zeph_core/
metrics.rs

1// SPDX-FileCopyrightText: 2026 Andrei G <bug-ops>
2// SPDX-License-Identifier: MIT OR Apache-2.0
3
4use std::collections::VecDeque;
5
6use tokio::sync::watch;
7
8pub use zeph_llm::{ClassifierMetricsSnapshot, TaskMetricsSnapshot};
9pub use zeph_memory::{CategoryScore, ProbeCategory, ProbeVerdict};
10
11/// Category of a security event for TUI display.
12#[derive(Debug, Clone, Copy, PartialEq, Eq)]
13pub enum SecurityEventCategory {
14    InjectionFlag,
15    /// ML classifier hard-blocked tool output (`enforcement_mode=block` only).
16    InjectionBlocked,
17    ExfiltrationBlock,
18    Quarantine,
19    Truncation,
20    RateLimit,
21    MemoryValidation,
22    PreExecutionBlock,
23    PreExecutionWarn,
24    ResponseVerification,
25    /// `TurnCausalAnalyzer` flagged behavioral deviation at tool-return boundary.
26    CausalIpiFlag,
27    /// MCP tool result crossing into an ACP-serving session boundary.
28    CrossBoundaryMcpToAcp,
29}
30
31impl SecurityEventCategory {
32    #[must_use]
33    pub fn as_str(self) -> &'static str {
34        match self {
35            Self::InjectionFlag => "injection",
36            Self::InjectionBlocked => "injection_blocked",
37            Self::ExfiltrationBlock => "exfil",
38            Self::Quarantine => "quarantine",
39            Self::Truncation => "truncation",
40            Self::RateLimit => "rate_limit",
41            Self::MemoryValidation => "memory_validation",
42            Self::PreExecutionBlock => "pre_exec_block",
43            Self::PreExecutionWarn => "pre_exec_warn",
44            Self::ResponseVerification => "response_verify",
45            Self::CausalIpiFlag => "causal_ipi",
46            Self::CrossBoundaryMcpToAcp => "cross_boundary_mcp_to_acp",
47        }
48    }
49}
50
51/// A single security event record for TUI display.
52#[derive(Debug, Clone)]
53pub struct SecurityEvent {
54    /// Unix timestamp (seconds since epoch).
55    pub timestamp: u64,
56    pub category: SecurityEventCategory,
57    /// Source that triggered the event (e.g., `web_scrape`, `mcp_response`).
58    pub source: String,
59    /// Short description, capped at 128 chars.
60    pub detail: String,
61}
62
63impl SecurityEvent {
64    #[must_use]
65    pub fn new(
66        category: SecurityEventCategory,
67        source: impl Into<String>,
68        detail: impl Into<String>,
69    ) -> Self {
70        // IMP-1: cap source at 64 chars and strip ASCII control chars.
71        let source: String = source
72            .into()
73            .chars()
74            .filter(|c| !c.is_ascii_control())
75            .take(64)
76            .collect();
77        // CR-1: UTF-8 safe truncation using floor_char_boundary (stable since Rust 1.82).
78        let detail = detail.into();
79        let detail = if detail.len() > 128 {
80            let end = detail.floor_char_boundary(127);
81            format!("{}…", &detail[..end])
82        } else {
83            detail
84        };
85        Self {
86            timestamp: std::time::SystemTime::now()
87                .duration_since(std::time::UNIX_EPOCH)
88                .unwrap_or_default()
89                .as_secs(),
90            category,
91            source,
92            detail,
93        }
94    }
95}
96
97/// Ring buffer capacity for security events.
98pub const SECURITY_EVENT_CAP: usize = 100;
99
100/// Lightweight snapshot of a single task row for TUI display.
101///
102/// Cloned from [`TaskGraph`] on each metrics tick; kept minimal on purpose.
103#[derive(Debug, Clone)]
104pub struct TaskSnapshotRow {
105    pub id: u32,
106    pub title: String,
107    /// Stringified `TaskStatus` (e.g. `"pending"`, `"running"`, `"completed"`).
108    pub status: String,
109    pub agent: Option<String>,
110    pub duration_ms: u64,
111    /// Truncated error message (first 80 chars) when the task failed.
112    pub error: Option<String>,
113}
114
115/// Lightweight snapshot of a `TaskGraph` for TUI display.
116#[derive(Debug, Clone, Default)]
117pub struct TaskGraphSnapshot {
118    pub graph_id: String,
119    pub goal: String,
120    /// Stringified `GraphStatus` (e.g. `"created"`, `"running"`, `"completed"`).
