assay-cli 3.31.0

Policy-as-code gate for MCP agent tool calls, with verifiable evidence and Linux kernel enforcement.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
use crate::exit_codes::{ReasonCode, RunOutcome};
use std::path::{Path, PathBuf};

fn reason_code_from_run_error_kind(kind: assay_core::errors::RunErrorKind) -> Option<ReasonCode> {
    use assay_core::errors::RunErrorKind;

    match kind {
        RunErrorKind::TraceNotFound => Some(ReasonCode::ETraceNotFound),
        RunErrorKind::MissingConfig => Some(ReasonCode::EMissingConfig),
        RunErrorKind::ConfigParse => Some(ReasonCode::ECfgParse),
        RunErrorKind::InvalidArgs => Some(ReasonCode::EInvalidArgs),
        RunErrorKind::ProviderRateLimit => Some(ReasonCode::ERateLimit),
        RunErrorKind::ProviderTimeout => Some(ReasonCode::ETimeout),
        RunErrorKind::ProviderServer => Some(ReasonCode::EProvider5xx),
        RunErrorKind::Network => Some(ReasonCode::ENetworkError),
        RunErrorKind::JudgeUnavailable => Some(ReasonCode::EJudgeUnavailable),
        RunErrorKind::Other => None,
    }
}

pub(crate) fn reason_code_from_run_error(err: &assay_core::errors::RunError) -> Option<ReasonCode> {
    reason_code_from_run_error_kind(err.kind.clone())
}

fn run_error_kind_from_details(
    details: &serde_json::Value,
) -> Option<assay_core::errors::RunErrorKind> {
    use assay_core::errors::RunErrorKind;

    let kind = details.get("run_error_kind")?.as_str()?;
    match kind {
        "trace_not_found" => Some(RunErrorKind::TraceNotFound),
        "missing_config" => Some(RunErrorKind::MissingConfig),
        "config_parse" => Some(RunErrorKind::ConfigParse),
        "invalid_args" => Some(RunErrorKind::InvalidArgs),
        "provider_rate_limit" => Some(RunErrorKind::ProviderRateLimit),
        "provider_timeout" => Some(RunErrorKind::ProviderTimeout),
        "provider_server" => Some(RunErrorKind::ProviderServer),
        "network" => Some(RunErrorKind::Network),
        "judge_unavailable" => Some(RunErrorKind::JudgeUnavailable),
        "other" => Some(RunErrorKind::Other),
        _ => None,
    }
}

fn reason_code_from_result_row(row: &assay_core::model::TestResultRow) -> Option<ReasonCode> {
    // Typed-first mapping in hot path; message classification remains explicit legacy fallback.
    if let Some(kind) = run_error_kind_from_details(&row.details) {
        return reason_code_from_run_error_kind(kind);
    }
    reason_code_from_error_message(&row.message)
}

pub(crate) fn reason_code_from_error_message(message: &str) -> Option<ReasonCode> {
    use assay_core::errors::RunError;

    // Legacy-only fallback for untyped paths.
    let classified = RunError::legacy_classify_message(message.to_string());
    reason_code_from_run_error(&classified)
}

pub(crate) fn decide_run_outcome(
    results: &[assay_core::model::TestResultRow],
    strict: bool,
    version: crate::exit_codes::ExitCodeVersion,
) -> crate::exit_codes::RunOutcome {
    use assay_core::model::TestStatus;

    // Helper to ensure exit code matches requested version
    let make_outcome = |reason: ReasonCode, msg: Option<String>, context: Option<&str>| {
        let mut o = RunOutcome::from_reason(reason, msg, context);
        o.exit_code = reason.exit_code_for(version);
        o
    };

    // Priority 1: Config/Argument Errors (Exit 2)
    for r in results {
        if let Some(reason) = reason_code_from_result_row(r) {
            if matches!(
                reason,
                ReasonCode::ETraceNotFound
                    | ReasonCode::EMissingConfig
                    | ReasonCode::ECfgParse
                    | ReasonCode::EInvalidArgs
            ) {
                return make_outcome(reason, Some(r.message.clone()), None);
            }
        }
    }

    // Priority 2: Infrastructure Failures (Refined Heuristics)
    let infra_errors: Vec<&assay_core::model::TestResultRow> = results
        .iter()
        .filter(|r| matches!(r.status, TestStatus::Error))
        .collect();

    if !infra_errors.is_empty() {
        let reason = pick_infra_reason(&infra_errors);
        return make_outcome(
            reason,
            Some("Infrastructure failures detected".into()),
            None,
        );
    }

