrived inside the quiet window.
Subscribe before the watcher starts and consume through the breakdown
event so the window begins after the last real change.

Second, the assertion treated any SessionUpdated as a failed
suppression, but the activity pass may legitimately announce an
Active→Idle transition in that window (ACTIVE_WINDOW is 3s), carrying
today's blocked_work unchanged. Only a changed indicator counts as a
re-emit now.

Fixes #113
2026-08-24 22:50:02 +01:00
Miguel Palhas efbccb4e11 fix(maestro): wait for settle in pristine natural-exit test
Dead status is observable a moment before the settle step removes the
worktree, so the assertions raced the removal under parallel load. Wait
for the tree to be gone and the store to stay empty, same as the
retained-store wait cfebfd4 added to the guarded sibling.

Fixes #113
2026-08-24 22:41:54 +01:00
5 changed files with 112 additions and 164 deletions
+19 -4
View File
@@ -189,8 +189,23 @@ fn a_foreground_pipeline_whose_leader_exited_is_still_awaiting_input() {
"never observed a foreground group whose leader had exited"
);
assert!(
awaiting_input(pane.pane_pid, &pane.tty, &ProcGroups::snapshot()),
"a surviving pipeline member still holds the tty"
);
// The leader can be gone while the surviving stage is still setting up:
// `true` exits as soon as both stages are forked, before the reader has
// opened the tty and parked. Poll for the answer, but only accept it
// while the state this test exists for still holds — otherwise a shell
// that regained the terminal would satisfy it vacuously.
let deadline = std::time::Instant::now() + Duration::from_secs(5);
let mut holds = false;
while std::time::Instant::now() < deadline {
let groups = ProcGroups::snapshot();
let still_leaderless = pgrp_and_tpgid(pane.pane_pid).is_some_and(|(pgrp, tpgid)| {
tpgid > 0 && tpgid != pgrp && pgrp_and_tpgid(tpgid).is_none()
});
if still_leaderless && awaiting_input(pane.pane_pid, &pane.tty, &groups) {
holds = true;
break;
}
std::thread::sleep(Duration::from_millis(20));
}
assert!(holds, "a surviving pipeline member still holds the tty");
}
+47 -6
View File
@@ -397,6 +397,12 @@ async fn watcher_refreshes_the_blocked_work_indicator() {
init_bare_repo(origin.path());
push_all_to_origin(repo.path(), origin.path());
// Subscribe before the watcher starts so every pass's emission is
// captured — otherwise the breakdown event can land between the
// snapshot observation below and the drain, and the quiet window
// would see it as an illegitimate re-emit.
let mut events = fx.manager.events();
fx.manager.start_liveness_watcher_with_blocked_work(
Duration::from_millis(50),
Duration::from_millis(100),
@@ -418,14 +424,38 @@ async fn watcher_refreshes_the_blocked_work_indicator() {
})
.await;
// A steady state must not re-emit every tick (#83 change-detection).
let mut events = fx.manager.events();
// Consume through the emission that carried the breakdown (#83
// change-detection): the quiet window below must start after the last
// legitimate change, not race it into the channel.
loop {
match tokio::time::timeout(Duration::from_secs(5), events.recv()).await {
Ok(Ok(CatalogEvent::SessionUpdated { session }))
if matches!(
&session.blocked_work,
Some(b) if b.uncommitted_files >= 1 && b.untracked_files >= 1
) =>
{
break;
}
Ok(Ok(_)) => continue,
_ => panic!("never saw the SessionUpdated carrying the breakdown"),
}
}
while events.try_recv().is_ok() {}
tokio::time::sleep(Duration::from_millis(400)).await;
assert!(
!matches!(events.try_recv(), Ok(CatalogEvent::SessionUpdated { .. })),
"an unchanged worktree must not re-emit SessionUpdated"
);
// Only a *changed* indicator counts as a re-emit (#83 change-detection).
// The activity pass may still legitimately announce e.g. Active→Idle
// (its debounce is time-based, not worktree-based), and its payload
// carries today's blocked_work unchanged.
while let Ok(CatalogEvent::SessionUpdated { session }) = events.try_recv() {
assert!(
matches!(
&session.blocked_work,
Some(b) if b.uncommitted_files >= 1 && b.untracked_files >= 1
),
"an unchanged worktree must not re-emit a changed indicator"
);
}
}
/// Cleanup on kill is unconditional once the guard passes (#79): even a spec
@@ -848,6 +878,17 @@ async fn natural_exit_of_a_pristine_worktree_removes_it() {
})
.await;
// (`Dead` is observable a moment before the settle step removes the tree.)
wait_for_async(Duration::from_secs(3), || async {
!wt_path.exists()
&& fx
.retained_store()
.load_all()
.map(|all| all.is_empty())
.unwrap_or(false)
})
.await;
assert!(
!wt_path.exists(),
"a pristine worktree leaves nothing behind"
+6 -28
View File
@@ -25,25 +25,13 @@ pub const HOOK_MARKER: &str = "maestro-claude-hook";
pub const SCRIPT_FILE_NAME: &str = "maestro-claude-hook.sh";
/// Claude Code hook event → `ActivityState` spelling posted to the daemon.