121    pub status: String,
122    pub tasks: Vec<TaskSnapshotRow>,
123    pub completed_at: Option<std::time::Instant>,
124}
125
126impl TaskGraphSnapshot {
127    /// Returns `true` if this snapshot represents a terminal plan that finished
128    /// more than 30 seconds ago and should no longer be shown in the TUI.
129    #[must_use]
130    pub fn is_stale(&self) -> bool {
131        self.completed_at
132            .is_some_and(|t| t.elapsed().as_secs() > 30)
133    }
134}
135
136/// Counters for the task orchestration subsystem.
137///
138/// Always present in [`MetricsSnapshot`]; zero-valued when orchestration is inactive.
139#[derive(Debug, Clone, Default)]
140pub struct OrchestrationMetrics {
141    pub plans_total: u64,
142    pub tasks_total: u64,
143    pub tasks_completed: u64,
144    pub tasks_failed: u64,
145    pub tasks_skipped: u64,
146}
147
148/// Connection status of a single MCP server for TUI display.
149#[derive(Debug, Clone, PartialEq, Eq)]
150pub enum McpServerConnectionStatus {
151    Connected,
152    Failed,
153}
154
155/// Per-server MCP status snapshot for TUI display.
156#[derive(Debug, Clone)]
157pub struct McpServerStatus {
158    pub id: String,
159    pub status: McpServerConnectionStatus,
160    /// Number of tools provided by this server (0 when failed).
161    pub tool_count: usize,
162    /// Human-readable failure reason. Empty when connected.
163    pub error: String,
164}
165
166/// Bayesian confidence data for a single skill, used by TUI confidence bar.
167#[derive(Debug, Clone, Default)]
168pub struct SkillConfidence {
169    pub name: String,
170    pub posterior: f64,
171    pub total_uses: u32,
172}
173
174/// Snapshot of a single sub-agent's runtime status.
175#[derive(Debug, Clone, Default)]
176pub struct SubAgentMetrics {
177    pub id: String,
178    pub name: String,
179    /// Stringified `TaskState`: "working", "completed", "failed", "canceled", etc.
180    pub state: String,
181    pub turns_used: u32,
182    pub max_turns: u32,
183    pub background: bool,
184    pub elapsed_secs: u64,
185    /// Stringified `PermissionMode`: `"default"`, `"accept_edits"`, `"dont_ask"`,
186    /// `"bypass_permissions"`, `"plan"`. Empty string when mode is `Default`.
187    pub permission_mode: String,
188    /// Path to the directory containing this agent's JSONL transcript file.
189    /// `None` when transcript writing is disabled for this agent.
190    pub transcript_dir: Option<String>,
191}
192
193#[derive(Debug, Clone, Default)]
194#[allow(clippy::struct_excessive_bools)]
195pub struct MetricsSnapshot {
196    pub prompt_tokens: u64,
197    pub completion_tokens: u64,
198    pub total_tokens: u64,
199    pub context_tokens: u64,
200    pub api_calls: u64,
201    pub active_skills: Vec<String>,
202    pub total_skills: usize,
203    /// Total configured MCP servers (connected + failed).
204    pub mcp_server_count: usize,
205    pub mcp_tool_count: usize,
206    /// Number of successfully connected MCP servers.
207    pub mcp_connected_count: usize,
208    /// Per-server connection status list.
209    pub mcp_servers: Vec<McpServerStatus>,
210    pub active_mcp_tools: Vec<String>,
211    pub sqlite_message_count: u64,
212    pub sqlite_conversation_id: Option<zeph_memory::ConversationId>,
213    pub qdrant_available: bool,
214    pub vector_backend: String,
215    pub embeddings_generated: u64,
216    pub last_llm_latency_ms: u64,
217    pub uptime_seconds: u64,
218    pub provider_name: String,
219    pub model_name: String,
220    pub summaries_count: u64,
221    pub context_compactions: u64,
222    /// Number of times the agent entered the Hard compaction tier, including cooldown-skipped
223    /// turns. Not equal to the actual LLM summarization count — reflects pressure, not action.
224    pub compaction_hard_count: u64,
225    /// User-message turns elapsed after each hard compaction event.
226    /// Entry i = turns between hard compaction i and hard compaction i+1 (or session end).
227    /// Empty when no hard compaction occurred during the session.
