switchyard_libsy/algorithms/util/
escalation.rs

1// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2// SPDX-License-Identifier: Apache-2.0
3
4//! Trajectory-judge components for the escalation router — the judge, its verdict policy, and
5//! the transcript condenser they read.
6//!
7//! [`build_judge`] is the whole surface; the confirmation policy that consumes its verdicts
8//! lives with the assembled algorithm in [`crate::algorithms::escalation`].
9
10use serde::Deserialize;
11use serde_json::Value;
12use switchyard_protocol::{Category, InstructionBlock, Message, Role};
13
14use super::classifier_contract::{ClassifierContract, ClassifierContractConfig, validate_prompt};
15use super::llm_judge::{
16    ClassifierInput, JudgeClassifier, JudgePolicy, JudgeRuntimeConfig, SerdeDecoder,
17    StructuredJudge,
18};
19use super::transcript::{collect_text, message_text};
20use crate::core::algorithm::Driver;
21use crate::core::classifier::{Classification, Score};
22use crate::core::state::State;
23use crate::{LibsyError, Result};
24use switchyard_protocol::Request;
25
26const PROMPT_TEMPLATE: &str = include_str!("../../prompts/escalation/prompt.md");
27const DEESCALATION_PROMPT: &str = include_str!("../../prompts/escalation/deescalation.md");
28const SCHEMA_TEMPLATE: &str = include_str!("../../prompts/escalation/schema.json");
29
30/// Separator marking where [`truncate_middle`] dropped a message's interior.
31const TRIM_MARKER: &str = " ...[trimmed] ";
32
33/// Suffix marking a transcript cut off by [`MAX_REQUEST_CHARS`].
34const TRUNCATION_SUFFIX: &str = "...<truncated>";
35
36/// Per-message cap for system and developer anchors, which carry no trajectory signal but
37/// which coding-agent harnesses make very large.
38const SYSTEM_CHARS: usize = 1_000;
39
40/// Per-message cap for task-framing user messages — every user message that precedes the first
41/// assistant reply. Coding-agent harnesses often send environment boilerplate as the first user
42/// message and the task itself as the second, so anchoring only the first would pin the
43/// boilerplate and let the task scroll out of the window. Feature specifications run to several
44/// thousand characters, so this gets the widest anchor budget.
45const TASK_CHARS: usize = 4_000;
46
47/// Backstop on the assembled transcript; the per-message caps normally bind first.
48const MAX_REQUEST_CHARS: usize = 18_000;
49
50/// Optional policy for returning an escalated session to the efficient tier.
51#[derive(Clone, Copy, Debug, Deserialize)]
52#[serde(deny_unknown_fields)]
53pub struct DeescalationConfig {
54    /// Minimum number of capable-tier turns before the judge may release the session.
55    pub strong_min_calls: u32,
56    /// Optional hard limit on capable-tier turns before a forced return.
57    #[serde(default)]
58    pub strong_max_calls: Option<u32>,
59    /// Consecutive judge declines required to return to the efficient tier.
60    pub confirmations: u32,
61    /// Efficient calls served without judging after a hard-limit return.
62    #[serde(default)]
63    pub weak_cooldown_calls: u32,
64}
65
66impl DeescalationConfig {
67    fn validate(&self) -> Result<()> {
68        let reject = |message: &str| {
69            Err(LibsyError::AlgorithmError {
70                message: message.to_string(),
71            })
72        };
73        if self.strong_min_calls == 0 {
74            return reject("deescalation.strong_min_calls must be at least 1");
75        }
76        if self.confirmations == 0 {
77            return reject("deescalation.confirmations must be at least 1");
78        }
79        if self
80            .strong_max_calls
81            .is_some_and(|strong_max_calls| strong_max_calls < self.strong_min_calls)
82        {
83            return reject(
84                "deescalation.strong_max_calls must be at least deescalation.strong_min_calls",
85            );
86        }
87        Ok(())
88    }
89}
90
91/// The tuning surface for the trajectory judge.
92///
93/// The routing settings retain their benchmarked defaults. Everything else is a fixed invariant
94/// (the constants above).
95#[derive(Clone, Debug, Deserialize)]
96#[serde(default, deny_unknown_fields)]
97pub struct EscalationJudgeConfig {
98    /// Consecutive fresh-evidence verdicts in the same category required before a turn moves to
99    /// the capable tier, which is also the turn that latches the session. Any decline or stale
100    /// evidence clears the streak.
