sentinel-agentos 0.3.6 → 0.3.8

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (97) hide show
  1. package/dist/src/dashboard.html +175 -0
  2. package/package.json +3 -2
  3. package/LICENSE +0 -21
  4. package/README.md +0 -1797
  5. package/dist/adapters/evaluation-bridge.d.ts +0 -78
  6. package/dist/adapters/evaluation-bridge.d.ts.map +0 -1
  7. package/dist/adapters/evaluation-bridge.js +0 -273
  8. package/dist/adapters/evaluation-bridge.js.map +0 -1
  9. package/dist/adapters/memory-bridge.d.ts +0 -110
  10. package/dist/adapters/memory-bridge.d.ts.map +0 -1
  11. package/dist/adapters/memory-bridge.js +0 -316
  12. package/dist/adapters/memory-bridge.js.map +0 -1
  13. package/dist/adapters/migrate.d.ts +0 -2
  14. package/dist/adapters/migrate.d.ts.map +0 -1
  15. package/dist/adapters/migrate.js +0 -63
  16. package/dist/adapters/migrate.js.map +0 -1
  17. package/dist/api.d.ts +0 -151
  18. package/dist/api.d.ts.map +0 -1
  19. package/dist/api.js +0 -179
  20. package/dist/api.js.map +0 -1
  21. package/dist/cli.d.ts +0 -16
  22. package/dist/cli.d.ts.map +0 -1
  23. package/dist/cli.js +0 -350
  24. package/dist/cli.js.map +0 -1
  25. package/dist/core.d.ts +0 -151
  26. package/dist/core.d.ts.map +0 -1
  27. package/dist/core.js +0 -341
  28. package/dist/core.js.map +0 -1
  29. package/dist/evaluator/exec-evaluator.d.ts +0 -102
  30. package/dist/evaluator/exec-evaluator.d.ts.map +0 -1
  31. package/dist/evaluator/exec-evaluator.js +0 -266
  32. package/dist/evaluator/exec-evaluator.js.map +0 -1
  33. package/dist/evaluator/feedback.d.ts +0 -57
  34. package/dist/evaluator/feedback.d.ts.map +0 -1
  35. package/dist/evaluator/feedback.js +0 -225
  36. package/dist/evaluator/feedback.js.map +0 -1
  37. package/dist/evaluator/profiler.d.ts +0 -55
  38. package/dist/evaluator/profiler.d.ts.map +0 -1
  39. package/dist/evaluator/profiler.js +0 -117
  40. package/dist/evaluator/profiler.js.map +0 -1
  41. package/dist/guard/audit-log.d.ts +0 -47
  42. package/dist/guard/audit-log.d.ts.map +0 -1
  43. package/dist/guard/audit-log.js +0 -199
  44. package/dist/guard/audit-log.js.map +0 -1
  45. package/dist/guard/container-sandbox.d.ts +0 -25
  46. package/dist/guard/container-sandbox.d.ts.map +0 -1
  47. package/dist/guard/container-sandbox.js +0 -145
  48. package/dist/guard/container-sandbox.js.map +0 -1
  49. package/dist/guard/risk-gate.d.ts +0 -101
  50. package/dist/guard/risk-gate.d.ts.map +0 -1
  51. package/dist/guard/risk-gate.js +0 -200
  52. package/dist/guard/risk-gate.js.map +0 -1
  53. package/dist/guard/sandbox.d.ts +0 -112
  54. package/dist/guard/sandbox.d.ts.map +0 -1
  55. package/dist/guard/sandbox.js +0 -379
  56. package/dist/guard/sandbox.js.map +0 -1
  57. package/dist/guard/schema-gate.d.ts +0 -90
  58. package/dist/guard/schema-gate.d.ts.map +0 -1
  59. package/dist/guard/schema-gate.js +0 -452
  60. package/dist/guard/schema-gate.js.map +0 -1
  61. package/dist/guard/snapshot-verify.d.ts +0 -111
  62. package/dist/guard/snapshot-verify.d.ts.map +0 -1
  63. package/dist/guard/snapshot-verify.js +0 -571
  64. package/dist/guard/snapshot-verify.js.map +0 -1
  65. package/dist/index.d.ts +0 -28
  66. package/dist/index.d.ts.map +0 -1
  67. package/dist/index.js +0 -59
  68. package/dist/index.js.map +0 -1
  69. package/dist/memory/episodic.d.ts +0 -76
  70. package/dist/memory/episodic.d.ts.map +0 -1
  71. package/dist/memory/episodic.js +0 -289
  72. package/dist/memory/episodic.js.map +0 -1
  73. package/dist/memory/semantic.d.ts +0 -68
  74. package/dist/memory/semantic.d.ts.map +0 -1
  75. package/dist/memory/semantic.js +0 -299
  76. package/dist/memory/semantic.js.map +0 -1
  77. package/dist/memory/working.d.ts +0 -53
  78. package/dist/memory/working.d.ts.map +0 -1
  79. package/dist/memory/working.js +0 -166
  80. package/dist/memory/working.js.map +0 -1
  81. package/dist/middleware/openclaw.d.ts +0 -45
  82. package/dist/middleware/openclaw.d.ts.map +0 -1
  83. package/dist/middleware/openclaw.js +0 -95
  84. package/dist/middleware/openclaw.js.map +0 -1
  85. package/dist/middleware/wrapper.d.ts +0 -54
  86. package/dist/middleware/wrapper.d.ts.map +0 -1
  87. package/dist/middleware/wrapper.js +0 -155
  88. package/dist/middleware/wrapper.js.map +0 -1
  89. package/dist/server.d.ts +0 -45
  90. package/dist/server.d.ts.map +0 -1
  91. package/dist/server.js +0 -246
  92. package/dist/server.js.map +0 -1
  93. package/dist/types/index.d.ts +0 -228
  94. package/dist/types/index.d.ts.map +0 -1
  95. package/dist/types/index.js +0 -23
  96. package/dist/types/index.js.map +0 -1
  97. package/scripts/sentinel-light.js +0 -234
package/dist/core.d.ts DELETED
@@ -1,151 +0,0 @@
1
- import { AgentOSConfig, GuardConfig, Snapshot, AuditEntry, PreExecMetrics, RuntimeMetrics, PostExecMetrics } from './types';
2
- import { SchemaGate } from './guard/schema-gate';
3
- import { RiskGate } from './guard/risk-gate';
4
- import { SnapshotGate, VerifyGate } from './guard/snapshot-verify';
5
- import { AuditLog } from './guard/audit-log';
6
- import { WorkingMemory } from './memory/working';
7
- import { EpisodicMemory } from './memory/episodic';
8
- import { SemanticMemoryStore } from './memory/semantic';
9
- import { PreExecEvaluator, RuntimeEvaluator, PostExecEvaluator } from './evaluator/exec-evaluator';
10
- import { ImplicitFeedbackEngine } from './evaluator/feedback';
11
- import { AgentProfiler } from './evaluator/profiler';
12
- import type { AgentProfile } from './evaluator/profiler';
13
- /**
14
- * AgentOS — the complete AI Agent Operating System.