228    pub compaction_turns_after_hard: Vec<u64>,
229    pub compression_events: u64,
230    pub compression_tokens_saved: u64,
231    pub tool_output_prunes: u64,
232    /// Compaction probe outcomes (#1609).
233    pub compaction_probe_passes: u64,
234    /// Compaction probe soft failures (summary borderline — compaction proceeded with warning).
235    pub compaction_probe_soft_failures: u64,
236    /// Compaction probe hard failures (compaction blocked due to lossy summary).
237    pub compaction_probe_failures: u64,
238    /// Compaction probe errors (LLM/timeout — non-blocking, compaction proceeded).
239    pub compaction_probe_errors: u64,
240    /// Last compaction probe verdict. `None` before the first probe completes.
241    pub last_probe_verdict: Option<zeph_memory::ProbeVerdict>,
242    /// Last compaction probe score in [0.0, 1.0]. `None` before the first probe
243    /// completes or after an Error verdict (errors produce no score).
244    pub last_probe_score: Option<f32>,
245    /// Per-category scores from the last completed probe.
246    pub last_probe_category_scores: Option<Vec<zeph_memory::CategoryScore>>,
247    /// Configured pass threshold for the compaction probe. Used by TUI for category color-coding.
248    pub compaction_probe_threshold: f32,
249    /// Configured hard-fail threshold for the compaction probe.
250    pub compaction_probe_hard_fail_threshold: f32,
251    pub cache_read_tokens: u64,
252    pub cache_creation_tokens: u64,
253    pub cost_spent_cents: f64,
254    pub filter_raw_tokens: u64,
255    pub filter_saved_tokens: u64,
256    pub filter_applications: u64,
257    pub filter_total_commands: u64,
258    pub filter_filtered_commands: u64,
259    pub filter_confidence_full: u64,
260    pub filter_confidence_partial: u64,
261    pub filter_confidence_fallback: u64,
262    pub cancellations: u64,
263    pub server_compaction_events: u64,
264    pub sanitizer_runs: u64,
265    pub sanitizer_injection_flags: u64,
266    /// Injection pattern hits on `ToolResult` (local) sources — likely false positives.
267    ///
268    /// Counts regex hits that fired on content from `shell`, `read_file`, `search_code`, etc.
269    /// These sources are user-owned and not adversarial; a non-zero value indicates a pattern
270    /// that needs tightening or a source reclassification.
271    pub sanitizer_injection_fp_local: u64,
272    pub sanitizer_truncations: u64,
273    pub quarantine_invocations: u64,
274    pub quarantine_failures: u64,
275    /// ML classifier hard-blocked tool outputs (`enforcement_mode=block` only).
276    pub classifier_tool_blocks: u64,
277    /// ML classifier suspicious tool outputs (both enforcement modes).
278    pub classifier_tool_suspicious: u64,
279    /// `TurnCausalAnalyzer` flags: behavioral deviation detected at tool-return boundary.
280    pub causal_ipi_flags: u64,
281    pub exfiltration_images_blocked: u64,
282    pub exfiltration_tool_urls_flagged: u64,
283    pub exfiltration_memory_guards: u64,
284    pub pii_scrub_count: u64,
285    /// Number of times the PII NER classifier timed out; input fell back to regex-only.
286    pub pii_ner_timeouts: u64,
287    pub memory_validation_failures: u64,
288    pub rate_limit_trips: u64,
289    pub pre_execution_blocks: u64,
290    pub pre_execution_warnings: u64,
291    /// `true` when a guardrail filter is active for this session.
292    #[cfg(feature = "guardrail")]
293    pub guardrail_enabled: bool,
294    /// `true` when guardrail is in warn-only mode (action = warn).
295    #[cfg(feature = "guardrail")]
296    pub guardrail_warn_mode: bool,
297    pub sub_agents: Vec<SubAgentMetrics>,
298    pub skill_confidence: Vec<SkillConfidence>,
299    /// Scheduled task summaries: `[name, kind, mode, next_run]`.
300    pub scheduled_tasks: Vec<[String; 4]>,
301    /// Thompson Sampling distribution snapshots: `(provider, alpha, beta)`.
302    pub router_thompson_stats: Vec<(String, f64, f64)>,
303    /// Ring buffer of recent security events (cap 100, FIFO eviction).
304    pub security_events: VecDeque<SecurityEvent>,
305    pub orchestration: OrchestrationMetrics,
306    /// Live snapshot of the currently active task graph. `None` when no plan is active.