101    /// `1` escalates on the first verdict; the router's main cost dial.
102    /// `2` or higher needs a session id, since the streak is retained per session.
103    pub confirmations: u32,
104    /// Trailing messages shown on top of the anchors. A loop longer than this is invisible.
105    pub recent_turn_window: usize,
106    /// Per-message cap inside the trailing window.
107    pub window_message_chars: usize,
108    /// De-escalation policy. `None` preserves permanent latching to the capable tier.
109    pub deescalation: Option<DeescalationConfig>,
110}
111
112impl EscalationJudgeConfig {
113    /// Rejects settings that would leave the judge with nothing useful to read.
114    fn validate(&self) -> Result<()> {
115        let reject = |message: String| Err(LibsyError::AlgorithmError { message });
116        if self.confirmations == 0 {
117            return reject("confirmations must be at least 1".to_string());
118        }
119        if self.recent_turn_window == 0 {
120            return reject("recent_turn_window must be at least 1".to_string());
121        }
122        if self.window_message_chars < 50 {
123            return reject(format!(
124                "window_message_chars must be at least 50, got {}",
125                self.window_message_chars
126            ));
127        }
128        if let Some(deescalation) = self.deescalation {
129            deescalation.validate()?;
130        }
131        Ok(())
132    }
133}
134
135impl Default for EscalationJudgeConfig {
136    fn default() -> Self {
137        Self {
138            confirmations: 2,
139            recent_turn_window: 28,
140            window_message_chars: 500,
141            deescalation: None,
142        }
143    }
144}
145
146/// Router-controlled phase attached to judge input when de-escalation is enabled.
147#[derive(Clone, Copy, Debug, Eq, PartialEq)]
148pub(crate) enum EvaluationPhase {
149    Efficient,
150    Strong,
151}
152
153impl EvaluationPhase {
154    fn marker(self) -> &'static str {
155        match self {
156            Self::Efficient => "EFFICIENT_EVALUATION",
157            Self::Strong => "STRONG_EVALUATION",
158        }
159    }
160}
161
162/// Bounded trouble pattern used to correlate escalation confirmations across turns.
163#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq)]
164#[serde(rename_all = "snake_case")]
165pub(crate) enum EscalationCategory {
166    None,
167    Repetition,
168    FalseProgress,
169    Drift,
170    Desperation,
171    CapabilityGap,
172}
173
174impl EscalationCategory {
175    /// Stable state and telemetry label.
176    pub(crate) const fn label(self) -> &'static str {
177        match self {
178            Self::None => "none",
179            Self::Repetition => "repetition",
180            Self::FalseProgress => "false_progress",
181            Self::Drift => "drift",
182            Self::Desperation => "desperation",
183            Self::CapabilityGap => "capability_gap",
184        }
185    }
186}
187
188/// The judge's typed verdict, including the evidence needed to confirm a stable pattern.
189#[derive(Deserialize)]
190pub(crate) struct EscalationVerdict {
191    pub(crate) escalate: bool,
192    pub(crate) category: EscalationCategory,
193    pub(crate) new_evidence: bool,
194    pub(crate) reason: String,
195}
196
197/// Builds the condensed trajectory presented to the escalation judge.
198pub(crate) struct EscalationInput {
199    config: EscalationJudgeConfig,
200    phase: Option<EvaluationPhase>,
201}
202
203impl ClassifierInput for EscalationInput {
204    fn build_messages(&self, _state: &State, request: &Request) -> Vec<Message> {
205        let summary = summarize_for_judge(
206            &request.llm_request.instructions,
207            &request.llm_request.messages,
208            conversation_turn(request),
209            self.phase,
210            &self.config,
211        );
212        vec![Message::text(Role::User, summary)]
213    }
214}
215
216/// Structured trajectory judge with a typed escalation verdict.
217pub(crate) type EscalationJudge = StructuredJudge<EscalationInput, SerdeDecoder<EscalationVerdict>>;
218
219/// Maps the judge's verdict to a classification. A verdict names the tier to serve — capable
220/// on escalate, efficient on decline — so the caller reads it straight off the winning score.
221/// [`Classification::Ambiguous`] carries the unavailable case, which names no tier: both a
222/// decline and an outage stay efficient, but only a decline is evidence, so only a decline
223/// clears the streak.