15
- *
16
- * Architecture:
17
- * ```
18
- * User Request
19
- * ↓
20
- * ┌─────────────┐
21
- * │ Memory Layer │ ← Semantic + Episodic + Working memory
22
- * ├─────────────┤
23
- * │ Guard Layer │ ← Schema → Risk → Snapshot
24
- * ├─────────────┤ ↓
25
- * │ Execute │ ← Tool call execution
26
- * ├─────────────┤ ↓
27
- * │ Verify Layer │ ← Snapshot diff → Verify checks
28
- * ├─────────────┤ ↓
29
- * │ Audit Log │ ← Immutable operation record
30
- * ├─────────────┤
31
- * │ Evaluator │ ← Three-phase metrics + feedback
32
- * └─────────────┘
33
- * ```
34
- */
35
- export declare class AgentOS {
36
- private config;
37
- readonly memory: {
38
- working: WorkingMemory;
39
- episodic: EpisodicMemory;
40
- semantic: SemanticMemoryStore;
41
- };
42
- readonly guard: {
43
- schema: SchemaGate;
44
- risk: RiskGate;
45
- snapshot: SnapshotGate;
46
- verify: VerifyGate;
47
- audit: AuditLog;
48
- };
49
- readonly evaluator: {
50
- preExec: PreExecEvaluator;
51
- runtime: RuntimeEvaluator;
52
- postExec: PostExecEvaluator;
53
- feedback: ImplicitFeedbackEngine;
54
- profiler: AgentProfiler;
55
- };
56
- constructor(config?: Partial<AgentOSConfig>);
57
- /**
58
- * Get the current AgentOS configuration.
59
- */
60
- getConfig(): Readonly<AgentOSConfig>;
61
- /**
62
- * Full pipeline: process a tool call through all layers.
63
- *
64
- * This is the main AgentOS orchestration method.
65
- * In production this would be called by the agent runtime
66
- * before/after every tool call.
67
- */
68
- executePipeline(options: {
69
- sessionId: string;
70
- agentId: string;
71
- toolName: string;
72
- parameters: Record<string, unknown>;
73
- affectedFiles?: string[];
74
- guardConfig?: GuardConfig;
75
- }): {
76
- preExec: PreExecMetrics;
77
- runtime?: RuntimeMetrics;
78
- postExec?: PostExecMetrics;
79
- snapshot: Snapshot | null;
80
- auditEntry?: AuditEntry;
81
- profile: AgentProfile;
82
- };
83
- /**
84
- * Complete the pipeline after tool execution.
85
- *
86
- * Called by the runtime after the tool call completes
87
- * (or fails, or times out).
88
- */
89
- completeExecution(options: {
90
- sessionId: string;
91
- agentId: string;
92
- toolName: string;
93
- toolParameters: Record<string, unknown>;
94
- toolResult: unknown;
95
- snapshot: Snapshot | null;
96
- startTime: number;
97
- endTime: number;
98
- retryCount: number;
99
- wasSelfCorrected: boolean;
100
- hadTimeout: boolean;
101
- userAccepted: boolean;
102
- userProvidedEdit: boolean;
103
- resultWasUsed: boolean;
104
- }): {
105
- runtime: RuntimeMetrics;
106
- postExec: PostExecMetrics;
107
- auditEntry: AuditEntry;
108
- profile: AgentProfile;
109
- };
110
- /**
111
- * Record implicit user feedback.
112
- */
113
- recordFeedback(signal: Parameters<ImplicitFeedbackEngine['record']>[0], sessionId: string, operationId?: string, confidence?: number, source?: string): void;
114
- /**
115
- * Get a JSON report for dashboard consumption.
116
- */
117
- getReport(): Record<string, unknown>;
118
- /**
119
- * Inject memory context at session startup.
120
- *
121
- * Call this at the beginning of every session to load
122
- * semantic + episodic context into the session prompt.
123
- */
124
- injectContext(): string;
125
- /**
126
- * End current session — auto-detect feedback, promote events to episodic,
127
- * append daily log, clear working memory.
128
- */
129
- endSession(sessionId: string, workspaceRoot?: string): void;
130
- /**
131
- * Append evaluation summary to daily log file.