307    pub orchestration_graph: Option<TaskGraphSnapshot>,
308    pub graph_community_detection_failures: u64,
309    pub graph_entities_total: u64,
310    pub graph_edges_total: u64,
311    pub graph_communities_total: u64,
312    pub graph_extraction_count: u64,
313    pub graph_extraction_failures: u64,
314    /// `true` when `config.llm.cloud.enable_extended_context = true`.
315    /// Never set for other providers to avoid false positives.
316    pub extended_context: bool,
317    /// Latest compression-guidelines version (0 = no guidelines yet).
318    pub guidelines_version: u32,
319    /// ISO 8601 timestamp of the latest guidelines update (empty if none).
320    pub guidelines_updated_at: String,
321    pub tool_cache_hits: u64,
322    pub tool_cache_misses: u64,
323    pub tool_cache_entries: usize,
324    /// Number of semantic-tier facts in memory (0 when tier promotion disabled).
325    pub semantic_fact_count: u64,
326    /// STT model name (e.g. "whisper-1"). `None` when STT is not configured.
327    pub stt_model: Option<String>,
328    /// Model used for context compaction/summarization. `None` when no summary provider is set.
329    pub compaction_model: Option<String>,
330    /// Temperature of the active provider when using Candle. `None` for API providers.
331    pub provider_temperature: Option<f32>,
332    /// Top-p of the active provider when using Candle. `None` for API providers.
333    pub provider_top_p: Option<f32>,
334    /// Embedding model name (e.g. `"nomic-embed-text"`). Empty when embeddings are disabled.
335    pub embedding_model: String,
336    /// Token budget for context window. `None` when not configured.
337    pub token_budget: Option<u64>,
338    /// Token threshold that triggers soft compaction. `None` when not configured.
339    pub compaction_threshold: Option<u32>,
340    /// Vault backend identifier: "age", "env", or "none".
341    pub vault_backend: String,
342    /// Active I/O channel name: `"cli"`, `"telegram"`, `"tui"`, `"discord"`, `"slack"`.
343    pub active_channel: String,
344    /// Whether self-learning (skill evolution) is enabled.
345    pub self_learning_enabled: bool,
346    /// Whether the semantic response cache is enabled.
347    pub semantic_cache_enabled: bool,
348    /// Whether semantic response caching is enabled (alias for `semantic_cache_enabled`).
349    pub cache_enabled: bool,
350    /// Whether assistant messages are auto-saved to memory.
351    pub autosave_enabled: bool,
352    /// Classifier p50/p95 latency metrics per task (injection, pii, feedback).
353    pub classifier: ClassifierMetricsSnapshot,
354}
355
356/// Strip ASCII control characters and ANSI escape sequences from a string for safe TUI display.
357///
358/// Allows tab, LF, and CR; removes everything else in the `0x00–0x1F` range including full
359/// ANSI CSI sequences (`ESC[...`). This prevents escape-sequence injection from LLM planner
360/// output into the TUI.
361fn strip_ctrl(s: &str) -> String {
362    let mut out = String::with_capacity(s.len());
363    let mut chars = s.chars().peekable();
364    while let Some(c) = chars.next() {
365        if c == '\x1b' {
366            // Consume an ANSI CSI sequence: ESC [ <params> <final-byte in 0x40–0x7E>
367            if chars.peek() == Some(&'[') {
368                chars.next(); // consume '['
369                for inner in chars.by_ref() {
370                    if ('\x40'..='\x7e').contains(&inner) {
371                        break;
372                    }
373                }
374            }
375            // Drop ESC and any consumed sequence — write nothing.
376        } else if c.is_control() && c != '\t' && c != '\n' && c != '\r' {
377            // drop other control chars
378        } else {
379            out.push(c);
380        }
381    }
382    out
383}
384
385/// Convert a live `TaskGraph` into a lightweight snapshot for TUI display.
386impl From<&crate::orchestration::TaskGraph> for TaskGraphSnapshot {
387    fn from(graph: &crate::orchestration::TaskGraph) -> Self {
388        let tasks = graph
389            .tasks
390            .iter()
391            .map(|t| {
392                let error = t
393                    .result
394                    .as_ref()
395                    .filter(|_| t.status == crate::orchestration::TaskStatus::Failed)
396                    .and_then(|r| {
397                        if r.output.is_empty() {
398                            None
399                        } else {
400                            // Strip control chars, then truncate at 80 chars (SEC-P6-01).
