hyperhive/hive-claude/src/classify.rs

329 lines
15 KiB
Rust

//! Sentinel detection: mapping claude-code CLI output onto [`crate::Error`]
//! variants.
//!
//! These marker strings are claude-code CLI knowledge, not app knowledge. They
//! are empirically stable across CLI versions; if one drifts the run degrades
//! gracefully (a clean turn, or a hard [`crate::Error::Exit`] on a non-zero
//! exit) rather than misbehaving.
//!
//! Stdout events are parsed into a typed [`StreamEvent`] (a serde-tagged enum)
//! rather than walked as a `serde_json::Value` — the event-type dispatch and
//! the terminal result event's fields (`is_error`, `api_error_status`) are
//! typed, so a shape drift is a clean fallback instead of a silent
//! mis-detection. The marker *strings* still gate detection (they're the
//! API's error text); typing removes the fragile field-probing around them.
use std::sync::atomic::{AtomicBool, Ordering};
use serde::Deserialize;
/// Emitted when the prompt/context exceeds the model's window.
const PROMPT_TOO_LONG_MARKER: &str = "Prompt is too long";
/// Substrings indicating the API refused for rate-limit / usage-cap / credit
/// reasons. On stdout these are only trusted inside a JSON `error` event (see
/// [`Sentinels::scan_stdout_json`] / [`Sentinels::scan_rate_limit_text`]) so a
/// model *discussing* a rate limit in prose can't trigger a false positive.
const RATE_LIMIT_MARKERS: [&str; 5] = [
"rate_limit_error",
"overloaded_error",
"Credit balance is too low",
"Usage limit reached",
"Request rate limit exceeded",
];
/// Substrings indicating the API rejected the request as unauthenticated (401)
/// — an expired/revoked OAuth session. Sourced from claude-code's `api_retry`
/// JSON events and its human-readable give-up line.
const AUTH_FAIL_MARKERS: [&str; 3] = [
"\"error\":\"authentication_failed\"",
"\"error_status\":401",
"Failed to authenticate. API Error: 401",
];
/// Substrings indicating `--resume` could not resolve its target: no session
/// with the given title, or no conversation with the given id.
const SESSION_NOT_FOUND_MARKERS: [&str; 2] = [
"does not match any session title",
"No conversation found with session ID",
];
/// The claude-code stdout stream-json events we classify, internally tagged
/// on `type`. Typing the dispatch (and the `result` event's fields below)
/// replaces walking a `serde_json::Value` with `get("field").and_then(...)`,
/// so a field-shape drift is a compile error / a clean fallback rather than a
/// silent mis-detection. Unknown event types fall into [`StreamEvent::Other`]
/// (scanned raw like any other control event).
#[derive(Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
enum StreamEvent {
/// Model-authored turn output — never scanned (it can quote any marker).
Assistant,
/// Model-authored tool results — never scanned.
User,
/// The terminal result event (see [`ResultEvent`]).
Result(ResultEvent),
/// A control `error` event — the only event a rate-limit marker is
/// trusted on (a model *discussing* a rate limit can't forge this type).
Error,
/// `system` and any other/unknown control event — claude-authored, so
/// scanned raw for the failure markers.
#[serde(other)]
Other,
}
/// The terminal `result` event's fields we act on. `subtype` is deliberately
/// absent: it reads `"success"` even on a hard failure, so it's not a usable
/// error signal — `is_error` is the discriminator (the CLI sets it from the
/// real outcome; the model can't forge it). On `is_error` the `result` /
/// `error` fields carry claude-code's own failure text.
#[derive(Deserialize)]
struct ResultEvent {
#[serde(default)]
is_error: bool,
/// claude-code's failure text on `is_error` (a genuine "Prompt is too
/// long" / auth message); the model's final answer on success.
#[serde(default)]
result: Option<String>,
/// Structured error payload (a string, or an object like
/// `{"type":"rate_limit_error"}`) present on some failures.
#[serde(default)]
error: Option<serde_json::Value>,
/// API status on an auth failure (e.g. `401`) — a typed signal that
/// doesn't depend on the human-readable string landing in `result`.
#[serde(default)]
api_error_status: Option<u32>,
}
/// Shared, lock-free sentinel flags accumulated while both output streams are
/// pumped concurrently. Read once after the child exits.