132
- */
133
- private appendDailyLog;
134
- /**
135
- * Detect if this tool call's parameters reference a previous result.
136
- */
137
- private detectResultUtilization;
138
- /**
139
- * Get audit statistics.
140
- */
141
- getAuditStats(): ReturnType<AuditLog['stats']>;
142
- /**
143
- * Get the current agent quality profile.
144
- */
145
- getProfile(sessionId?: string): AgentProfile;
146
- /**
147
- * Get a summarized status report.
148
- */
149
- statusReport(): string;
150
- }
151
- //# sourceMappingURL=core.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"core.d.ts","sourceRoot":"","sources":["../src/core.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,aAAa,EACb,WAAW,EACX,QAAQ,EACR,UAAU,EACV,cAAc,EACd,cAAc,EACd,eAAe,EAChB,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AACjD,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAC7C,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAC7C,OAAO,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACjD,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AACnD,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MAAM,4BAA4B,CAAC;AACnG,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAEzD;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,qBAAa,OAAO;IAClB,OAAO,CAAC,MAAM,CAAgB;IAG9B,QAAQ,CAAC,MAAM,EAAE;QACf,OAAO,EAAE,aAAa,CAAC;QACvB,QAAQ,EAAE,cAAc,CAAC;QACzB,QAAQ,EAAE,mBAAmB,CAAC;KAC/B,CAAC;IAGF,QAAQ,CAAC,KAAK,EAAE;QACd,MAAM,EAAE,UAAU,CAAC;QACnB,IAAI,EAAE,QAAQ,CAAC;QACf,QAAQ,EAAE,YAAY,CAAC;QACvB,MAAM,EAAE,UAAU,CAAC;QACnB,KAAK,EAAE,QAAQ,CAAC;KACjB,CAAC;IAGF,QAAQ,CAAC,SAAS,EAAE;QAClB,OAAO,EAAE,gBAAgB,CAAC;QAC1B,OAAO,EAAE,gBAAgB,CAAC;QAC1B,QAAQ,EAAE,iBAAiB,CAAC;QAC5B,QAAQ,EAAE,sBAAsB,CAAC;QACjC,QAAQ,EAAE,aAAa,CAAC;KACzB,CAAC;gBAEU,MAAM,CAAC,EAAE,OAAO,CAAC,aAAa,CAAC;IAqD3C;;OAEG;IACH,SAAS,IAAI,QAAQ,CAAC,aAAa,CAAC;IAIpC;;;;;;OAMG;IACH,eAAe,CAAC,OAAO,EAAE;QACvB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QACpC,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;QACzB,WAAW,CAAC,EAAE,WAAW,CAAC;KAC3B,GAAG;QACF,OAAO,EAAE,cAAc,CAAC;QACxB,OAAO,CAAC,EAAE,cAAc,CAAC;QACzB,QAAQ,CAAC,EAAE,eAAe,CAAC;QAC3B,QAAQ,EAAE,QAAQ,GAAG,IAAI,CAAC;QAC1B,UAAU,CAAC,EAAE,UAAU,CAAC;QACxB,OAAO,EAAE,YAAY,CAAC;KACvB;IAyBD;;;;;OAKG;IACH,iBAAiB,CAAC,OAAO,EAAE;QACzB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QACxC,UAAU,EAAE,OAAO,CAAC;QACpB,QAAQ,EAAE,QAAQ,GAAG,IAAI,CAAC;QAC1B,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,MAAM,CAAC;QACnB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,UAAU,EAAE,OAAO,CAAC;QACpB,YAAY,EAAE,OAAO,CAAC;QACtB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,aAAa,EAAE,OAAO,CAAC;KACxB,GAAG;QACF,OAAO,EAAE,cAAc,CAAC;QACxB,QAAQ,EAAE,eAAe,CAAC;QAC1B,UAAU,EAAE,UAAU,CAAC;QACvB,OAAO,EAAE,YAAY,CAAC;KACvB;IAyED;;OAEG;IACH,cAAc,CACZ,MAAM,EAAE,UAAU,CAAC,sBAAsB,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,EACvD,SAAS,EAAE,MAAM,EACjB,WAAW,CAAC,EAAE,MAAM,EACpB,UAAU,CAAC,EAAE,MAAM,EACnB,MAAM,CAAC,EAAE,MAAM,GACd,IAAI;IAIP;;OAEG;IACH,SAAS,IAAI,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC;IAcpC;;;;;OAKG;IACH,aAAa,IAAI,MAAM;IAYvB;;;OAGG;IACH,UAAU,CAAC,SAAS,EAAE,MAAM,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI;IAsD3D;;OAEG;IACH,OAAO,CAAC,cAAc;IAiCtB;;OAEG;IACH,OAAO,CAAC,uBAAuB;IAoB/B;;OAEG;IACH,aAAa,IAAI,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;IAI9C;;OAEG;IACH,UAAU,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,YAAY;IAI5C;;OAEG;IACH,YAAY,IAAI,MAAM;CAgCvB"}
package/dist/core.js DELETED
@@ -1,341 +0,0 @@
1
- "use strict";
2
- Object.defineProperty(exports, "__esModule", { value: true });
3
- exports.AgentOS = void 0;
4
- const schema_gate_1 = require("./guard/schema-gate");
5
- const risk_gate_1 = require("./guard/risk-gate");
6
- const snapshot_verify_1 = require("./guard/snapshot-verify");
7
- const audit_log_1 = require("./guard/audit-log");
8
- const working_1 = require("./memory/working");
9
- const episodic_1 = require("./memory/episodic");
10
- const semantic_1 = require("./memory/semantic");
11
- const exec_evaluator_1 = require("./evaluator/exec-evaluator");
12
- const feedback_1 = require("./evaluator/feedback");
13
- const profiler_1 = require("./evaluator/profiler");
14
- /**
15
- * AgentOS — the complete AI Agent Operating System.