401                            let s = strip_ctrl(&r.output);
402                            if s.len() > 80 {
403                                let end = s.floor_char_boundary(79);
404                                Some(format!("{}…", &s[..end]))
405                            } else {
406                                Some(s)
407                            }
408                        }
409                    });
410                let duration_ms = t.result.as_ref().map_or(0, |r| r.duration_ms);
411                TaskSnapshotRow {
412                    id: t.id.as_u32(),
413                    title: strip_ctrl(&t.title),
414                    status: t.status.to_string(),
415                    agent: t.assigned_agent.as_deref().map(strip_ctrl),
416                    duration_ms,
417                    error,
418                }
419            })
420            .collect();
421        Self {
422            graph_id: graph.id.to_string(),
423            goal: strip_ctrl(&graph.goal),
424            status: graph.status.to_string(),
425            tasks,
426            completed_at: None,
427        }
428    }
429}
430
431pub struct MetricsCollector {
432    tx: watch::Sender<MetricsSnapshot>,
433}
434
435impl MetricsCollector {
436    #[must_use]
437    pub fn new() -> (Self, watch::Receiver<MetricsSnapshot>) {
438        let (tx, rx) = watch::channel(MetricsSnapshot::default());
439        (Self { tx }, rx)
440    }
441
442    pub fn update(&self, f: impl FnOnce(&mut MetricsSnapshot)) {
443        self.tx.send_modify(f);
444    }
445}
446
447#[cfg(test)]
448mod tests {
449    #![allow(clippy::field_reassign_with_default)]
450
451    use super::*;
452
453    #[test]
454    fn default_metrics_snapshot() {
455        let m = MetricsSnapshot::default();
456        assert_eq!(m.total_tokens, 0);
457        assert_eq!(m.api_calls, 0);
458        assert!(m.active_skills.is_empty());
459        assert!(m.active_mcp_tools.is_empty());
460        assert_eq!(m.mcp_tool_count, 0);
461        assert_eq!(m.mcp_server_count, 0);
462        assert!(m.provider_name.is_empty());
463        assert_eq!(m.summaries_count, 0);
464        // Phase 2 fields
465        assert!(m.stt_model.is_none());
466        assert!(m.compaction_model.is_none());
467        assert!(m.provider_temperature.is_none());
468        assert!(m.provider_top_p.is_none());
469        assert!(m.active_channel.is_empty());
470        assert!(m.embedding_model.is_empty());
471        assert!(m.token_budget.is_none());
472        assert!(!m.self_learning_enabled);
473        assert!(!m.semantic_cache_enabled);
474    }
475
476    #[test]
477    fn metrics_collector_update_phase2_fields() {
478        let (collector, rx) = MetricsCollector::new();
479        collector.update(|m| {
480            m.stt_model = Some("whisper-1".into());
481            m.compaction_model = Some("haiku".into());
482            m.provider_temperature = Some(0.7);
483            m.provider_top_p = Some(0.95);
484            m.active_channel = "tui".into();
485            m.embedding_model = "nomic-embed-text".into();
486            m.token_budget = Some(200_000);
487            m.self_learning_enabled = true;
488            m.semantic_cache_enabled = true;
489        });
490        let s = rx.borrow();
491        assert_eq!(s.stt_model.as_deref(), Some("whisper-1"));
492        assert_eq!(s.compaction_model.as_deref(), Some("haiku"));
493        assert_eq!(s.provider_temperature, Some(0.7));
494        assert_eq!(s.provider_top_p, Some(0.95));
495        assert_eq!(s.active_channel, "tui");
496        assert_eq!(s.embedding_model, "nomic-embed-text");
497        assert_eq!(s.token_budget, Some(200_000));
498        assert!(s.self_learning_enabled);
499        assert!(s.semantic_cache_enabled);
500    }
501
502    #[test]
503    fn metrics_collector_update() {