224pub(crate) struct EscalationPolicy {
225    phase: Option<EvaluationPhase>,
226}
227
228impl JudgePolicy for EscalationPolicy {
229    type Verdict = EscalationVerdict;
230
231    fn to_classification(
232        &self,
233        verdict: Option<&EscalationVerdict>,
234        driver: &Driver,
235    ) -> Result<Classification> {
236        if let Some(verdict) = verdict {
237            tracing::debug!(
238                escalate = verdict.escalate,
239                reason = %verdict.reason,
240                "escalation judge verdict"
241            );
242        }
243        match verdict {
244            Some(verdict) if verdict.escalate => Ok(Classification::Scores(vec![Score {
245                target: driver.first_model_for(&Category::Capable)?.clone(),
246                confidence: 1.0,
247                category: Some(Category::Capable),
248            }])),
249            Some(_) => Ok(Classification::Scores(vec![Score {
250                target: driver.first_model_for(&Category::Efficient)?.clone(),
251                confidence: 1.0,
252                category: Some(Category::Efficient),
253            }])),
254            None => Ok(Classification::Ambiguous(Vec::new())),
255        }
256    }
257}
258
259/// Maps present verdicts to phase-aware evidence; absent verdicts add nothing.
260fn escalation_evidence(
261    policy: &EscalationPolicy,
262    verdict: Option<&EscalationVerdict>,
263) -> Option<Value> {
264    verdict.map(|verdict| {
265        let verdict = match (policy.phase, verdict.escalate) {
266            (Some(EvaluationPhase::Strong), true) => "retain",
267            (Some(EvaluationPhase::Strong), false) => "deescalate",
268            (_, true) => "escalate",
269            (_, false) => "continue",
270        };
271        serde_json::json!({
272            "source": "escalation",
273            "verdict": verdict,
274        })
275    })
276}
277
278/// Builds the trajectory judge, scoring the runtime capable category when it escalates.
279///
280/// Loads the packaged prompt and schema, so an unusable asset or an unusable `config` value
281/// fails here rather than on the first request.
282pub(crate) fn build_judge(
283    contract_config: &ClassifierContractConfig,
284    config: EscalationJudgeConfig,
285    phase: Option<EvaluationPhase>,
286    max_output_tokens: u64,
287) -> Result<JudgeClassifier<EscalationJudge, EscalationPolicy>> {
288    config.validate()?;
289    let contract = build_contract(contract_config, phase.is_some())?;
290    Ok(JudgeClassifier::new(
291        StructuredJudge::new(
292            EscalationInput { config, phase },
293            contract,
294            SerdeDecoder::new(),
295            JudgeRuntimeConfig::new(max_output_tokens)?,
296        ),
297        EscalationPolicy { phase },
298    )
299    .with_evidence(escalation_evidence))
300}
301
302fn build_contract(
303    contract_config: &ClassifierContractConfig,
304    phase_aware: bool,
305) -> Result<ClassifierContract> {
306    let prompt = contract_config.prompt().unwrap_or(PROMPT_TEMPLATE);
307    validate_prompt(prompt)?;
308    let phase_aware_config = phase_aware.then(|| {
309        contract_config.clone().with_prompt(format!(
310            "{}\n\n{}",
311            prompt.trim_end(),
312            DEESCALATION_PROMPT.trim()
313        ))
314    });
315    let contract_config = phase_aware_config.as_ref().unwrap_or(contract_config);
316    ClassifierContract::from_config(contract_config, PROMPT_TEMPLATE, SCHEMA_TEMPLATE)
317}
318
319/// The 1-indexed model invocation the transcript ends on: one per assistant reply.
320///
321/// The judge reads the turn *including* the reply it is judging, so the newest assistant
322/// message is this turn — no `+ 1`. Counting the caller's request instead would report the
323/// turn after the one under judgement.
324///
325/// Messages are already normalized by `switchyard-protocol`, so this needs no
326/// per-format branching.
327pub(crate) fn conversation_turn(request: &Request) -> usize {
328    request
329        .llm_request
330        .messages
331        .iter()
332        .filter(|message| message.role == Role::Assistant)
333        .count()
334}
335
336/// Keeps the head and tail of `text` within `limit` characters.
337///
338/// The head gets two thirds of the surviving budget: for a trajectory judge the command or
339/// error signature that opens a message carries more signal than its trailing output.
