icebox-gov 0.1.0

Runtime governance for autonomous offensive security — the single seam every human operator, REST client, and LLM agent must pass through before anything touches a target.
Documentation
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
606
607
608
609
610
611
use std::collections::HashMap;
use std::sync::Arc;

use serde::{Deserialize, Serialize};
use serde_json::Value;
use tokio::sync::{broadcast, Mutex};

use crate::core::governance::{ApprovalQueue, ApprovalRequest, ApprovalStatus, Role};
use crate::core::module::{Capability, Intent};
use crate::core::safety::{
    now_secs, Charter, ConfigPolicy, CvssScore, DecisionRecord, PolicyContext, PolicyDecision,
    PolicyEngine, PolicyRequest, PolicyRule, PolicySet, RiskLevel, ScopeManager,
};

#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct TaskSpec {
    pub name: String,
    pub target: String,
    pub capabilities: Vec<Capability>,
    pub impact: RiskLevel,
    pub destructive: bool,
    #[serde(default)]
    pub options: HashMap<String, String>,
    #[serde(default)]
    pub agent_id: Option<String>,
    #[serde(default)]
    pub context: PolicyContext,
    #[serde(default)]
    pub approved: bool,
    #[serde(default)]
    pub cvss: Option<CvssScore>,
}

impl Default for TaskSpec {
    fn default() -> Self {
        TaskSpec {
            name: String::new(),
            target: String::new(),
            capabilities: Vec::new(),
            impact: RiskLevel::Low,
            destructive: false,
            options: HashMap::new(),
            agent_id: None,
            context: PolicyContext::Autonomous,
            approved: false,
            cvss: None,
        }
    }
}

#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct GovernanceConfig {
    pub charter: Charter,
    pub scope: ScopeManager,
    #[serde(default)]
    pub max_risk: RiskLevel,
    #[serde(default)]
    pub role: Role,
    #[serde(default)]
    pub policy_set: PolicySet,
    /// Per-task-name rate ceilings, in invocations per minute.
    #[serde(default)]
    pub rate_limits: HashMap<String, u64>,
}

impl Default for GovernanceConfig {
    fn default() -> Self {
        GovernanceConfig {
            charter: Charter::default(),
            scope: ScopeManager::default(),
            max_risk: RiskLevel::Critical,
            role: Role::Operator,
            policy_set: PolicySet::default(),
            rate_limits: HashMap::new(),
        }
    }
}

#[derive(Debug, Clone, Serialize, Deserialize)]
pub enum GovernedOutcome {
    Allowed {
        result: Value,
        decision_id: u64,
    },
    Blocked {
        reason: String,
        decision_id: u64,
    },
    NeedsApproval {
        reason: String,
        decision_id: u64,
        approval_id: u64,
    },
}

#[derive(Debug, Default)]
struct RateState {
    count: u64,
    window_start: u64,
}

#[derive(Debug)]
struct RuntimeState {
    config: GovernanceConfig,
    decisions: Vec<DecisionRecord>,
    approvals: ApprovalQueue,
    rate: HashMap<String, RateState>,
    audit_tx: broadcast::Sender<DecisionRecord>,
    next_decision_id: u64,
}

impl RuntimeState {
    /// Enforces the per-name rate ceiling (fixed 60s window). Returns a denial
    /// reason and consumes a token when allowed.
    fn check_rate_limit(&mut self, name: &str) -> Option<String> {
        let limit = *self.config.rate_limits.get(name)?;
        let now = now_secs();
        let st = self.rate.entry(name.to_string()).or_default();
        if now.saturating_sub(st.window_start) >= 60 {
            st.count = 0;
            st.window_start = now;
        }
        if st.count >= limit {
            return Some(format!("rate limit of {limit}/min exceeded for {name}"));
        }
        st.count += 1;
        None
    }
}

fn derive_intents(caps: &[Capability]) -> Vec<Intent> {
    caps.iter().map(|c| c.intent()).collect()
}

/// A governed runtime backed by `Arc<Mutex<...>>`, safe to share across tasks.
#[derive(Debug, Clone)]
pub struct GovernanceRuntime {
    state: Arc<Mutex<RuntimeState>>,
}

impl GovernanceRuntime {
    pub fn builder() -> GovernanceBuilder {
        GovernanceBuilder::new()
    }

    /// Supervised run -- destructive / high-risk tasks and any `RequireApproval`
    /// rule come back as `NeedsApproval` and are queued rather than executed.
    pub async fn execute<F, Fut>(&self, task: TaskSpec, action: F) -> GovernedOutcome
    where
        F: FnOnce() -> Fut,
        Fut: std::future::Future<Output = Result<Value, String>> + Send + 'static,
    {
        self.enforce(task, action, false).await
    }