504        let (collector, rx) = MetricsCollector::new();
505        collector.update(|m| {
506            m.api_calls = 5;
507            m.total_tokens = 1000;
508        });
509        let snapshot = rx.borrow().clone();
510        assert_eq!(snapshot.api_calls, 5);
511        assert_eq!(snapshot.total_tokens, 1000);
512    }
513
514    #[test]
515    fn metrics_collector_multiple_updates() {
516        let (collector, rx) = MetricsCollector::new();
517        collector.update(|m| m.api_calls = 1);
518        collector.update(|m| m.api_calls += 1);
519        assert_eq!(rx.borrow().api_calls, 2);
520    }
521
522    #[test]
523    fn metrics_snapshot_clone() {
524        let mut m = MetricsSnapshot::default();
525        m.provider_name = "ollama".into();
526        let cloned = m.clone();
527        assert_eq!(cloned.provider_name, "ollama");
528    }
529
530    #[test]
531    fn filter_metrics_tracking() {
532        let (collector, rx) = MetricsCollector::new();
533        collector.update(|m| {
534            m.filter_raw_tokens += 250;
535            m.filter_saved_tokens += 200;
536            m.filter_applications += 1;
537        });
538        collector.update(|m| {
539            m.filter_raw_tokens += 100;
540            m.filter_saved_tokens += 80;
541            m.filter_applications += 1;
542        });
543        let s = rx.borrow();
544        assert_eq!(s.filter_raw_tokens, 350);
545        assert_eq!(s.filter_saved_tokens, 280);
546        assert_eq!(s.filter_applications, 2);
547    }
548
549    #[test]
550    fn filter_confidence_and_command_metrics() {
551        let (collector, rx) = MetricsCollector::new();
552        collector.update(|m| {
553            m.filter_total_commands += 1;
554            m.filter_filtered_commands += 1;
555            m.filter_confidence_full += 1;
556        });
557        collector.update(|m| {
558            m.filter_total_commands += 1;
559            m.filter_confidence_partial += 1;
560        });
561        let s = rx.borrow();
562        assert_eq!(s.filter_total_commands, 2);
563        assert_eq!(s.filter_filtered_commands, 1);
564        assert_eq!(s.filter_confidence_full, 1);
565        assert_eq!(s.filter_confidence_partial, 1);
566        assert_eq!(s.filter_confidence_fallback, 0);
567    }
568
569    #[test]
570    fn summaries_count_tracks_summarizations() {
571        let (collector, rx) = MetricsCollector::new();
572        collector.update(|m| m.summaries_count += 1);
573        collector.update(|m| m.summaries_count += 1);
574        assert_eq!(rx.borrow().summaries_count, 2);
575    }
576
577    #[test]
578    fn cancellations_counter_increments() {
579        let (collector, rx) = MetricsCollector::new();
580        assert_eq!(rx.borrow().cancellations, 0);
581        collector.update(|m| m.cancellations += 1);
582        collector.update(|m| m.cancellations += 1);
583        assert_eq!(rx.borrow().cancellations, 2);
584    }
585
586    #[test]
587    fn security_event_detail_exact_128_not_truncated() {
588        let s = "a".repeat(128);
589        let ev = SecurityEvent::new(SecurityEventCategory::InjectionFlag, "src", s.clone());
590        assert_eq!(ev.detail, s, "128-char string must not be truncated");
591    }
592
593    #[test]
594    fn security_event_detail_129_is_truncated() {
595        let s = "a".repeat(129);
596        let ev = SecurityEvent::new(SecurityEventCategory::InjectionFlag, "src", s);
597        assert!(
598            ev.detail.ends_with('…'),
599            "129-char string must end with ellipsis"
600        );
601        assert!(
602            ev.detail.len() <= 130,
603            "truncated detail must be at most 130 bytes"
604        );
605    }
606
607    #[test]
608    fn security_event_detail_multibyte_utf8_no_panic() {
609        // Each '中' is 3 bytes. 43 chars = 129 bytes — triggers truncation at a multi-byte boundary.