    // Priority 3: Judge uncertain (abstain) — exit 1, E_JUDGE_UNCERTAIN
    let abstain_count = results
        .iter()
        .filter(|r| has_judge_verdict_abstain(&r.details))
        .count();
    if abstain_count > 0 {
        let mut o = RunOutcome::judge_uncertain(abstain_count);
        o.exit_code = ReasonCode::EJudgeUncertain.exit_code_for(version);
        return o;
    }

    // Priority 4: Test Failures
    let fails = results
        .iter()
        .filter(|r| matches!(r.status, TestStatus::Fail))
        .count();
    if fails > 0 {
        let mut o = RunOutcome::test_failure(fails);
        o.exit_code = ReasonCode::ETestFailed.exit_code_for(version);
        return o;
    }

    // Priority 5: Strict Mode Violations
    if strict {
        let violations = results
            .iter()
            .filter(|r| {
                matches!(
                    r.status,
                    TestStatus::Warn | TestStatus::Flaky | TestStatus::Unstable
                )
            })
            .count();
        if violations > 0 {
            return make_outcome(
                ReasonCode::EPolicyViolation,
                Some(format!("Strict mode: {} policy violations", violations)),
                None,
            );
        }
    }

    // Priority 6: Success (ensure version compliance though Success is usually 0 in all versions)
    let mut o = RunOutcome::success();
    o.exit_code = ReasonCode::Success.exit_code_for(version);
    o
}

/// True if this result row has any judge metric with verdict "Abstain" (E7.5).
pub(crate) fn has_judge_verdict_abstain(details: &serde_json::Value) -> bool {
    let Some(metrics) = details.get("metrics").and_then(|m| m.as_object()) else {
        return false;
    };
    for (_name, metric_val) in metrics {
        if let Some(inner) = metric_val.get("details").and_then(|d| d.get("verdict")) {
            if inner.as_str() == Some("Abstain") {
                return true;
            }
        }
    }
    false
}

fn pick_infra_reason(
    errors: &[&assay_core::model::TestResultRow],
) -> crate::exit_codes::ReasonCode {
    for r in errors {
        if let Some(reason) = reason_code_from_result_row(r) {
            if matches!(
                reason,
                ReasonCode::ERateLimit
                    | ReasonCode::ETimeout
                    | ReasonCode::EProvider5xx
                    | ReasonCode::ENetworkError
                    | ReasonCode::EJudgeUnavailable
            ) {
                return reason;
            }
        }
    }
    ReasonCode::EJudgeUnavailable
}

/// Build a Summary from RunOutcome for writing summary.json (same dir as run.json).
pub(crate) fn summary_from_outcome(
    outcome: &crate::exit_codes::RunOutcome,
    verify_enabled: bool,
) -> assay_core::report::summary::Summary {
    let assay_version = env!("CARGO_PKG_VERSION");
    if outcome.exit_code == 0 {
        assay_core::report::summary::Summary::success(assay_version, verify_enabled)
    } else {
        assay_core::report::summary::Summary::failure(
            outcome.exit_code,
            &outcome.reason_code,
            outcome.message.as_deref().unwrap_or(""),
            outcome.next_step.as_deref().unwrap_or(""),
            assay_version,
            verify_enabled,
        )
    }
}

pub(crate) fn write_extended_run_json(
    artifacts: &assay_core::report::RunArtifacts,
    outcome: &crate::exit_codes::RunOutcome,
    path: &Path,
    sarif_omitted: Option<u64>,
) -> anyhow::Result<()> {
    // Manually construct the JSON to inject outcome fields
    let mut v = serde_json::to_value(artifacts)?;
    if let Some(obj) = v.as_object_mut() {
        // Inject top-level outcome fields for machine-readability (Canonical Contract)
        obj.insert(
            "exit_code".to_string(),
            serde_json::json!(outcome.exit_code),
        );
        obj.insert(
            "reason_code".to_string(),
            serde_json::json!(outcome.reason_code),
        );
        obj.insert(
            "reason_code_version".to_string(),
            serde_json::json!(assay_core::report::summary::REASON_CODE_VERSION),
        );

        // E7.2: seeds always present; order_seed/judge_seed as string or null (SPEC: avoid JSON number precision loss)
        obj.insert(
            "seed_version".to_string(),
            serde_json::json!(assay_core::report::summary::SEED_VERSION),
        );
        let order_seed_json = match artifacts.order_seed {
            Some(n) => serde_json::Value::String(n.to_string()),
            None => serde_json::Value::Null,
        };
        obj.insert("order_seed".to_string(), order_seed_json);
        obj.insert("judge_seed".to_string(), serde_json::Value::Null);