    /// Unsupervised run -- policy still hard-blocks denials, but
    /// approval-gated tasks are auto-granted and run.
    pub async fn run<F, Fut>(&self, task: TaskSpec, action: F) -> GovernedOutcome
    where
        F: FnOnce() -> Fut,
        Fut: std::future::Future<Output = Result<Value, String>> + Send + 'static,
    {
        self.enforce(task, action, true).await
    }

    async fn enforce<F, Fut>(
        &self,
        task: TaskSpec,
        action: F,
        auto_approve: bool,
    ) -> GovernedOutcome
    where
        F: FnOnce() -> Fut,
        Fut: std::future::Future<Output = Result<Value, String>> + Send + 'static,
    {
        let mut st = self.state.lock().await;

        if let Some(reason) = st.check_rate_limit(&task.name) {
            return st.record_and_block(task, reason);
        }

        let approved = auto_approve || task.approved;
        let req = PolicyRequest {
            target: task.target.clone(),
            capabilities: task.capabilities.clone(),
            impact: task.impact,
            destructive: task.destructive,
            charter_accepted: st.config.charter.accepted,
            in_scope: st.config.scope.is_in_scope(&task.target),
            approved,
            context: task.context,
            cvss: task.cvss.clone(),
        };
        let policy = ConfigPolicy {
            max_risk: st.config.policy_set.max_risk(st.config.max_risk),
            context: task.context,
            rules: st.config.policy_set.clone(),
        };
        let decision = policy.evaluate(&req);
        let decision_id = st.next_decision_id;
        st.next_decision_id += 1;
        let rec = DecisionRecord {
            at: now_secs(),
            target: task.target.clone(),
            module: task.name.clone(),
            capabilities: task.capabilities.clone(),
            intents: derive_intents(&task.capabilities),
            impact: task.impact,
            context: task.context,
            decision: decision.clone(),
        };
        st.decisions.push(rec.clone());
        let _ = st.audit_tx.send(rec);

        match decision {
            PolicyDecision::Deny(reason) => GovernedOutcome::Blocked {
                reason,
                decision_id,
            },
            PolicyDecision::RequireApproval(reason) if !approved => {
                let approval_id = st.approvals.request(
                    task.name.clone(),
                    task.target.clone(),
                    reason.clone(),
                    task.options.clone(),
                );
                GovernedOutcome::NeedsApproval {
                    reason,
                    decision_id,
                    approval_id,
                }
            }
            PolicyDecision::RequireApproval(_) | PolicyDecision::Allow => {
                drop(st);
                match action().await {
                    Ok(result) => GovernedOutcome::Allowed {
                        result,
                        decision_id,
                    },
                    Err(e) => GovernedOutcome::Blocked {
                        reason: format!("task action failed: {e}"),
                        decision_id,
                    },
                }
            }
        }
    }

    pub async fn approve(&self, id: u64) -> bool {
        self.state.lock().await.approvals.approve(id)
    }

    pub async fn deny(&self, id: u64) -> bool {
        self.state.lock().await.approvals.deny(id)
    }

    pub async fn pending_approvals(&self) -> Vec<ApprovalRequest> {
        self.state
            .lock()
            .await
            .approvals
            .list()
            .into_iter()
            .filter(|a| matches!(a.status, ApprovalStatus::Pending))
            .collect()
    }

    pub async fn audit(&self) -> Vec<DecisionRecord> {
        self.state.lock().await.decisions.clone()
    }

    pub async fn export_audit_json(&self) -> String {
        let d = self.state.lock().await.decisions.clone();
        serde_json::to_string_pretty(&d).unwrap_or_else(|_| "[]".into())
    }

    pub async fn export_audit_csv(&self) -> String {
        let d = self.state.lock().await.decisions.clone();
        crate::core::governance::audit_to_csv(&d)
    }

    /// Live audit stream -- each decision is delivered to every subscriber
    /// as it happens.
    pub async fn audit_stream(&self) -> broadcast::Receiver<DecisionRecord> {
        self.state.lock().await.audit_tx.subscribe()
    }

    pub async fn add_rule(&self, rule: PolicyRule) {
        self.state.lock().await.config.policy_set.add_rule(rule);
    }

    pub async fn policy_set(&self) -> PolicySet {
        self.state.lock().await.config.policy_set.clone()
    }

    pub async fn role(&self) -> Role {
        self.state.lock().await.config.role
    }

    pub async fn config(&self) -> GovernanceConfig {
        self.state.lock().await.config.clone()
    }
}