16
- *
17
- * Architecture:
18
- * ```
19
- * User Request
20
- * ↓
21
- * ┌─────────────┐
22
- * │ Memory Layer │ ← Semantic + Episodic + Working memory
23
- * ├─────────────┤
24
- * │ Guard Layer │ ← Schema → Risk → Snapshot
25
- * ├─────────────┤ ↓
26
- * │ Execute │ ← Tool call execution
27
- * ├─────────────┤ ↓
28
- * │ Verify Layer │ ← Snapshot diff → Verify checks
29
- * ├─────────────┤ ↓
30
- * │ Audit Log │ ← Immutable operation record
31
- * ├─────────────┤
32
- * │ Evaluator │ ← Three-phase metrics + feedback
33
- * └─────────────┘
34
- * ```
35
- */
36
- class AgentOS {
37
- config;
38
- // Memory Layer
39
- memory;
40
- // Guard Layer
41
- guard;
42
- // Evaluator Layer
43
- evaluator;
44
- constructor(config) {
45
- this.config = {
46
- workspaceRoot: process.cwd(),
47
- maxWorkingTokens: 50000,
48
- maxEpisodicSizeKb: 500,
49
- guardConfig: {},
50
- ...config,
51
- };
52
- // --- Memory Layer Init ---
53
- const semantic = new semantic_1.SemanticMemoryStore();
54
- semantic.enablePersistence(this.config.workspaceRoot);
55
- const episodic = new episodic_1.EpisodicMemory(this.config.maxEpisodicSizeKb);
56
- episodic.enablePersistence(this.config.workspaceRoot);
57
- const working = new working_1.WorkingMemory(this.config.maxWorkingTokens);
58
- this.memory = { working, episodic, semantic };
59
- // --- Guard Layer Init ---
60
- const schema = new schema_gate_1.SchemaGate();
61
- const risk = new risk_gate_1.RiskGate();
62
- const snapshot = new snapshot_verify_1.SnapshotGate(this.config.workspaceRoot);
63
- const verify = new snapshot_verify_1.VerifyGate(this.config.workspaceRoot);
64
- const audit = new audit_log_1.AuditLog(this.config.workspaceRoot, schema, risk);
65
- this.guard = { schema, risk, snapshot, verify, audit };
66
- // --- Evaluator Layer Init ---
67
- const preExecEval = new exec_evaluator_1.PreExecEvaluator(schema, risk, working);
68
- const runtimeEval = new exec_evaluator_1.RuntimeEvaluator();
69
- const postExecEval = new exec_evaluator_1.PostExecEvaluator();
70
- const feedbackEngine = new feedback_1.ImplicitFeedbackEngine();
71
- const profiler = new profiler_1.AgentProfiler(feedbackEngine);
72
- this.evaluator = {
73
- preExec: preExecEval,
74
- runtime: runtimeEval,
75
- postExec: postExecEval,
76
- feedback: feedbackEngine,
77
- profiler,
78
- };
79
- }
80
- /**
81
- * Get the current AgentOS configuration.
82
- */
83
- getConfig() {
84
- return this.config;
85
- }
86
- /**
87
- * Full pipeline: process a tool call through all layers.
88
- *
89
- * This is the main AgentOS orchestration method.
90
- * In production this would be called by the agent runtime
91
- * before/after every tool call.
92
- */
93
- executePipeline(options) {
94
- const { sessionId, toolName, parameters, affectedFiles } = options;
95
- // --- Phase 1: Pre-exec evaluation ---
96
- const preExec = this.evaluator.preExec.evaluate(toolName, parameters);
97
- this.memory.working.addMessage('tool', JSON.stringify(parameters));
98
- // --- Phase 2: Snapshot before execution ---
99
- const snapshot = this.guard.snapshot.takeSnapshot(`call_${Date.now()}`, toolName, affectedFiles ?? [], 'file');
100
- // --- Phase 3: Execute (delegated to runtime — here we return pre-exec state) ---
101
- // Return the pre-exec state so the runtime can complete
102
- return {
103
- preExec,
104
- snapshot,
105
- profile: this.evaluator.profiler.getProfile(sessionId),
106
- };
107
- }
108
- /**
109
- * Complete the pipeline after tool execution.
110
- *
111
- * Called by the runtime after the tool call completes
112
- * (or fails, or times out).