        // E7.3: judge metrics when present
        if let Some(metrics) =
            assay_core::report::summary::judge_metrics_from_results(&artifacts.results)
        {
            obj.insert("judge_metrics".to_string(), serde_json::to_value(metrics)?);
        }

        // E2.3: SARIF truncation metadata when SARIF was truncated
        if let Some(n) = sarif_omitted {
            if n > 0 {
                obj.insert("sarif".to_string(), serde_json::json!({ "omitted": n }));
            }
        }

        // Conflict avoidance: Move full details to 'resolution' object
        // Do NOT inject 'message' or 'next_step' top-level to avoid collisions with artifact fields.
        obj.insert("resolution".to_string(), serde_json::to_value(outcome)?);

        if !outcome.warnings.is_empty() {
            obj.insert("warnings".into(), serde_json::json!(outcome.warnings));
        }
    }

    write_sanitized_run_json(path, v)
}

/// Sanitize a run.json document and write it. Render-safety (MCP01a-5): EVERY run.json writer routes
/// its assembled value through the same `render_details_safe` walk as the `--format json` stdout
/// renderer (`assay_core::report::json::render_json`), so the untrusted result `message` /
/// `details.*` / `resolution.message` content is redacted on disk too — the file-writers used to
/// serialize raw artifacts and bypass redaction. As a record sink it redacts and control-strips but
/// does not truncate (`usize::MAX`); assay-owned keys (ids, status, reason_code, exit_code, digests,
/// seeds) stay byte-stable. Shared by both the extended and minimal (early-error) writers so neither
/// path can drift back to raw.
fn write_sanitized_run_json(path: &Path, v: serde_json::Value) -> anyhow::Result<()> {
    let v = assay_core::render_safety::render_details_safe(
        assay_core::render_safety::Sink::Json,
        &v,
        usize::MAX,
    );
    if let Some(parent) = path.parent() {
        std::fs::create_dir_all(parent)?;
    }
    std::fs::write(path, serde_json::to_string_pretty(&v)?)?;
    Ok(())
}

pub(crate) fn write_run_json_minimal(
    outcome: &crate::exit_codes::RunOutcome,
    path: &Path,
) -> anyhow::Result<()> {
    // Minimal JSON for early exits (no artifacts available). E7.2: seed fields present for schema stability (null when unknown).
    // `resolution` carries the outcome (incl. its `message`, which on config/arg errors echoes
    // external path/parse text), so this goes through the shared sanitizer like the extended writer.
    let v = serde_json::json!({
        "exit_code": outcome.exit_code,
        "reason_code": outcome.reason_code,
        "reason_code_version": assay_core::report::summary::REASON_CODE_VERSION,
        "seed_version": assay_core::report::summary::SEED_VERSION,
        "order_seed": null,
        "judge_seed": null,
        "resolution": outcome
    });
    write_sanitized_run_json(path, v)
}

pub(crate) fn export_baseline(
    path: &PathBuf,
    config_path: &Path,
    cfg: &assay_core::model::EvalConfig,
    results: &[assay_core::model::TestResultRow],
) -> anyhow::Result<()> {
    let mut entries = Vec::new();

    for r in results {
        if let Some(metrics) = r.details.get("metrics").and_then(|v| v.as_object()) {
            for (metric_name, m_val) in metrics {
                if let Some(score) = m_val.get("score").and_then(|s| s.as_f64()) {
                    entries.push(assay_core::baseline::BaselineEntry {
                        test_id: r.test_id.clone(),
                        metric: metric_name.clone(),
                        score,
                        meta: None,
                    });
                }
            }
        }
    }

    let b = assay_core::baseline::Baseline {
        schema_version: 1,
        suite: cfg.suite.clone(),
        assay_version: env!("CARGO_PKG_VERSION").to_string(),
        created_at: chrono::Utc::now().to_rfc3339(),
        config_fingerprint: assay_core::baseline::compute_config_fingerprint(config_path),
        git_info: None,
        entries,
    };

    b.save(path)?;
    eprintln!("exported baseline to {}", path.display());
    Ok(())
}

#[cfg(test)]
mod run_json_render_safety_tests {
    use super::write_extended_run_json;
    use crate::exit_codes::{ReasonCode, RunOutcome};
    use assay_core::model::{TestResultRow, TestStatus};
    use assay_core::report::RunArtifacts;