impl RuntimeState {
    fn record_and_block(&mut self, task: TaskSpec, reason: String) -> GovernedOutcome {
        let decision_id = self.next_decision_id;
        self.next_decision_id += 1;
        let rec = DecisionRecord {
            at: now_secs(),
            target: task.target.clone(),
            module: task.name.clone(),
            capabilities: task.capabilities.clone(),
            intents: derive_intents(&task.capabilities),
            impact: task.impact,
            context: task.context,
            decision: PolicyDecision::Deny(reason.clone()),
        };
        self.decisions.push(rec.clone());
        let _ = self.audit_tx.send(rec);
        GovernedOutcome::Blocked {
            reason,
            decision_id,
        }
    }
}

pub fn govern(config: GovernanceConfig) -> GovernanceRuntime {
    let (audit_tx, _rx) = broadcast::channel(1024);
    GovernanceRuntime {
        state: Arc::new(Mutex::new(RuntimeState {
            config,
            decisions: Vec::new(),
            approvals: ApprovalQueue::default(),
            rate: HashMap::new(),
            audit_tx,
            next_decision_id: 1,
        })),
    }
}

pub struct GovernanceBuilder {
    config: GovernanceConfig,
}

impl GovernanceBuilder {
    pub fn new() -> Self {
        GovernanceBuilder {
            config: GovernanceConfig::default(),
        }
    }

    pub fn charter(mut self, charter: Charter) -> Self {
        self.config.charter = charter;
        self
    }

    pub fn scope(mut self, allow: Vec<String>) -> Self {
        self.config.scope = ScopeManager::new(allow);
        self
    }

    pub fn max_risk(mut self, max_risk: RiskLevel) -> Self {
        self.config.max_risk = max_risk;
        self
    }

    pub fn role(mut self, role: Role) -> Self {
        self.config.role = role;
        self
    }

    pub fn deny_capability(mut self, cap: Capability) -> Self {
        self.config
            .policy_set
            .add_rule(PolicyRule::DenyCapability(cap));
        self
    }

    pub fn allow_capability(mut self, cap: Capability) -> Self {
        self.config
            .policy_set
            .add_rule(PolicyRule::AllowCapability(cap));
        self
    }

    pub fn require_approval(mut self, cap: Capability, target_pattern: &str) -> Self {
        self.config
            .policy_set
            .add_rule(PolicyRule::RequireApproval {
                capability: cap,
                target_pattern: target_pattern.to_string(),
            });
        self
    }

    pub fn max_risk_ceiling(mut self, max_risk: RiskLevel) -> Self {
        self.config
            .policy_set
            .add_rule(PolicyRule::MaxRisk(max_risk));
        self
    }

    pub fn deny_if_cvss_above(mut self, threshold: f64) -> Self {
        self.config
            .policy_set
            .add_rule(PolicyRule::DenyIfCvssAbove(threshold));
        self
    }

    pub fn require_approval_if(
        mut self,
        cvss_above: Option<f64>,
        epss_above: Option<f64>,
        kev: bool,
    ) -> Self {
        self.config
            .policy_set
            .add_rule(PolicyRule::RequireApprovalIf {
                cvss_above,
                epss_above,
                kev,
            });
        self
    }

    pub fn rate_limit(mut self, name: &str, per_minute: u64) -> Self {
        self.config.rate_limits.insert(name.to_string(), per_minute);
        self
    }

    pub fn build(self) -> GovernanceRuntime {
        govern(self.config)
    }
}

impl Default for GovernanceBuilder {
    fn default() -> Self {
        Self::new()
    }
}

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

    fn base_runtime() -> GovernanceRuntime {
        GovernanceRuntime::builder()
            .charter(Charter::accept("eng", vec!["no destruction".into()]))
            .scope(vec!["10.0.0.0/8".into()])
            .max_risk(RiskLevel::Critical)
            .role(Role::Admin)
            .build()
    }

    fn low_risk_task(target: &str) -> TaskSpec {
        TaskSpec {
            name: "scan".into(),
            target: target.into(),
            capabilities: vec![Capability::NetworkScan],
            impact: RiskLevel::Low,
            ..Default::default()
        }
    }

    #[tokio::test]
    async fn supervised_low_risk_runs_and_audits() {
        let rt = base_runtime();
        let out = rt
            .execute(low_risk_task("10.0.0.5"), || async {
                Ok(json!({"open_ports": [22, 80]}))
            })
            .await;
        match out {
            GovernedOutcome::Allowed { result, .. } => {
                assert_eq!(result["open_ports"][0], 22);
            }
            other => panic!("expected Allowed, got {other:?}"),
        }
        assert_eq!(rt.audit().await.len(), 1);
    }