///
/// `PreToolUse` is what makes a session `Active` *during* a long tool call —
/// `PostToolUse` alone only fires once it ends (#114). `StopFailure` is
/// Claude Code's own name for "the turn ended on an API error"; there is no
/// failure signal in `Stop` or `SessionEnd` payloads, so those never report
/// `Error`.
pub const HOOK_EVENTS: [(&str, &str); 5] = [
// Tool starting: the agent is demonstrably producing work, even if the
// tool itself runs for minutes.
("PreToolUse", "Active"),
// Tool finished successfully.
pub const HOOK_EVENTS: [(&str, &str); 3] = [
// Tool finished: the agent is demonstrably producing work.
("PostToolUse", "Active"),
// Permission / idle notices: Claude is blocked on the user.
("Notification", "AwaitingInput"),
// Turn ended: waiting on the user again.
("Stop", "AwaitingInput"),
// Turn ended on an API error; the payload's `error` field is forwarded
// as the report's `detail` (#114).
("StopFailure", "Error"),
];
/// The shipped hook script. POSIX sh so any `/bin/sh` works.
@@ -51,27 +39,17 @@ pub const HOOK_SCRIPT: &str = r#"#!/bin/sh
# Installed by maestro (#106): report Claude Code activity to its daemon.
# Exits silently unless running inside a session maestro spawned (#103 env).
# Event -> ActivityState mapping:
# PreToolUse -> Active (tool started; agent is working)
# PostToolUse -> Active (tool finished)
# PostToolUse -> Active (tool finished; agent is working)
# Notification -> AwaitingInput (permission / idle notice)
# Stop -> AwaitingInput (turn ended; waiting on the user)
# StopFailure -> Error (turn ended on an API error)
state="$1"
[ -n "$MAESTRO_SESSION_ID" ] || exit 0
[ -n "$MAESTRO_URL" ] || exit 0
payload=$(cat 2>/dev/null) || true
body="{\"state\":\"$state\"}"
# StopFailure payloads carry the outcome in "error". Only a plain enum
# spelling is reported as the detail — anything else (quotes, backslashes,
# prose) would risk breaking the JSON body, and the spelling already names
# the failure.
detail=$(printf '%s' "$payload" | sed -n 's/.*"error"[[:space:]]*:[[:space:]]*"\([a-z_]*\)".*/\1/p')
if [ -n "$detail" ]; then
body="{\"state\":\"$state\",\"detail\":\"$detail\"}"
fi
# Drain the hook payload so the parent never sees a broken pipe.
cat >/dev/null 2>&1
curl -fsS --connect-timeout 1 --max-time 2 \
-H 'Content-Type: application/json' \
-d "$body" \
-d "{\"state\":\"$state\"}" \
"$MAESTRO_URL/api/sessions/$MAESTRO_SESSION_ID/activity" \
>/dev/null 2>&1 || true
exit 0
+40 -112
View File
@@ -7,7 +7,7 @@ use std::process::{Command, Stdio};
use super::*;
fn sh(script: &str, args: &[&str], envs: &[(&str, &str)], payload: &[u8]) -> std::process::Output {
fn sh(script: &str, args: &[&str], envs: &[(&str, &str)]) -> std::process::Output {
let mut cmd = Command::new("/bin/sh");
// `sh -c CMD NAME ARGS…`: NAME fills $0, so the real arguments start at $1.
let mut argv = vec!["-c", script, HOOK_MARKER];
@@ -26,48 +26,21 @@ fn sh(script: &str, args: &[&str], envs: &[(&str, &str)], payload: &[u8]) -> std
.spawn()
.expect("spawn /bin/sh");
// Feed a Claude-Code-style payload on stdin; the script must drain it.
child.stdin.take().unwrap().write_all(payload).ok();
child
.stdin
.take()
.unwrap()
.write_all(br#"{"session_id":"x"}"#)
.ok();
child.wait_with_output().expect("wait")
}
const SESSION: &str = "01234567-89ab-cdef-0123-456789abcdef";
/// Accept exactly one HTTP request on a throwaway listener, answer 204, and
/// hand back the raw request text. Runs `run` with the base URL once the
/// acceptor thread is in place.