340fn truncate_middle(text: &str, limit: usize) -> String {
341    let chars: Vec<char> = text.chars().collect();
342    if chars.len() <= limit {
343        return text.to_string();
344    }
345    let keep = limit
346        .saturating_sub(TRIM_MARKER.chars().count())
347        .max(20)
348        .min(chars.len());
349    let head = keep * 2 / 3;
350    let tail = keep - head;
351    let mut out: String = chars[..head].iter().collect();
352    out.push_str(TRIM_MARKER);
353    out.extend(chars[chars.len() - tail..].iter());
354    out
355}
356
357/// Renders a compact role-labelled transcript for the judge.
358///
359/// Task-framing user messages are capped individually. System/developer instructions share
360/// the remaining budget after reserving space for task framing and the newest window entry.
361/// The trailing window carries recent activity. A coverage header states how much history is
362/// not shown, so the judge can reason about pace rather than assuming it sees everything.
363///
364/// When the assembled text still exceeds `max_request_chars`, the oldest window lines go
365/// first: for a trajectory judge the newest evidence is strictly the most valuable.
366fn summarize_for_judge(
367    instructions: &[InstructionBlock],
368    messages: &[Message],
369    turn: usize,
370    phase: Option<EvaluationPhase>,
371    config: &EscalationJudgeConfig,
372) -> String {
373    let mut instruction_anchors: Vec<String> = Vec::new();
374    let mut anchors: Vec<String> = Vec::new();
375    let mut window: Vec<String> = Vec::new();
376    let mut assistant_seen = false;
377
378    for instruction in instructions {
379        let mut parts = Vec::new();
380        collect_text(&instruction.content, &mut parts, &[]);
381        instruction_anchors.push(format!(
382            "[{}] {}",
383            role_label(instruction.role),
384            truncate_middle(&parts.join(" "), SYSTEM_CHARS)
385        ));
386    }
387
388    for message in messages {
389        let text = message_text(message);
390        match message.role {
391            Role::System | Role::Developer => instruction_anchors.push(format!(
392                "[{}] {}",
393                role_label(message.role),
394                truncate_middle(&text, SYSTEM_CHARS)
395            )),
396            // Everything the user said before the agent first replied is task framing.
397            Role::User if !assistant_seen => {
398                anchors.push(format!(
399                    "[user (task)] {}",
400                    truncate_middle(&text, TASK_CHARS)
401                ));
402            }
403            role => {
404                if role == Role::Assistant {
405                    assistant_seen = true;
406                }
407                window.push(format!(
408                    "[{}] {}",
409                    role_label(role),
410                    truncate_middle(&text, config.window_message_chars)
411                ));
412            }
413        }
414    }
415
416    if window.len() > config.recent_turn_window {
417        window.drain(..window.len() - config.recent_turn_window);
418    }
419
420    let assemble = |instructions: Option<&str>, window: &[String]| {
421        let header = format!(
422            "Conversation turn {turn}; showing the last {} of {} messages after the task framing.",
423            window.len(),
424            messages.len(),
425        );
426        phase
427            .map(|phase| format!("Routing phase: {}", phase.marker()))
428            .into_iter()
429            .chain(std::iter::once(header))
430            .chain(instructions.map(str::to_owned))
431            .chain(anchors.iter().cloned())
432            .chain(window.iter().cloned())
433            .collect::<Vec<_>>()
434            .join("\n")
435    };
436
437    let reserved = assemble(None, &window[window.len().saturating_sub(1)..])
438        .chars()
439        .count();
440    let instruction_budget = MAX_REQUEST_CHARS.saturating_sub(reserved + 1);
441    // The remaining budget may be smaller than truncate_middle's minimum retained span.
442    let instruction_text = truncate_middle(&instruction_anchors.join("\n"), instruction_budget)
443        .chars()
444        .take(instruction_budget)
445        .collect::<String>();
446    let instructions = (!instruction_text.is_empty()).then_some(instruction_text.as_str());
447    let mut text = assemble(instructions, &window);
448    while text.chars().count() > MAX_REQUEST_CHARS && !window.is_empty() {
449        window.remove(0);
450        text = assemble(instructions, &window);
451    }
452    if text.chars().count() > MAX_REQUEST_CHARS {
453        let keep = MAX_REQUEST_CHARS.saturating_sub(TRUNCATION_SUFFIX.chars().count() + 1);
454        text = text.chars().take(keep).collect::<String>() + TRUNCATION_SUFFIX;
455    }
456    text
457}
458
459/// The transcript label for a role.
