car-multi 0.9.0

//! AdversarialReview — fresh agent reviews work against a spec.
//!
//! Key property: the reviewer gets NO prior context from the author.
//! It receives only the work output and the acceptance criteria, then
//! evaluates pass/fail with evidence (file:line references).
//!
//! Inspired by metaswarm's 4-phase execution loop where adversarial
//! reviewers are ALWAYS fresh Task() instances — never teammates,
//! never resumed, never given prior context.

use crate::error::MultiError;
use crate::mailbox::Mailbox;
use crate::runner::AgentRunner;
use crate::shared::SharedInfra;
use crate::types::{AgentOutput, AgentSpec};
use serde::{Deserialize, Serialize};
use std::sync::Arc;
use tracing::instrument;

/// A single review criterion with pass/fail and evidence.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ReviewFinding {
    pub criterion: String,
    pub passed: bool,
    pub evidence: String,
    pub severity: String, // "blocker", "major", "minor", "info"
}

/// Result of an adversarial review.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct AdversarialReviewResult {
    /// The task/spec being reviewed against.
    pub spec: String,
    /// Overall pass/fail.
    pub passed: bool,
    /// Per-criterion findings.
    pub findings: Vec<ReviewFinding>,
    /// The reviewer's raw output.
    pub reviewer_output: AgentOutput,
    /// Number of blockers found.
    pub blocker_count: usize,
}

/// Configuration for adversarial review.
pub struct AdversarialReview {
    /// The reviewer agent spec. Must be a different agent from the author.
    pub reviewer: AgentSpec,
    /// The acceptance criteria / spec to review against.
    pub criteria: Vec<String>,
    /// Whether blockers auto-fail the review.
    pub fail_on_blockers: bool,
}

impl AdversarialReview {
    pub fn new(reviewer: AgentSpec, criteria: Vec<String>) -> Self {
        Self {
            reviewer,
            criteria,
            fail_on_blockers: true,
        }
    }

    /// Run an adversarial review of the given work output.
    ///
    /// The reviewer is always a fresh agent with NO context from the author.
    /// It only sees: the work output and the acceptance criteria.
    #[instrument(name = "multi.adversarial_review", skip_all)]
    pub async fn run(
        &self,
        work_output: &str,
        runner: &Arc<dyn AgentRunner>,
        infra: &SharedInfra,
    ) -> Result<AdversarialReviewResult, MultiError> {
        let criteria_text = self.criteria.iter()
            .enumerate()
            .map(|(i, c)| format!("{}. {}", i + 1, c))
            .collect::<Vec<_>>()
            .join("\n");

        let review_task = format!(
            r#"You are an adversarial reviewer. Your job is to find problems.

## Acceptance Criteria
{criteria}

## Work Output to Review
{work}

## Instructions
Evaluate the work output against EACH acceptance criterion above.
For each criterion, determine if it PASSES or FAILS. Provide specific evidence
(file paths, line numbers, code snippets, or direct quotes from the output).

Be strict. If a criterion is ambiguous, assume it should be fully met.
Flag anything suspicious as a "blocker" or "major" finding.

Respond with a JSON object:
```json
{{
  "passed": true/false,
  "findings": [
    {{
      "criterion": "criterion text",
      "passed": true/false,
      "evidence": "specific evidence with file:line references",
      "severity": "blocker|major|minor|info"
    }}
  ]
}}
```"#,
            criteria = criteria_text,
            work = work_output,
        );

        let mailbox = Mailbox::default();
        let rt = infra.make_runtime();
        let output = runner
            .run(&self.reviewer, &review_task, &rt, &mailbox)
            .await
            .map_err(|e| MultiError::AgentFailed(
                self.reviewer.name.clone(),
                format!("adversarial review failed: {}", e),
            ))?;

        // Parse the review response
        let findings = Self::parse_findings(&output.answer);
        let blocker_count = findings.iter().filter(|f| f.severity == "blocker").count();
        let passed = if self.fail_on_blockers {
            blocker_count == 0 && findings.iter().all(|f| f.passed || f.severity != "major")
        } else {
            findings.iter().filter(|f| f.passed).count() > findings.len() / 2
        };

        Ok(AdversarialReviewResult {
            spec: criteria_text,
            passed,
            findings,
            reviewer_output: output,
            blocker_count,
        })
    }

    fn parse_findings(response: &str) -> Vec<ReviewFinding> {
        // Try to extract JSON
        if let Some(json_str) = car_ir::json_extract::extract_json_object(response) {
            if let Ok(parsed) = serde_json::from_str::<serde_json::Value>(&json_str) {
                if let Some(findings) = parsed.get("findings").and_then(|f| f.as_array()) {
                    return findings.iter().filter_map(|f| {
                        Some(ReviewFinding {
                            criterion: f.get("criterion")?.as_str()?.to_string(),
                            passed: f.get("passed")?.as_bool()?,
                            evidence: f.get("evidence")?.as_str()?.to_string(),
                            severity: f.get("severity").and_then(|s| s.as_str())
                                .unwrap_or("major").to_string(),
                        })
                    }).collect();
                }
            }
        }
        // Fallback: treat the whole response as a single finding
        vec![ReviewFinding {
            criterion: "overall".to_string(),
            passed: response.to_lowercase().contains("pass"),
            evidence: response.to_string(),
            severity: "major".to_string(),
        }]
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn parse_findings_from_json() {
        let response = r#"```json
{
  "passed": false,
  "findings": [
    {"criterion": "all tests pass", "passed": true, "evidence": "cargo test: 50 passed", "severity": "info"},
    {"criterion": "no hardcoded secrets", "passed": false, "evidence": "src/config.rs:42 contains API key", "severity": "blocker"}
  ]
}
```"#;
        let findings = AdversarialReview::parse_findings(response);
        assert_eq!(findings.len(), 2);
        assert!(findings[0].passed);
        assert!(!findings[1].passed);
        assert_eq!(findings[1].severity, "blocker");
    }

    #[test]
    fn parse_findings_fallback() {
        let response = "This looks good overall. PASS.";
        let findings = AdversarialReview::parse_findings(response);
        assert_eq!(findings.len(), 1);
        assert!(findings[0].passed);
    }
}