113
- */
114
- completeExecution(options) {
115
- const { sessionId, agentId, toolName, toolParameters, toolResult, snapshot, startTime, endTime, retryCount, wasSelfCorrected, hadTimeout, userAccepted, userProvidedEdit, resultWasUsed, } = options;
116
- // --- Phase 1: Runtime evaluation ---
117
- const runtime = this.evaluator.runtime.evaluate({
118
- toolName,
119
- startTime,
120
- endTime,
121
- retryCount,
122
- wasSelfCorrected,
123
- hadTimeout,
124
- toolResult,
125
- });
126
- // --- Phase 2: Post-exec verification ---
127
- const verifyResult = this.guard.verify.verify(toolName, snapshot, { files: toolParameters['path'] ? [String(toolParameters['path'])] : undefined });
128
- // --- Phase 3: Post-exec evaluation ---
129
- const postExec = this.evaluator.postExec.evaluate({
130
- verifyPassed: verifyResult.status === 'PASS',
131
- verifyChecks: verifyResult.checks.length,
132
- verifyFailures: verifyResult.checks.filter((c) => c.status === 'FAIL').length,
133
- userAccepted,
134
- userProvidedEdit,
135
- resultWasUsed,
136
- });
137
- // --- Phase 4: Audit log ---
138
- const auditEntry = this.guard.audit.record({
139
- sessionId,
140
- agentId,
141
- startedAt: startTime,
142
- completedAt: endTime,
143
- toolName,
144
- toolParameters,
145
- toolResult,
146
- snapshot,
147
- verifyStatus: verifyResult.status,
148
- verifyChecks: verifyResult.checks,
149
- });
150
- // --- Phase 5: Track result for utilization scoring ---
151
- this.evaluator.postExec.trackResult(auditEntry.id, toolResult);
152
- // --- Phase 5.5: Auto-detect result utilization (after track, avoids race) ---
153
- this.detectResultUtilization(toolParameters, sessionId);
154
- // --- Phase 6: Record in profiler (reuse pre-exec from pipeline start) ---
155
- const preExecMetric = this.evaluator.preExec.evaluate(toolName, toolParameters);
156
- this.evaluator.profiler.recordCycle(sessionId, preExecMetric, runtime, postExec);
157
- return {
158
- runtime,
159
- postExec,
160
- auditEntry,
161
- profile: this.evaluator.profiler.getProfile(sessionId),
162
- };
163
- }
164
- /**
165
- * Record implicit user feedback.
166
- */
167
- recordFeedback(signal, sessionId, operationId, confidence, source) {
168
- this.evaluator.feedback.record(signal, sessionId, operationId, confidence, source);
169
- }
170
- /**
171
- * Get a JSON report for dashboard consumption.
172
- */
173
- getReport() {
174
- const profile = this.getProfile();
175
- const audit = this.getAuditStats();
176
- return {
177
- quality: { overallScore: Math.round(profile.overallScore), ...profile.breakdown },
178
- audit: { ...audit },
179
- satisfaction: profile.breakdown.userSatisfaction || 0,
180
- workingMemory: { messages: this.memory.working.getState().recentMessages.length },
181
- episodicEvents: this.memory.episodic.count || 0,
182
- warnings: profile.warnings,
183
- strengths: profile.strengths,
184
- };
185
- }
186
- /**
187
- * Inject memory context at session startup.
188
- *
189
- * Call this at the beginning of every session to load
190
- * semantic + episodic context into the session prompt.
191
- */
192
- injectContext() {
193
- const semanticSummary = this.memory.semantic.generateContextSummary(2000);
194
- const episodicSummary = this.memory.episodic.generateContextSummary(1500);
195
- const parts = [];
196
- if (semanticSummary)
197
- parts.push(semanticSummary);
198
- if (episodicSummary)
199
- parts.push(episodicSummary);
200
- return parts.join('\n\n---\n\n');
201
- }
202
- /**
203
- * End current session — auto-detect feedback, promote events to episodic,
204
- * append daily log, clear working memory.
205
- */
206
- endSession(sessionId, workspaceRoot) {
207
- // Phase 1: Auto-detect feedback from audit log
208
- const recentAudit = this.guard.audit.query({ sessionId, limit: 100 });
209
- const detected = this.evaluator.feedback.autoDetect(recentAudit, sessionId);
210
- // Phase 2: Promote important working memory items to episodic
211
- if (this.memory.working.currentTask) {
212
- this.memory.episodic.record('milestone', `Task completed: ${this.memory.working.currentTask.description}`, ['task', 'session-end'], [sessionId]);
213
- }
214
- // Phase 3: Log learned rules from this session
215
- const rules = this.memory.semantic.getRules(0.6);
216
- for (const rule of rules.slice(0, 3)) {
217
- this.memory.episodic.record('note', `Rule: ${rule.rule} (confidence: ${Math.round(rule.confidence * 100)}%)`, ['rule', 'semantic'], []);
218
- }
219
- // Phase 4: Record session end event with stats
220
- const profile = this.evaluator.profiler.getProfile(sessionId);
221
- const auditStats = this.guard.audit.stats();
222
- this.memory.episodic.record('note', `Session ended — Score: ${profile.overallScore}/100 | Ops: ${auditStats.totalOperations} | Feedback signals detected: ${detected}`, ['session-end', 'summary'], [sessionId]);
223
- // Phase 5: Append daily log if workspaceRoot provided
224
- if (workspaceRoot) {
225
- this.appendDailyLog(sessionId, workspaceRoot);
226
- }
227
- // Phase 5: Decay unused semantic rules
228
- this.memory.semantic.decayUnusedRules();
229
- // Phase 5.3: Clean up profiler session scores
230
- this.evaluator.profiler.clearSession(sessionId);
231
- // Phase 5.5: Force Episodic compression
232
- this.memory.episodic.record('note', 'Session end', ['session-end'], []);
233
- // Phase 6: Clear working memory for next session
234
- this.memory.working.clear();
235
- }
236
- /**
237
- * Append evaluation summary to daily log file.