460fn role_label(role: Role) -> &'static str {
461    match role {
462        Role::System => "system",
463        Role::Developer => "developer",
464        Role::User => "user",
465        Role::Assistant => "assistant",
466        Role::Tool => "tool",
467    }
468}
469
470/// A request whose conversation sits at `turn`: `turn - 1` prior assistant replies, each
471/// answered by a further user message.
472///
473/// Shared with the assembled router's tests, which drive the same conversation shape.
474#[cfg(test)]
475pub(crate) fn request_at_turn(session_id: Option<&str>, turn: usize) -> Request {
476    use switchyard_protocol::{LlmRequest, Metadata};
477
478    let mut messages = vec![Message::text(Role::User, "What is 2+2?")];
479    for attempt in 1..turn {
480        messages.push(Message::text(Role::Assistant, format!("attempt {attempt}")));
481        messages.push(Message::text(Role::User, format!("still wrong {attempt}")));
482    }
483    Request {
484        llm_request: LlmRequest {
485            model: Some("auto".to_string()),
486            messages,
487            ..LlmRequest::default()
488        },
489        raw_request: None,
490        metadata: session_id.map(|id| Metadata {
491            session_id: Some(id.to_string()),
492            ..Metadata::default()
493        }),
494    }
495}
496
497#[cfg(test)]
498mod tests {
499    use switchyard_protocol::{ContentBlock, Message, Role};
500
501    use super::*;
502    use crate::algorithms::util::llm_judge::Judge;
503
504    fn escalation_judge(
505        max_output_tokens: u64,
506        phase: Option<EvaluationPhase>,
507        contract_config: &ClassifierContractConfig,
508    ) -> Result<EscalationJudge> {
509        Ok(StructuredJudge::new(
510            EscalationInput {
511                config: EscalationJudgeConfig::default(),
512                phase,
513            },
514            build_contract(contract_config, phase.is_some())?,
515            SerdeDecoder::new(),
516            JudgeRuntimeConfig::new(max_output_tokens)?,
517        ))
518    }
519
520    #[test]
521    fn judge_request_is_rubric_plus_summary_under_a_completion_cap() -> Result<()> {
522        let judge = escalation_judge(
523            super::super::DEFAULT_JUDGE_MAX_OUTPUT_TOKENS,
524            None,
525            &ClassifierContractConfig::default(),
526        )?;
527
528        // As the classifier calls it: the turn's reply is already on the transcript.
529        let mut judged = request_at_turn(None, 4);
530        judged.llm_request.instructions = [
531            (Role::System, "system constraint"),
532            (Role::Developer, "developer constraint"),
533        ]
534        .into_iter()
535        .map(|(role, text)| InstructionBlock {
536            role,
537            content: Message::text(role, text).content,
538        })
539        .collect();
540        judged
541            .llm_request
542            .messages
543            .push(Message::text(Role::Assistant, "this turn's reply"));
544        let built = judge.build_request(&State::default(), &judged);
545
546        // Rubric in instructions, condensed trajectory as the sole user message.
547        assert_eq!(built.llm_request.instructions.len(), 1);
548        assert_eq!(built.llm_request.instructions[0].role, Role::System);
549        assert_eq!(
550            built.llm_request.instructions[0].content.as_slice(),
551            &[ContentBlock::Text {
552                text: PROMPT_TEMPLATE.to_string()
553            }]
554        );
555        assert_eq!(built.llm_request.messages.len(), 1);
556        assert_eq!(built.llm_request.messages[0].role, Role::User);
557        let summary = built.llm_request.messages[0]
558            .text_content("")
559            .expect("summary");
560        assert!(summary.contains("Conversation turn 4"));
561        assert!(summary.contains(
562            "[system] system constraint\n[developer] developer constraint\n[user (task)] What is 2+2?"
563        ));
564        assert!(summary.contains("[assistant] this turn's reply"));
565        assert!(!summary.contains("Routing phase:"));
566        // Bounded output, so a reasoning judge cannot run away mid-verdict.