fn capture_request(run: impl FnOnce(String) -> std::process::Output) -> String {
let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
let addr = listener.local_addr().unwrap();
let handle = std::thread::spawn(move || {
let (mut sock, _) = listener.accept().unwrap();
let mut buf = [0u8; 4096];
let n = sock.read(&mut buf).unwrap();
sock.write_all(b"HTTP/1.1 204 No Content\r\nContent-Length: 0\r\n\r\n")
.unwrap();
String::from_utf8_lossy(&buf[..n]).to_string()
});
let out = run(format!("http://{addr}"));
assert!(out.status.success(), "{:?}", out.stderr);
handle.join().unwrap()
}
fn session_envs(url: &str) -> Vec<(&'static str, String)> {
vec![
("MAESTRO_URL", url.to_string()),
("MAESTRO_SESSION_ID", SESSION.to_string()),
]
}
fn owned_envs<'a>(envs: &'a [(&'static str, String)]) -> Vec<(&'static str, &'a str)> {
envs.iter().map(|(k, v)| (*k, v.as_str())).collect()
}
/// The hook must be invisible outside maestro: no env vars (#103 unset) ⇒
/// exit 0, no output, stdin drained.
#[test]
fn script_exits_silently_without_maestro_env() {
for (event, _) in HOOK_EVENTS {
let out = sh(HOOK_SCRIPT, &[event], &[], br#"{"session_id":"x"}"#);
let out = sh(HOOK_SCRIPT, &[event], &[]);
assert!(out.status.success(), "{event:?}: {:?}", out.stderr);
assert!(out.stdout.is_empty(), "{event:?}: {:?}", out.stdout);
assert!(out.stderr.is_empty(), "{event:?}: {:?}", out.stderr);
@@ -78,82 +51,46 @@ fn script_exits_silently_without_maestro_env() {
/// activity endpoint and still exits 0.
#[test]
fn script_posts_mapped_state_to_the_activity_endpoint() {
let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
let addr = listener.local_addr().unwrap();
let server = move || {
let (mut sock, _) = listener.accept().unwrap();
let mut buf = [0u8; 4096];
let n = sock.read(&mut buf).unwrap();
let raw = String::from_utf8_lossy(&buf[..n]).to_string();
sock.write_all(b"HTTP/1.1 204 No Content\r\nContent-Length: 0\r\n\r\n")
.unwrap();
raw
};
let handle = std::thread::spawn(server);
let url = format!("http://{addr}");
let session = "01234567-89ab-cdef-0123-456789abcdef";
// The installer bakes the mapped state into the command's argv; the
// script relays whatever it is given.
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["Active"],
&owned_envs(&session_envs(&url)),
br#"{"session_id":"x"}"#,
)
});
let out = sh(
HOOK_SCRIPT,
&["Active"],
&[("MAESTRO_URL", &url), ("MAESTRO_SESSION_ID", session)],
);
assert!(out.status.success());
let request = handle.join().unwrap();
assert!(request.starts_with("POST "), "{request}");
assert!(
request.contains(&format!("/api/sessions/{SESSION}/activity")),
request.contains(&format!("/api/sessions/{session}/activity")),
"{request}"
);
assert!(request.contains(r#""state":"Active""#), "{request}");
}
/// Every event maps to the state the contract specifies.
/// Every event maps to the state the issue specifies.
#[test]
fn event_state_mapping_matches_the_contract() {
let map: std::collections::HashMap<&str, &str> = HOOK_EVENTS.into_iter().collect();
assert_eq!(map["PreToolUse"], "Active");
assert_eq!(map["PostToolUse"], "Active");
assert_eq!(map["Notification"], "AwaitingInput");
assert_eq!(map["Stop"], "AwaitingInput");
assert_eq!(map["StopFailure"], "Error");
}
/// A `StopFailure` payload names the API error that ended the turn; the
/// script reports its `error` spelling as the report's `detail` so a card
/// can show why (#114).
#[test]
fn script_reports_the_error_spelling_as_detail() {
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["Error"],
&owned_envs(&session_envs(&url)),
br#"{"session_id":"abc","hook_event_name":"StopFailure","error":"rate_limit","error_details":"429 Too Many Requests"}"#,
)
});
assert!(request.contains(r#""state":"Error""#), "{request}");
assert!(request.contains(r#""detail":"rate_limit""#), "{request}");
}
/// An `error` value that is not a plain enum spelling (or a payload without
/// one at all) must never produce a broken body — no detail is posted.