#[derive(Default)]
pub(crate) struct Sentinels {
prompt_too_long: AtomicBool,
rate_limited: AtomicBool,
auth_failed: AtomicBool,
session_not_found: AtomicBool,
}
impl Sentinels {
/// Scan a CLI-authored line — stderr, or a non-JSON stdout line — for every
/// marker. These bytes are always claude-code's own output, never model
/// conversation, so all markers are trusted.
pub(crate) fn scan_cli_line(&self, line: &str) {
self.scan_failure_markers(line);
if RATE_LIMIT_MARKERS.iter().any(|m| line.contains(m)) {
self.rate_limited.store(true, Ordering::Relaxed);
}
}
/// Scan a parsed stdout JSON event. **Skips model-authored `assistant` /
/// `user` message events**, whose serialized content can quote any marker
/// verbatim (an agent discussing this very code, say) — a false positive
/// that would otherwise trip a needless compact/retry or a spurious
/// auth/session error. Every real signal here is emitted *instead of* a
/// model turn (the API rejected the prompt, the auth failed, or `--resume`
/// missed before any inference), so it can only appear on a control event
/// (`error` / `result` / `system`) or as raw non-JSON text — never inside
/// an assistant/user message. This holds whatever exact shape claude-code
/// uses for the message, so the gate can't suppress a genuine signal.
pub(crate) fn scan_stdout_json(&self, event: &serde_json::Value, raw: &str) {
// Type the event instead of walking the `Value`. A parse failure
// (missing / non-string `type`) is rare and falls through to a raw
// scan — the safe default, matching the old "unknown type" branch.
match StreamEvent::deserialize(event) {
// Model-authored output — its serialized content can quote any
// marker verbatim (an agent discussing this very code), so it is
// never scanned. A genuine signal is emitted *instead of* a model
// turn, so it never rides an assistant/user event.
Ok(StreamEvent::Assistant | StreamEvent::User) => {}
// The terminal result event. On success its `result` field is the
// model's final answer (same trust level as an assistant message
// — must NOT scan). On a FAILED result (`is_error`) the `result` /
// `error` fields carry claude-code's OWN failure text — the real
// event (verified against captured stream-json) is
// {"type":"result","is_error":true,"subtype":"success",
// "result":"Prompt is too long","terminal_reason":"blocking_limit"}
// — note `subtype` is "success" even on a hard failure, so
// `is_error` is the discriminator (the model can't forge it).
Ok(StreamEvent::Result(r)) => {
if r.is_error {
if let Some(text) = &r.result {
self.scan_failure_markers(text);
}
if let Some(err) = &r.error {
// `to_string()` re-serializes the JSON value (a string
// value round-trips to `"...quoted..."`), but the marker
// substring still appears inside the quoted form, so the
// scan finds it either way.
self.scan_failure_markers(&err.to_string());
}
// Typed 401: an auth failure that set the status code but
// may not have put the human-readable string in `result`.
if r.api_error_status == Some(401) {
self.auth_failed.store(true, Ordering::Relaxed);
}
}
}
// A control `error` event — claude-authored end to end, and the
// only place a rate-limit marker is trusted.
Ok(StreamEvent::Error) => {
self.scan_failure_markers(raw);
if RATE_LIMIT_MARKERS.iter().any(|m| raw.contains(m)) {
self.rate_limited.store(true, Ordering::Relaxed);
}
}
// `system` / unknown control events (claude-authored) → scan raw.
Ok(StreamEvent::Other) | Err(_) => self.scan_failure_markers(raw),
}
}
/// The prompt-too-long / auth-failed / session-not-found markers. Callers
/// gate *where* this runs (see `scan_cli_line` / `scan_stdout_json`).
fn scan_failure_markers(&self, line: &str) {
if line.contains(PROMPT_TOO_LONG_MARKER) {
self.prompt_too_long.store(true, Ordering::Relaxed);
}
if AUTH_FAIL_MARKERS.iter().any(|m| line.contains(m)) {
self.auth_failed.store(true, Ordering::Relaxed);
}
if SESSION_NOT_FOUND_MARKERS.iter().any(|m| line.contains(m)) {
self.session_not_found.store(true, Ordering::Relaxed);
}
}
/// The recognized-sentinel error, if any fired — `None` means no sentinel
/// (so the run either completed or failed hard on its exit code). The
/// sentinels keep a fixed priority (too-long > rate > auth); a
/// session-not-found can only arise on a resume that made no model call,
/// so it never coincides with the others.