610        let s = "中".repeat(43);
611        let ev = SecurityEvent::new(SecurityEventCategory::InjectionFlag, "src", s);
612        assert!(ev.detail.ends_with('…'));
613    }
614
615    #[test]
616    fn security_event_source_capped_at_64_chars() {
617        let long_source = "x".repeat(200);
618        let ev = SecurityEvent::new(SecurityEventCategory::InjectionFlag, long_source, "detail");
619        assert_eq!(ev.source.len(), 64);
620    }
621
622    #[test]
623    fn security_event_source_strips_control_chars() {
624        let source = "tool\x00name\x1b[31m";
625        let ev = SecurityEvent::new(SecurityEventCategory::InjectionFlag, source, "detail");
626        assert!(!ev.source.contains('\x00'));
627        assert!(!ev.source.contains('\x1b'));
628    }
629
630    #[test]
631    fn security_event_category_as_str() {
632        assert_eq!(SecurityEventCategory::InjectionFlag.as_str(), "injection");
633        assert_eq!(SecurityEventCategory::ExfiltrationBlock.as_str(), "exfil");
634        assert_eq!(SecurityEventCategory::Quarantine.as_str(), "quarantine");
635        assert_eq!(SecurityEventCategory::Truncation.as_str(), "truncation");
636        assert_eq!(
637            SecurityEventCategory::CrossBoundaryMcpToAcp.as_str(),
638            "cross_boundary_mcp_to_acp"
639        );
640    }
641
642    #[test]
643    fn ring_buffer_respects_cap_via_update() {
644        let (collector, rx) = MetricsCollector::new();
645        for i in 0..110u64 {
646            let event = SecurityEvent::new(
647                SecurityEventCategory::InjectionFlag,
648                "src",
649                format!("event {i}"),
650            );
651            collector.update(|m| {
652                if m.security_events.len() >= SECURITY_EVENT_CAP {
653                    m.security_events.pop_front();
654                }
655                m.security_events.push_back(event);
656            });
657        }
658        let snap = rx.borrow();
659        assert_eq!(snap.security_events.len(), SECURITY_EVENT_CAP);
660        // FIFO: earliest events evicted, last one present
661        assert!(snap.security_events.back().unwrap().detail.contains("109"));
662    }
663
664    #[test]
665    fn security_events_empty_by_default() {
666        let m = MetricsSnapshot::default();
667        assert!(m.security_events.is_empty());
668    }
669
670    #[test]
671    fn orchestration_metrics_default_zero() {
672        let m = OrchestrationMetrics::default();
673        assert_eq!(m.plans_total, 0);
674        assert_eq!(m.tasks_total, 0);
675        assert_eq!(m.tasks_completed, 0);
676        assert_eq!(m.tasks_failed, 0);
677        assert_eq!(m.tasks_skipped, 0);
678    }
679
680    #[test]
681    fn metrics_snapshot_includes_orchestration_default_zero() {
682        let m = MetricsSnapshot::default();
683        assert_eq!(m.orchestration.plans_total, 0);
684        assert_eq!(m.orchestration.tasks_total, 0);
685        assert_eq!(m.orchestration.tasks_completed, 0);
686    }
687
688    #[test]
689    fn orchestration_metrics_update_via_collector() {
690        let (collector, rx) = MetricsCollector::new();
691        collector.update(|m| {
692            m.orchestration.plans_total += 1;
693            m.orchestration.tasks_total += 5;
694            m.orchestration.tasks_completed += 3;
695            m.orchestration.tasks_failed += 1;
696            m.orchestration.tasks_skipped += 1;
697        });
698        let s = rx.borrow();
699        assert_eq!(s.orchestration.plans_total, 1);
700        assert_eq!(s.orchestration.tasks_total, 5);
701        assert_eq!(s.orchestration.tasks_completed, 3);
702        assert_eq!(s.orchestration.tasks_failed, 1);
703        assert_eq!(s.orchestration.tasks_skipped, 1);
704    }
705
706    #[test]
707    fn strip_ctrl_removes_escape_sequences() {
708        let input = "hello\x1b[31mworld\x00end";
709        let result = strip_ctrl(input);
710        assert_eq!(result, "helloworldend");
711    }
712
713    #[test]
714    fn strip_ctrl_allows_tab_lf_cr() {
715        let input = "a\tb\nc\rd";
716        let result = strip_ctrl(input);
717        assert_eq!(result, "a\tb\nc\rd");
718    }
719
720    #[test]
721    fn task_graph_snapshot_is_stale_after_30s() {
722        let mut snap = TaskGraphSnapshot::default();
723        // Not stale if no completed_at.
724        assert!(!snap.is_stale());
725        // Not stale if just completed.
726        snap.completed_at = Some(std::time::Instant::now());
727        assert!(!snap.is_stale());
728        // Stale if completed more than 30s ago.
729        snap.completed_at = Some(
730            std::time::Instant::now()
731                .checked_sub(std::time::Duration::from_secs(31))
732                .unwrap(),
733        );
734        assert!(snap.is_stale());
735    }
736
737    // T1: From<&TaskGraph> correctly maps fields including duration_ms and error truncation.