#[test]
fn script_posts_no_detail_for_an_unexpected_error_value() {
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["Error"],
&owned_envs(&session_envs(&url)),
br#"{"hook_event_name":"StopFailure","error":"429 \"Too Many\" \\ requests"}"#,
)
});
assert!(request.contains(r#""state":"Error""#), "{request}");
assert!(!request.contains("detail"), "{request}");
}
/// Events whose payload carries no error fields post a bare body — never an
/// empty or stale detail.
#[test]
fn script_posts_no_detail_for_events_without_an_error() {
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["AwaitingInput"],
&owned_envs(&session_envs(&url)),
br#"{"hook_event_name":"Stop","stop_hook_active":false}"#,
)
});
assert!(request.contains(r#""state":"AwaitingInput""#), "{request}");
assert!(!request.contains("detail"), "{request}");
assert_eq!(map["Notification"], "AwaitingInput");
}
/// A dead or unreachable daemon must never fail — or slow down — a session.
@@ -167,7 +104,6 @@ fn script_ignores_daemon_failure() {
("MAESTRO_URL", "http://127.0.0.1:1"),
("MAESTRO_SESSION_ID", "00000000-0000-0000-0000-000000000000"),
],
br#"{"session_id":"x"}"#,
);
assert!(out.status.success());
}
@@ -202,10 +138,9 @@ fn fresh_install_registers_every_event() {
install(&settings_path(home), &script_path(home)).unwrap();
let cmds = installed_commands(&settings_path(home));
assert_eq!(cmds.len(), HOOK_EVENTS.len());
assert_eq!(cmds.len(), 3);
let joined = cmds.join("\n");
assert_eq!(joined.matches("Active").count(), 2);
assert!(joined.contains("Error"));
assert!(joined.contains("Active"));
assert_eq!(joined.matches("AwaitingInput").count(), 2);
for cmd in &cmds {
assert!(cmd.contains(SCRIPT_FILE_NAME));
@@ -223,10 +158,7 @@ fn install_is_idempotent() {
install(&settings_path(home), &script_path(home)).unwrap();
let twice = std::fs::read_to_string(settings_path(home)).unwrap();
assert_eq!(once, twice);
assert_eq!(
installed_commands(&settings_path(home)).len(),
HOOK_EVENTS.len()
);
assert_eq!(installed_commands(&settings_path(home)).len(), 3);
}
/// An older maestro entry under the same events is updated in place, not
@@ -256,11 +188,7 @@ fn install_updates_an_older_maestro_hook_in_place() {
install(&settings, &new_script).unwrap();
let cmds = installed_commands(&settings);
assert_eq!(
cmds.len(),
HOOK_EVENTS.len(),
"old entry replaced, not kept: {cmds:?}"
);
assert_eq!(cmds.len(), 3, "old entry replaced, not kept: {cmds:?}");
assert!(!cmds.iter().any(|c| c == old_script));
let root: serde_json::Value =
serde_json::from_str(&std::fs::read_to_string(&settings).unwrap()).unwrap();
@@ -303,11 +231,11 @@ fn install_preserves_foreign_hooks_and_removes_stale_maestro_ones() {
assert_eq!(pre[0]["hooks"][0]["command"], "my-own-linter.sh");
// Stale maestro event gone.
assert!(root["hooks"].get("UserPromptSubmit").is_none());
// Our events present, nothing else maestro-related anywhere.
// Our three events present, nothing else maestro-related anywhere.
let cmds = installed_commands(&settings);
assert_eq!(
cmds.iter().filter(|c| c.contains(HOOK_MARKER)).count(),
HOOK_EVENTS.len(),
3,
"{cmds:?}"
);
}
-14
View File
@@ -107,20 +107,6 @@ When the TTL expires, inference resumes and the session's reported detail —
if any — clears with the report. Clients never blend the two: whichever source
is in force supplies both the colour and the clock for `activity_since`.
Two rulings specific to a reported `Error` (#114):
- **The five-minute TTL applies to `Error` like to any other state.** A
session that failed is still failed after five minutes, but a card pinned
red forever lies once the session moves on; when the report lapses the card
returns to the daemon's own inference — usually `Idle` for a session that
has gone quiet.
- **Only a new report clears an `Error` before its TTL.** The next
`ActivityState` of any kind replaces it (with #106's hook that is typically
`PreToolUse` → `Active` when the user prompts again). PTY output and
process changes do not clear it — only the agent itself may withdraw what
the agent itself reported. Re-reporting `Error` extends the TTL rather than
stacking reports.
If a client receives an `ActivityState` variant it does not know, it must not
fail rendering: treat it as `Idle` (neutral/dim, line-2 rules apply as usual)
until the daemon sends a variant it does know. Rust clients deserialize