pub(crate) fn soft_error(&self) -> Option<crate::Error> {
use crate::Error;
if self.prompt_too_long.load(Ordering::Relaxed) {
Some(Error::PromptTooLong)
} else if self.rate_limited.load(Ordering::Relaxed) {
Some(Error::RateLimited)
} else if self.auth_failed.load(Ordering::Relaxed) {
Some(Error::AuthFailed)
} else if self.session_not_found.load(Ordering::Relaxed) {
Some(Error::SessionNotFound)
} else {
None
}
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::Error;
fn json(raw: &str) -> serde_json::Value {
serde_json::from_str(raw).unwrap()
}
#[test]
fn assistant_content_quoting_marker_is_ignored() {
// An agent discussing this code emits the marker verbatim in an
// assistant message — must NOT trip a sentinel.
let s = Sentinels::default();
let raw = r#"{"type":"assistant","message":{"content":[{"type":"text","text":"the CLI prints Prompt is too long on overflow"}]}}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(s.soft_error().is_none());
}
#[test]
fn genuine_failure_result_is_detected() {
// Both fixtures are the real claude-code failure shape, verified against
// captured stream-json: the marker lives in the `result` field with
// `is_error: true` and — counterintuitively — `subtype: "success"`. The
// earlier unconditional `result` scrub blinded this; `is_error` is the
// discriminator that restores it (and the model can't forge it).
let s = Sentinels::default();
let raw = r#"{"type":"result","is_error":true,"subtype":"success","result":"Prompt is too long","terminal_reason":"blocking_limit","stop_reason":"stop_sequence"}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(matches!(s.soft_error(), Some(Error::PromptTooLong)));
let s = Sentinels::default();
let raw = r#"{"type":"result","is_error":true,"subtype":"success","result":"Failed to authenticate. API Error: 401 Invalid authentication credentials","api_error_status":401}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(matches!(s.soft_error(), Some(Error::AuthFailed)));
}
#[test]
fn typed_api_error_status_401_trips_auth_without_string_marker() {
// The typed `api_error_status` field catches an auth failure even when
// the human-readable 401 string isn't in `result` — a signal the old
// substring scan would have missed.
let s = Sentinels::default();
let raw = r#"{"type":"result","is_error":true,"subtype":"success","result":"request failed","api_error_status":401}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(matches!(s.soft_error(), Some(Error::AuthFailed)));
}
#[test]
fn error_payload_object_on_result_is_scanned() {
// A failure result can carry the marker in a structured `error`
// payload rather than `result`; the typed `error: Value` is stringified
// and scanned.
let s = Sentinels::default();
let raw = r#"{"type":"result","is_error":true,"error":"Prompt is too long","result":"ok"}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(matches!(s.soft_error(), Some(Error::PromptTooLong)));
}
#[test]
fn result_field_model_text_quoting_marker_is_ignored() {
// The terminal `result` event's `result` field is the model's final
// answer. A marker quoted there (the model discussing this very code,
// or echoing an API error string) must NOT trip a sentinel — that was
// a turn-kill DoS. Covers prompt-too-long, auth, and session markers.
for marker in [
"Prompt is too long",
"Failed to authenticate. API Error: 401",
"does not match any session title",
] {
let s = Sentinels::default();
let raw = format!(
r#"{{"type":"result","subtype":"success","is_error":false,"result":"the harness scans stdout for {marker} — see classify.rs"}}"#
);
s.scan_stdout_json(&json(&raw), &raw);
assert!(
s.soft_error().is_none(),
"marker {marker:?} in the model-authored result field must be ignored"
);
}
}
#[test]
fn result_control_field_still_trips_even_with_clean_result_text() {
// Scrubbing `result` must not blind us to a genuine signal in a
// control field of the same event.
let s = Sentinels::default();
let raw = r#"{"type":"result","subtype":"error","is_error":true,"error":"Failed to authenticate. API Error: 401","result":"ok"}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(matches!(s.soft_error(), Some(Error::AuthFailed)));
}
#[test]
fn raw_non_json_marker_is_detected() {
let s = Sentinels::default();
s.scan_cli_line("API Error: Prompt is too long");
assert!(matches!(s.soft_error(), Some(Error::PromptTooLong)));
}
#[test]
fn rate_limit_still_only_on_error_event() {
// A non-error control event mentioning the marker must not trip it.
let s = Sentinels::default();
let raw = r#"{"type":"result","summary":"we hit a rate_limit_error earlier"}"#;
s.scan_stdout_json(&json(raw), raw);
assert!(s.soft_error().is_none());
// A genuine error event does.
let raw2 = r#"{"type":"error","error":{"type":"rate_limit_error"}}"#;
s.scan_stdout_json(&json(raw2), raw2);
assert!(matches!(s.soft_error(), Some(Error::RateLimited)));
}
}