    /// MCP01a-5 regression: the on-disk run.json file writer must redact untrusted result
    /// `message` / `details.*` content, matching the `--format json` stdout renderer. This guards the
    /// gap where `write_extended_run_json` serialized raw artifacts and leaked secrets to the file.
    #[test]
    fn on_disk_run_json_is_render_safe() {
        let secret = format!("ghp_{}", "A".repeat(36));
        let artifacts = RunArtifacts {
            run_id: 7,
            suite: "owned-suite".into(),
            results: vec![TestResultRow {
                test_id: "t_owned".into(),
                status: TestStatus::Fail,
                score: Some(0.0),
                cached: false,
                message: format!("failed: leaked {secret}"),
                details: serde_json::json!({
                    "prompt": format!("ask {secret} alice@example.com"),
                    "metrics": { "must_contain": { "details": { "message": format!("missing {secret}") } } },
                }),
                duration_ms: Some(3),
                fingerprint: Some("fp_owned".into()),
                skip_reason: None,
                attempts: None,
                error_policy_applied: None,
            }],
            order_seed: None,
            runner_clone_ms: None,
        };
        let outcome = RunOutcome::from_reason(ReasonCode::ETestFailed, None, None);
        let dir = tempfile::tempdir().unwrap();
        let path = dir.path().join("run.json");
        write_extended_run_json(&artifacts, &outcome, &path, None).unwrap();
        let out = std::fs::read_to_string(&path).unwrap();

        // Structure preserved + hostile values gone, markers present.
        let parsed: serde_json::Value = serde_json::from_str(&out).unwrap();
        assert!(!out.contains(&secret), "on-disk run.json leaked secret");
        assert!(!out.contains("ghp_"), "raw token prefix leaked to run.json");
        assert!(
            !out.contains("alice@example.com"),
            "on-disk run.json leaked pii"
        );
        assert!(
            out.contains("<redacted:"),
            "on-disk run.json fired no redaction"
        );
        // Assay-owned fields stay byte-stable.
        assert_eq!(parsed["suite"], "owned-suite");
        assert_eq!(parsed["results"][0]["test_id"], "t_owned");
        assert_eq!(parsed["results"][0]["fingerprint"], "fp_owned");
        assert_eq!(parsed["reason_code"], "E_TEST_FAILED");
    }

    /// MCP01a-5 regression: the early-error / replay-failure minimal writer shares the same sanitizer,
    /// so a config/arg-error `resolution.message` (which can echo external path/parse text) is redacted
    /// on disk too — not just the artifact-bearing extended writer.
    #[test]
    fn minimal_run_json_resolution_message_is_render_safe() {
        let secret = format!("ghp_{}", "Z".repeat(36));
        let outcome = RunOutcome::from_reason(
            ReasonCode::ETraceNotFound,
            Some(format!("trace not found: {secret}")),
            None,
        );
        let dir = tempfile::tempdir().unwrap();
        let path = dir.path().join("run.json");
        super::write_run_json_minimal(&outcome, &path).unwrap();
        let out = std::fs::read_to_string(&path).unwrap();
        let parsed: serde_json::Value = serde_json::from_str(&out).unwrap();
        assert!(
            !out.contains(&secret),
            "minimal run.json leaked secret in resolution.message"
        );
        assert!(!out.contains("ghp_"), "raw token prefix leaked to run.json");
        assert!(
            out.contains("<redacted:"),
            "minimal run.json fired no redaction"
        );
        // Assay-owned reason_code stays byte-stable.
        assert_eq!(parsed["reason_code"], "E_TRACE_NOT_FOUND");
    }
}

#[cfg(test)]
mod run_outcome_tests {
    use super::{
        decide_run_outcome, has_judge_verdict_abstain, reason_code_from_error_message,
        reason_code_from_run_error,
    };
    use crate::exit_codes::{ExitCodeVersion, ReasonCode, EXIT_INFRA_ERROR};
    use assay_core::errors::RunError;
    use assay_core::model::{TestResultRow, TestStatus};