238
- */
239
- appendDailyLog(sessionId, workspaceRoot) {
240
- try {
241
- const { appendFileSync, existsSync, mkdirSync } = require('fs');
242
- const path = require('path');
243
- const dateKey = new Date().toISOString().split('T')[0];
244
- const memDir = path.join(workspaceRoot, 'memory');
245
- if (!existsSync(memDir))
246
- mkdirSync(memDir, { recursive: true });
247
- const dailyFile = path.join(memDir, `${dateKey}.md`);
248
- const profile = this.evaluator.profiler.getProfile(sessionId);
249
- const auditStats = this.guard.audit.stats();
250
- const satisfaction = this.evaluator.feedback.getSatisfactionScore(sessionId);
251
- const report = [
252
- '',
253
- '## 📊 AgentOS Evaluator 今日评估',
254
- '',
255
- `**综合评分**: ${profile.overallScore}/100 | Pre:${profile.breakdown.preExec ?? 'N/A'}/100 | Run:${profile.breakdown.runtime ?? 'N/A'}/100 | Post:${profile.breakdown.postExec ?? 'N/A'}/100`,
256
- `**趋势**: ${profile.trends.improving ? '📈 上升' : '📉 下降'} | **操作数**: ${auditStats.totalOperations} | **满意度**: ${satisfaction}`,
257
- profile.warnings.length > 0 ? `**⚠️**: ${profile.warnings.join('; ')}` : '',
258
- profile.strengths.length > 0 ? `**✅**: ${profile.strengths.join('; ')}` : '',
259
- '',
260
- ].filter(Boolean).join('\n');
261
- const existing = existsSync(dailyFile) ? require('fs').readFileSync(dailyFile, 'utf-8') : '';
262
- if (!existing.includes('AgentOS Evaluator')) {
263
- appendFileSync(dailyFile, report, 'utf-8');
264
- }
265
- }
266
- catch (e) {
267
- console.warn('[AgentOS] Failed to append daily log:', e);
268
- }
269
- }
270
- /**
271
- * Detect if this tool call's parameters reference a previous result.
272
- */
273
- detectResultUtilization(params, sessionId) {
274
- try {
275
- const allValues = Object.values(params).join(' ');
276
- const tracker = this.evaluator.postExec['resultReferenceTracker'];
277
- if (!tracker)
278
- return;
279
- for (const [opId, entry] of tracker) {
280
- if (entry.referenced)
281
- continue;
282
- const resultStr = JSON.stringify(entry.result);
283
- // Check if parameters contain a reference to previous result
284
- if (resultStr.length > 10 && allValues.includes(resultStr.substring(0, 20))) {
285
- this.evaluator.postExec.markResultReferenced(opId);
286
- this.evaluator.feedback.record('user_used_result', sessionId, opId, 0.5, 'auto-param-match');
287
- }
288
- }
289
- }
290
- catch {
291
- // Non-critical, silently ignore detection failures
292
- }
293
- }
294
- /**
295
- * Get audit statistics.
296
- */
297
- getAuditStats() {
298
- return this.guard.audit.stats();
299
- }
300
- /**
301
- * Get the current agent quality profile.
302
- */
303
- getProfile(sessionId) {
304
- return this.evaluator.profiler.getProfile(sessionId);
305
- }
306
- /**
307
- * Get a summarized status report.
308
- */
309
- statusReport() {
310
- const profile = this.getProfile();
311
- const audit = this.getAuditStats();
312
- const lines = [
313
- '=== AgentOS Status Report ===',
314
- '',
315
- `Quality Score: ${Math.round(profile.overallScore)}/100 ${profile.trends.improving ? '📈' : '📉'}`,
316
- `Total Operations: ${profile.totalOps} (${profile.trends.recentOps} in last 24h)`,
317
- '',
318
- '--- Breakdown ---',
319
- `Pre-Exec: ${profile.breakdown.preExec}/100`,
320
- `Runtime: ${profile.breakdown.runtime}/100`,
321
- `Post-Exec: ${profile.breakdown.postExec}/100`,
322
- `Satisfaction: ${profile.breakdown.userSatisfaction}/100`,
323
- '',
324
- '--- Audit ---',
325
- `Total: ${audit.totalOperations} | Failures: ${audit.verifyFailures} | High-Risk: ${audit.highRiskOps}`,
326
- ];
327
- if (profile.warnings.length > 0) {
328
- lines.push('', '--- ⚠️ Warnings ---');
329
- for (const w of profile.warnings)
330
- lines.push(`- ${w}`);
331
- }
332
- if (profile.strengths.length > 0) {
333
- lines.push('', '--- ✅ Strengths ---');
334
- for (const s of profile.strengths)
335
- lines.push(`- ${s}`);
336
- }
337
- return lines.join('\n');
338
- }
339
- }
340
- exports.AgentOS = AgentOS;
341
- //# sourceMappingURL=core.js.map
package/dist/core.js.map DELETED
@@ -1 +0,0 @@
1