738    #[test]
739    fn task_graph_snapshot_from_task_graph_maps_fields() {
740        use crate::orchestration::{GraphStatus, TaskGraph, TaskNode, TaskResult, TaskStatus};
741
742        let mut graph = TaskGraph::new("My goal");
743        let mut task = TaskNode::new(0, "Do work", "description");
744        task.status = TaskStatus::Failed;
745        task.assigned_agent = Some("agent-1".into());
746        task.result = Some(TaskResult {
747            output: "error occurred here".into(),
748            artifacts: vec![],
749            duration_ms: 1234,
750            agent_id: None,
751            agent_def: None,
752        });
753        graph.tasks.push(task);
754        graph.status = GraphStatus::Failed;
755
756        let snap = TaskGraphSnapshot::from(&graph);
757        assert_eq!(snap.goal, "My goal");
758        assert_eq!(snap.status, "failed");
759        assert_eq!(snap.tasks.len(), 1);
760        let row = &snap.tasks[0];
761        assert_eq!(row.title, "Do work");
762        assert_eq!(row.status, "failed");
763        assert_eq!(row.agent.as_deref(), Some("agent-1"));
764        assert_eq!(row.duration_ms, 1234);
765        assert!(row.error.as_deref().unwrap().contains("error occurred"));
766    }
767
768    // T2: From impl compiles with orchestration feature active.
769    #[test]
770    fn task_graph_snapshot_from_compiles_with_feature() {
771        use crate::orchestration::TaskGraph;
772        let graph = TaskGraph::new("feature flag test");
773        let snap = TaskGraphSnapshot::from(&graph);
774        assert_eq!(snap.goal, "feature flag test");
775        assert!(snap.tasks.is_empty());
776        assert!(!snap.is_stale());
777    }
778
779    // T1-extra: long error is truncated with ellipsis.
780    #[test]
781    fn task_graph_snapshot_error_truncated_at_80_chars() {
782        use crate::orchestration::{TaskGraph, TaskNode, TaskResult, TaskStatus};
783
784        let mut graph = TaskGraph::new("goal");
785        let mut task = TaskNode::new(0, "t", "d");
786        task.status = TaskStatus::Failed;
787        task.result = Some(TaskResult {
788            output: "e".repeat(100),
789            artifacts: vec![],
790            duration_ms: 0,
791            agent_id: None,
792            agent_def: None,
793        });
794        graph.tasks.push(task);
795
796        let snap = TaskGraphSnapshot::from(&graph);
797        let err = snap.tasks[0].error.as_ref().unwrap();
798        assert!(err.ends_with('…'), "truncated error must end with ellipsis");
799        assert!(
800            err.len() <= 83,
801            "truncated error must not exceed 80 chars + ellipsis"
802        );
803    }
804
805    // SEC-P6-01: control chars in task title are stripped.
806    #[test]
807    fn task_graph_snapshot_strips_control_chars_from_title() {
808        use crate::orchestration::{TaskGraph, TaskNode};
809
810        let mut graph = TaskGraph::new("goal\x1b[31m");
811        let task = TaskNode::new(0, "title\x00injected", "d");
812        graph.tasks.push(task);
813
814        let snap = TaskGraphSnapshot::from(&graph);
815        assert!(!snap.goal.contains('\x1b'), "goal must not contain escape");
816        assert!(
817            !snap.tasks[0].title.contains('\x00'),
818            "title must not contain null byte"
819        );
820    }
821
822    #[test]
823    fn graph_metrics_default_zero() {
824        let m = MetricsSnapshot::default();
825        assert_eq!(m.graph_entities_total, 0);
826        assert_eq!(m.graph_edges_total, 0);
827        assert_eq!(m.graph_communities_total, 0);
828        assert_eq!(m.graph_extraction_count, 0);
829        assert_eq!(m.graph_extraction_failures, 0);
830    }
831
832    #[test]
833    fn graph_metrics_update_via_collector() {
834        let (collector, rx) = MetricsCollector::new();
835        collector.update(|m| {
836            m.graph_entities_total = 5;
837            m.graph_edges_total = 10;
838            m.graph_communities_total = 2;
839            m.graph_extraction_count = 7;
840            m.graph_extraction_failures = 1;
841        });
842        let snapshot = rx.borrow().clone();
843        assert_eq!(snapshot.graph_entities_total, 5);
844        assert_eq!(snapshot.graph_edges_total, 10);
845        assert_eq!(snapshot.graph_communities_total, 2);
846        assert_eq!(snapshot.graph_extraction_count, 7);
847        assert_eq!(snapshot.graph_extraction_failures, 1);
848    }
849}
zeph_core/metrics.rs

zeph_core/
metrics.rs