    #[test]
    fn test_infra_beats_abstain_precedence() {
        // Precedence: infra (exit 3) must win over abstain (exit 1). One Error + one Abstain → exit 3.
        let results = vec![
            TestResultRow {
                test_id: "infra".into(),
                status: TestStatus::Error,
                score: None,
                cached: false,
                message: "Request timeout".into(),
                details: serde_json::json!({}),
                duration_ms: None,
                fingerprint: None,
                skip_reason: None,
                attempts: None,
                error_policy_applied: None,
            },
            TestResultRow {
                test_id: "abstain".into(),
                status: TestStatus::Pass,
                score: Some(0.5),
                cached: false,
                message: String::new(),
                details: serde_json::json!({
                    "metrics": {
                        "faithfulness": {
                            "details": { "verdict": "Abstain", "score": 0.5 }
                        }
                    }
                }),
                duration_ms: None,
                fingerprint: None,
                skip_reason: None,
                attempts: None,
                error_policy_applied: None,
            },
        ];
        let outcome = decide_run_outcome(&results, false, ExitCodeVersion::V2);
        assert_eq!(
            outcome.exit_code, EXIT_INFRA_ERROR,
            "infra must beat abstain: expected exit 3"
        );
        assert!(
            outcome.reason_code == ReasonCode::ETimeout.as_str()
                || outcome.reason_code == ReasonCode::EJudgeUnavailable.as_str(),
            "reason should be infra (E_TIMEOUT or E_JUDGE_UNAVAILABLE), got {}",
            outcome.reason_code
        );
    }

    #[test]
    fn test_has_judge_verdict_abstain_detects_abstain() {
        let details = serde_json::json!({
            "metrics": {
                "faithfulness": {
                    "score": 0.5,
                    "passed": false,
                    "unstable": true,
                    "details": { "verdict": "Abstain", "score": 0.5 }
                }
            }
        });
        assert!(has_judge_verdict_abstain(&details));
    }

    #[test]
    fn test_has_judge_verdict_abstain_ignores_pass() {
        let details = serde_json::json!({
            "metrics": {
                "faithfulness": {
                    "score": 1.0,
                    "passed": true,
                    "unstable": false,
                    "details": { "verdict": "Pass", "score": 1.0 }
                }
            }
        });
        assert!(!has_judge_verdict_abstain(&details));
    }

    #[test]
    fn test_has_judge_verdict_abstain_no_metrics() {
        let details = serde_json::json!({});
        assert!(!has_judge_verdict_abstain(&details));
    }

    #[test]
    fn test_reason_code_from_error_message_maps_config_family() {
        assert_eq!(
            reason_code_from_error_message("trace not found: traces/missing.jsonl"),
            Some(ReasonCode::ETraceNotFound)
        );
        assert_eq!(
            reason_code_from_error_message("Config file not found: eval.yaml"),
            Some(ReasonCode::EMissingConfig)
        );
        assert_eq!(
            reason_code_from_error_message("config error: unknown field `foo`"),
            Some(ReasonCode::ECfgParse)
        );
    }

    #[test]
    fn test_reason_code_from_error_message_maps_infra_family() {
        assert_eq!(
            reason_code_from_error_message("provider returned 429 rate limit"),
            Some(ReasonCode::ERateLimit)
        );
        assert_eq!(
            reason_code_from_error_message("request timeout while calling judge"),
            Some(ReasonCode::ETimeout)
        );
        assert_eq!(
            reason_code_from_error_message("provider error: 503"),
            Some(ReasonCode::EProvider5xx)
        );
        assert_eq!(
            reason_code_from_error_message("network connection reset by peer"),
            Some(ReasonCode::ENetworkError)
        );
    }

    #[test]
    fn test_reason_code_from_run_error_uses_typed_kind() {
        let typed = RunError::missing_config("eval.yaml", "missing");
        assert_eq!(
            reason_code_from_run_error(&typed),
            Some(ReasonCode::EMissingConfig)
        );
        assert!(!typed.legacy_classified);
    }

    #[test]
    fn test_decide_outcome_uses_typed_details_before_legacy_message_fallback() {
        let row = TestResultRow {
            test_id: "typed".into(),
            status: TestStatus::Error,
            score: None,
            cached: false,
            message: "untyped error text".into(),
            details: serde_json::json!({
                "run_error_kind": "invalid_args"
            }),
            duration_ms: None,
            fingerprint: None,
            skip_reason: None,
            attempts: None,
            error_policy_applied: None,
        };

        let outcome = decide_run_outcome(&[row], true, ExitCodeVersion::V2);
        assert_eq!(outcome.reason_code, ReasonCode::EInvalidArgs.as_str());
        assert_eq!(
            outcome.exit_code,
            ReasonCode::EInvalidArgs.exit_code_for(ExitCodeVersion::V2)
        );
    }
}