    #[tokio::test]
    async fn out_of_scope_is_blocked() {
        let rt = base_runtime();
        let out = rt
            .execute(low_risk_task("8.8.8.8"), || async { Ok(json!(null)) })
            .await;
        assert!(matches!(out, GovernedOutcome::Blocked { .. }));
    }

    #[tokio::test]
    async fn deny_capability_blocks() {
        let rt = GovernanceRuntime::builder()
            .charter(Charter::accept("eng", vec![]))
            .scope(vec!["10.0.0.0/8".into()])
            .deny_capability(Capability::Persistence)
            .build();
        let task = TaskSpec {
            name: "implant".into(),
            target: "10.0.0.9".into(),
            capabilities: vec![Capability::Persistence],
            impact: RiskLevel::High,
            ..Default::default()
        };
        let out = rt.execute(task, || async { Ok(json!(null)) }).await;
        assert!(matches!(out, GovernedOutcome::Blocked { .. }));
    }

    #[tokio::test]
    async fn destructive_requires_approval_then_runs_after_approve() {
        let rt = base_runtime();
        let task = TaskSpec {
            name: "wipe".into(),
            target: "10.0.0.7".into(),
            capabilities: vec![Capability::FilesystemModification],
            impact: RiskLevel::High,
            destructive: true,
            ..Default::default()
        };
        let out = rt
            .execute(task.clone(), || async { Ok(json!({"wiped": true})) })
            .await;
        let id = match out {
            GovernedOutcome::NeedsApproval { approval_id, .. } => approval_id,
            other => panic!("expected NeedsApproval, got {other:?}"),
        };
        assert!(rt.approve(id).await);
        assert!(rt.pending_approvals().await.is_empty());
    }

    #[tokio::test]
    async fn require_approval_rule_gates_specific_target() {
        let rt = GovernanceRuntime::builder()
            .charter(Charter::accept("eng", vec![]))
            .scope(vec!["0.0.0.0/0".into()])
            .require_approval(Capability::CredentialAccess, "192.168.*")
            .build();
        let in_scope = TaskSpec {
            name: "dump".into(),
            target: "192.168.1.5".into(),
            capabilities: vec![Capability::CredentialAccess],
            impact: RiskLevel::Low,
            ..Default::default()
        };
        let out = rt.execute(in_scope, || async { Ok(json!(null)) }).await;
        assert!(matches!(out, GovernedOutcome::NeedsApproval { .. }));

        let other = TaskSpec {
            name: "dump".into(),
            target: "10.0.0.5".into(),
            capabilities: vec![Capability::CredentialAccess],
            impact: RiskLevel::Low,
            ..Default::default()
        };
        let out = rt.execute(other, || async { Ok(json!(null)) }).await;
        assert!(matches!(out, GovernedOutcome::Allowed { .. }));
    }

    #[tokio::test]
    async fn unsupervised_run_auto_grants_approval() {
        let rt = base_runtime();
        let task = TaskSpec {
            name: "wipe".into(),
            target: "10.0.0.7".into(),
            capabilities: vec![Capability::FilesystemModification],
            impact: RiskLevel::High,
            destructive: true,
            ..Default::default()
        };
        let out = rt.run(task, || async { Ok(json!({"ok": true})) }).await;
        assert!(matches!(out, GovernedOutcome::Allowed { .. }));
    }

    #[tokio::test]
    async fn rate_limit_blocks_after_ceiling() {
        let rt = GovernanceRuntime::builder()
            .charter(Charter::accept("eng", vec![]))
            .scope(vec!["0.0.0.0/0".into()])
            .rate_limit("scan", 2)
            .build();
        let task = low_risk_task("10.0.0.5");
        for _ in 0..2 {
            let out = rt.execute(task.clone(), || async { Ok(json!(null)) }).await;
            assert!(matches!(out, GovernedOutcome::Allowed { .. }));
        }
        let out = rt.execute(task.clone(), || async { Ok(json!(null)) }).await;
        assert!(matches!(out, GovernedOutcome::Blocked { .. }));
    }

    #[tokio::test]
    async fn audit_stream_delivers_records() {
        let rt = base_runtime();
        let mut rx = rt.audit_stream().await;
        let _ = rt
            .execute(low_risk_task("10.0.0.5"), || async { Ok(json!(null)) })
            .await;
        let rec = rx.recv().await.expect("audit record");
        assert_eq!(rec.target, "10.0.0.5");
    }

    #[tokio::test]
    async fn config_round_trips_through_json() {
        let rt = base_runtime();
        let json = serde_json::to_string(&rt.config().await).unwrap();
        let cfg: GovernanceConfig = serde_json::from_str(&json).unwrap();
        assert!(cfg.charter.accepted);
        assert_eq!(cfg.role, Role::Admin);
    }
}