567        assert_eq!(
568            built.llm_request.output.max_output_tokens,
569            Some(super::super::DEFAULT_JUDGE_MAX_OUTPUT_TOKENS)
570        );
571        assert!(built.llm_request.output.response_format.is_some());
572
573        judged.llm_request.instructions.extend(vec![
574            InstructionBlock {
575                role: Role::Developer,
576                content: Message::text(Role::Developer, "x".repeat(SYSTEM_CHARS)).content,
577            };
578            MAX_REQUEST_CHARS / SYSTEM_CHARS + 1
579        ]);
580        let built = judge.build_request(&State::default(), &judged);
581        let summary = built.llm_request.messages[0]
582            .text_content("")
583            .expect("summary");
584        assert!(summary.chars().count() <= MAX_REQUEST_CHARS);
585        assert!(summary.contains(TRIM_MARKER));
586        assert!(summary.contains("[system] system constraint"));
587        assert!(summary.contains("[developer] developer constraint"));
588        assert!(summary.contains("[user (task)] What is 2+2?"));
589        assert!(summary.contains("[assistant] this turn's reply"));
590        Ok(())
591    }
592
593    #[test]
594    fn judge_request_uses_the_configured_completion_cap() -> Result<()> {
595        let judge = escalation_judge(512, None, &ClassifierContractConfig::default())?;
596
597        let built = judge.build_request(&State::default(), &request_at_turn(None, 1));
598
599        assert_eq!(built.llm_request.output.max_output_tokens, Some(512));
600        Ok(())
601    }
602
603    #[test]
604    fn conversation_turn_counts_assistant_replies() {
605        // The judge is handed the transcript with this turn's reply already appended, which
606        // is the shape asserted here: a request entering turn N plus its reply *is* turn N.
607        for turn in [1, 5] {
608            let mut judged = request_at_turn(None, turn);
609            judged
610                .llm_request
611                .messages
612                .push(Message::text(Role::Assistant, "this turn's reply"));
613            assert_eq!(conversation_turn(&judged), turn);
614        }
615    }
616
617    #[test]
618    fn truncate_middle_keeps_head_and_tail() {
619        let text = "a".repeat(40) + &"z".repeat(40);
620        let trimmed = truncate_middle(&text, 50);
621        assert!(trimmed.chars().count() <= 50, "{trimmed}");
622        assert!(trimmed.starts_with('a'));
623        assert!(trimmed.ends_with('z'));
624        assert!(trimmed.contains("[trimmed]"));
625
626        // Under the limit the text is returned untouched.
627        assert_eq!(truncate_middle("short", 50), "short");
628    }
629
630    #[test]
631    fn deescalation_settings_must_be_valid() {
632        let zero = EscalationJudgeConfig {
633            deescalation: Some(DeescalationConfig {
634                strong_min_calls: 0,
635                strong_max_calls: None,
636                confirmations: 2,
637                weak_cooldown_calls: 0,
638            }),
639            ..EscalationJudgeConfig::default()
640        };
641        assert!(
642            zero.validate()
643                .is_err_and(|error| error.to_string().contains("at least 1"))
644        );
645
646        let inverted = EscalationJudgeConfig {
647            deescalation: Some(DeescalationConfig {
648                strong_min_calls: 4,
649                strong_max_calls: Some(3),
650                confirmations: 2,
651                weak_cooldown_calls: 0,
652            }),
653            ..EscalationJudgeConfig::default()
654        };
655        assert!(
656            inverted
657                .validate()
658                .is_err_and(|error| error.to_string().contains("at least deescalation"))
659        );
660    }
661
662    #[test]
663    fn deescalation_contract_marks_both_routing_phases() -> Result<()> {
664        let contract = ClassifierContractConfig::default().with_prompt("Custom trajectory rubric.");
665        for (phase, marker) in [
666            (EvaluationPhase::Efficient, "EFFICIENT_EVALUATION"),
667            (EvaluationPhase::Strong, "STRONG_EVALUATION"),
668        ] {
669            let judge = escalation_judge(512, Some(phase), &contract)?;
670            let built = judge.build_request(&State::default(), &request_at_turn(None, 1));
671            let system_prompt = built.llm_request.instructions[0].content.iter().find_map(
672                |content| match content {
673                    ContentBlock::Text { text } => Some(text.as_str()),
674                    _ => None,
675                },
676            );
677            assert!(system_prompt.is_some_and(|prompt| {
678                prompt.starts_with("Custom trajectory rubric.")
679                    && prompt.contains("EFFICIENT_EVALUATION")
680                    && prompt.contains("STRONG_EVALUATION")
681            }));
682            assert!(
683                built.llm_request.messages[0]
684                    .text_content("")
685                    .is_some_and(|summary| summary.starts_with(&format!(
686                        "Routing phase: {marker}"
687                    )))
688            );
689        }
690        Ok(())
691    }
692
693    #[test]
694    fn summary_keeps_anchors_and_the_recent_window() {
695        let mut messages = vec![
696            Message::text(Role::System, "you are a coding agent"),
697            Message::text(Role::User, "fix the failing test"),
698        ];
699        for i in 0..10 {
700            messages.push(Message::text(Role::Assistant, format!("step {i}")));
701        }
702        let config = EscalationJudgeConfig {
703            recent_turn_window: 3,
704            ..EscalationJudgeConfig::default()
705        };
706
707        let summary = summarize_for_judge(&[], &messages, 11, None, &config);
708
709        assert!(
710            summary.contains("[system] you are a coding agent"),
711            "{summary}"
712        );
713        assert!(
714            summary.contains("[user (task)] fix the failing test"),
715            "{summary}"
716        );
717        assert!(summary.contains("Conversation turn 11; showing the last 3 of 12 messages"));
718        // Only the newest window entries survive.
719        assert!(summary.contains("step 9"), "{summary}");
720        assert!(summary.contains("step 7"), "{summary}");
721        assert!(!summary.contains("step 6"), "{summary}");
722    }
723
724    #[test]
725    fn summary_anchors_every_user_message_before_the_first_reply() {
726        // Codex sends environment boilerplate as the first user message and the task as the
727        // second. Both are framing; the task must stay visible after the window has moved on.
728        let mut messages = vec![
729            Message::text(
730                Role::Developer,
731                "<skills_instructions>...</skills_instructions>",
732            ),
733            Message::text(
734                Role::User,
735                "<environment_context><cwd>/app</cwd></environment_context>",
736            ),
737            Message::text(Role::User, "Implement RFC 5545 timezone interop in rrule."),
738        ];
739        for i in 0..40 {
740            messages.push(Message::text(Role::Assistant, format!("step {i}")));
741            messages.push(Message::text(Role::User, format!("later user note {i}")));
742        }
743        let config = EscalationJudgeConfig {
744            recent_turn_window: 3,
745            ..EscalationJudgeConfig::default()
746        };
747
748        let summary = summarize_for_judge(&[], &messages, 40, None, &config);
749
750        assert!(
751            summary.contains("[user (task)] <environment_context>"),
752            "{summary}"
753        );
754        assert!(
755            summary.contains("[user (task)] Implement RFC 5545 timezone interop in rrule."),
756            "{summary}"
757        );
758        // User messages after the first reply are ordinary window entries, not anchors.
759        assert!(
760            !summary.contains("[user (task)] later user note"),
761            "{summary}"
762        );
763        assert!(summary.contains("[user] later user note 39"), "{summary}");
764        assert!(!summary.contains("later user note 0\n"), "{summary}");
765    }
766
767    #[test]
768    fn summary_drops_oldest_window_lines_under_the_char_cap() {
769        // MAX_REQUEST_CHARS is a backstop, not a dial: at default settings the window caps
770        // bind first (28 x 500 plus anchors sits under it), so reaching it takes an unusually
771        // wide per-message cap. That is the point — it only fires on pathological input.
772        let mut messages = vec![
773            Message::text(Role::System, "framing"),
774            Message::text(Role::User, "task"),
775        ];
776        for i in 0..20 {
777            messages.push(Message::text(
778                Role::Assistant,
779                format!("{i} {}", "x".repeat(2_000)),
780            ));
781        }
782        let config = EscalationJudgeConfig {
783            window_message_chars: 2_000,
784            ..EscalationJudgeConfig::default()
785        };
786
787        let summary = summarize_for_judge(&[], &messages, 21, None, &config);
788
789        assert!(
790            summary.chars().count() <= MAX_REQUEST_CHARS,
791            "{}",
792            summary.chars().count()
793        );
794        // Anchors are never dropped, and the newest activity outlives the oldest.
795        assert!(summary.contains("[system] framing"), "{summary}");
796        assert!(summary.contains("[user (task)] task"), "{summary}");
797        assert!(summary.contains("19 xxx"), "{summary}");
798        assert!(!summary.contains("0 xxx"), "{summary}");
799    }
800}
switchyard_libsy/algorithms/util/escalation.rs

switchyard_libsy/algorithms/util/
escalation.rs