- {"version":3,"file":"core.js","sourceRoot":"","sources":["../src/core.ts"],"names":[],"mappings":";;;AASA,qDAAiD;AACjD,iDAA6C;AAC7C,6DAAmE;AACnE,iDAA6C;AAC7C,8CAAiD;AACjD,gDAAmD;AACnD,gDAAwD;AACxD,+DAAmG;AACnG,mDAA8D;AAC9D,mDAAqD;AAGrD;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAa,OAAO;IACV,MAAM,CAAgB;IAE9B,eAAe;IACN,MAAM,CAIb;IAEF,cAAc;IACL,KAAK,CAMZ;IAEF,kBAAkB;IACT,SAAS,CAMhB;IAEF,YAAY,MAA+B;QACzC,IAAI,CAAC,MAAM,GAAG;YACZ,aAAa,EAAE,OAAO,CAAC,GAAG,EAAE;YAC5B,gBAAgB,EAAE,KAAK;YACvB,iBAAiB,EAAE,GAAG;YACtB,WAAW,EAAE,EAAE;YACf,GAAG,MAAM;SACV,CAAC;QAEF,4BAA4B;QAC5B,MAAM,QAAQ,GAAG,IAAI,8BAAmB,EAAE,CAAC;QAC3C,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAEvD,MAAM,QAAQ,GAAG,IAAI,yBAAc,CAAC,IAAI,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;QACnE,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAEvD,MAAM,OAAO,GAAG,IAAI,uBAAa,CAAC,IAAI,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC;QAEhE,IAAI,CAAC,MAAM,GAAG,EAAE,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC;QAE9C,2BAA2B;QAC3B,MAAM,MAAM,GAAG,IAAI,wBAAU,EAAE,CAAC;QAEhC,MAAM,IAAI,GAAG,IAAI,oBAAQ,EAAE,CAAC;QAE5B,MAAM,QAAQ,GAAG,IAAI,8BAAY,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAE9D,MAAM,MAAM,GAAG,IAAI,4BAAU,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAE1D,MAAM,KAAK,GAAG,IAAI,oBAAQ,CACxB,IAAI,CAAC,MAAM,CAAC,aAAc,EAC1B,MAAM,EACN,IAAI,CACL,CAAC;QAEF,IAAI,CAAC,KAAK,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;QAEvD,+BAA+B;QAC/B,MAAM,WAAW,GAAG,IAAI,iCAAgB,CAAC,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAChE,MAAM,WAAW,GAAG,IAAI,iCAAgB,EAAE,CAAC;QAC3C,MAAM,YAAY,GAAG,IAAI,kCAAiB,EAAE,CAAC;QAC7C,MAAM,cAAc,GAAG,IAAI,iCAAsB,EAAE,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,wBAAa,CAAC,cAAc,CAAC,CAAC;QAEnD,IAAI,CAAC,SAAS,GAAG;YACf,OAAO,EAAE,WAAW;YACpB,OAAO,EAAE,WAAW;YACpB,QAAQ,EAAE,YAAY;YACtB,QAAQ,EAAE,cAAc;YACxB,QAAQ;SACT,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,SAAS;QACP,OAAO,IAAI,CAAC,MAAM,CAAC;IACrB,CAAC;IAED;;;;;;OAMG;IACH,eAAe,CAAC,OAOf;QAQC,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,EAAE,GAAG,OAAO,CAAC;QAEnE,uCAAuC;QACvC,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;QACtE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC,CAAC;QAEnE,6CAA6C;QAC7C,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,YAAY,CAC/C,QAAQ,IAAI,CAAC,GAAG,EAAE,EAAE,EACpB,QAAQ,EACR,aAAa,IAAI,EAAE,EACnB,MAAM,CACP,CAAC;QAEF,kFAAkF;QAElF,wDAAwD;QACxD,OAAO;YACL,OAAO;YACP,QAAQ;YACR,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC;SACvD,CAAC;IACJ,CAAC;IAED;;;;;OAKG;IACH,iBAAiB,CAAC,OAejB;QAMC,MAAM,EACJ,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,UAAU,EACxD,QAAQ,EAAE,SAAS,EAAE,OAAO,EAC5B,UAAU,EAAE,gBAAgB,EAAE,UAAU,EACxC,YAAY,EAAE,gBAAgB,EAAE,aAAa,GAC9C,GAAG,OAAO,CAAC;QAEZ,sCAAsC;QACtC,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC;YAC9C,QAAQ;YACR,SAAS;YACT,OAAO;YACP,UAAU;YACV,gBAAgB;YAChB,UAAU;YACV,UAAU;SACX,CAAC,CAAC;QAEH,0CAA0C;QAC1C,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAC3C,QAAQ,EACR,QAAS,EACT,EAAE,KAAK,EAAE,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CACjF,CAAC;QAEF,wCAAwC;QACxC,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC;YAChD,YAAY,EAAE,YAAY,CAAC,MAAM,KAAK,MAAM;YAC5C,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,MAAM;YACxC,cAAc,EAAE,YAAY,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM;YAC7E,YAAY;YACZ,gBAAgB;YAChB,aAAa;SACd,CAAC,CAAC;QAEH,6BAA6B;QAC7B,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC;YACzC,SAAS;YACT,OAAO;YACP,SAAS,EAAE,SAAS;YACpB,WAAW,EAAE,OAAO;YACpB,QAAQ;YACR,cAAc;YACd,UAAU;YACV,QAAQ;YACR,YAAY,EAAE,YAAY,CAAC,MAAM;YACjC,YAAY,EAAE,YAAY,CAAC,MAAM;SAClC,CAAC,CAAC;QAEH,wDAAwD;QACxD,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,WAAW,CAAC,UAAU,CAAC,EAAE,EAAE,UAAU,CAAC,CAAC;QAE/D,+EAA+E;QAC/E,IAAI,CAAC,uBAAuB,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;QAExD,2EAA2E;QAC3E,MAAM,aAAa,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,EAAE,cAAc,CAAC,CAAC;QAChF,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,WAAW,CACjC,SAAS,EACT,aAAa,EACb,OAAO,EACP,QAAQ,CACT,CAAC;QAEF,OAAO;YACL,OAAO;YACP,QAAQ;YACR,UAAU;YACV,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC;SACvD,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,cAAc,CACZ,MAAuD,EACvD,SAAiB,EACjB,WAAoB,EACpB,UAAmB,EACnB,MAAe;QAEf,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;IACrF,CAAC;IAED;;OAEG;IACH,SAAS;QACP,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;QAClC,MAAM,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QACnC,OAAO;YACL,OAAO,EAAE,EAAE,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,GAAG,OAAO,CAAC,SAAS,EAAE;YACjF,KAAK,EAAE,EAAE,GAAG,KAAK,EAAE;YACnB,YAAY,EAAE,OAAO,CAAC,SAAS,CAAC,gBAAgB,IAAI,CAAC;YACrD,aAAa,EAAE,EAAE,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC,cAAc,CAAC,MAAM,EAAE;YACjF,cAAc,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,IAAI,CAAC;YAC/C,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;SAC7B,CAAC;IACJ,CAAC;IAED;;;;;OAKG;IACH,aAAa;QACX,MAAM,eAAe,GAAG,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,sBAAsB,CAAC,IAAI,CAAC,CAAC;QAC1E,MAAM,eAAe,GAAG,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,sBAAsB,CAAC,IAAI,CAAC,CAAC;QAE1E,MAAM,KAAK,GAAa,EAAE,CAAC;QAE3B,IAAI,eAAe;YAAE,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QACjD,IAAI,eAAe;YAAE,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAEjD,OAAO,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IACnC,CAAC;IAED;;;OAGG;IACH,UAAU,CAAC,SAAiB,EAAE,aAAsB;QAClD,+CAA+C;QAC/C,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,GAAG,EAAE,CAAC,CAAC;QACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;QAE5E,8DAA8D;QAC9D,IAAI,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC;YACpC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CACzB,WAAW,EACX,mBAAmB,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,WAAW,EAAE,EAChE,CAAC,MAAM,EAAE,aAAa,CAAC,EACvB,CAAC,SAAS,CAAC,CACZ,CAAC;QACJ,CAAC;QAED,+CAA+C;QAC/C,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QACjD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;YACrC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CACzB,MAAM,EACN,SAAS,IAAI,CAAC,IAAI,iBAAiB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,GAAG,CAAC,IAAI,EACxE,CAAC,MAAM,EAAE,UAAU,CAAC,EACpB,EAAE,CACH,CAAC;QACJ,CAAC;QAED,+CAA+C;QAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC9D,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QAC5C,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CACzB,MAAM,EACN,0BAA0B,OAAO,CAAC,YAAY,eAAe,UAAU,CAAC,eAAe,iCAAiC,QAAQ,EAAE,EAClI,CAAC,aAAa,EAAE,SAAS,CAAC,EAC1B,CAAC,SAAS,CAAC,CACZ,CAAC;QAEF,sDAAsD;QACtD,IAAI,aAAa,EAAE,CAAC;YAClB,IAAI,CAAC,cAAc,CAAC,SAAS,EAAE,aAAa,CAAC,CAAC;QAChD,CAAC;QAED,uCAAuC;QACvC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,gBAAgB,EAAE,CAAC;QAExC,8CAA8C;QAC9C,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,YAAY,CAAC,SAAS,CAAC,CAAC;QAEhD,wCAAwC;QACxC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,aAAa,EAAE,CAAC,aAAa,CAAC,EAAE,EAAE,CAAC,CAAC;QAExE,iDAAiD;QACjD,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;IAC9B,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,SAAiB,EAAE,aAAqB;QAC7D,IAAI,CAAC;YACH,MAAM,EAAE,cAAc,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;YAC7B,MAAM,OAAO,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YACvD,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;YAClD,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC;gBAAE,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAEhE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,OAAO,KAAK,CAAC,CAAC;YACrD,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;YAC9D,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;YAC5C,MAAM,YAAY,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,oBAAoB,CAAC,SAAS,CAAC,CAAC;YAE7E,MAAM,MAAM,GAAG;gBACb,EAAE;gBACF,8BAA8B;gBAC9B,EAAE;gBACF,aAAa,OAAO,CAAC,YAAY,cAAc,OAAO,CAAC,SAAS,CAAC,OAAO,IAAI,KAAK,cAAc,OAAO,CAAC,SAAS,CAAC,OAAO,IAAI,KAAK,eAAe,OAAO,CAAC,SAAS,CAAC,QAAQ,IAAI,KAAK,MAAM;gBACzL,WAAW,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,eAAe,UAAU,CAAC,eAAe,eAAe,YAAY,EAAE;gBAC7H,OAAO,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;gBAC3E,OAAO,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;gBAC5E,EAAE;aACH,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAE7B,MAAM,QAAQ,GAAG,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAC7F,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,mBAAmB,CAAC,EAAE,CAAC;gBAC5C,cAAc,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;YAC7C,CAAC;QACH,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO,CAAC,IAAI,CAAC,uCAAuC,EAAE,CAAC,CAAC,CAAC;QAC3D,CAAC;IACH,CAAC;IAED;;OAEG;IACK,uBAAuB,CAAC,MAA+B,EAAE,SAAiB;QAChF,IAAI,CAAC;YACH,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAClD,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,wBAAwB,CAAC,CAAC;YAClE,IAAI,CAAC,OAAO;gBAAE,OAAO;YAErB,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,OAAO,EAAE,CAAC;gBACpC,IAAI,KAAK,CAAC,UAAU;oBAAE,SAAS;gBAC/B,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;gBAC/C,6DAA6D;gBAC7D,IAAI,SAAS,CAAC,MAAM,GAAG,EAAE,IAAI,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC;oBAC5E,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,oBAAoB,CAAC,IAAI,CAAC,CAAC;oBACnD,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,kBAAkB,EAAE,SAAS,EAAE,IAAI,EAAE,GAAG,EAAE,kBAAkB,CAAC,CAAC;gBAC/F,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,mDAAmD;QACrD,CAAC;IACH,CAAC;IAED;;OAEG;IACH,aAAa;QACX,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;IAClC,CAAC;IAED;;OAEG;IACH,UAAU,CAAC,SAAkB;QAC3B,OAAO,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;IACvD,CAAC;IAED;;OAEG;IACH,YAAY;QACV,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;QAClC,MAAM,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QAEnC,MAAM,KAAK,GAAG;YACZ,+BAA+B;YAC/B,EAAE;YACF,kBAAkB,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,QAAQ,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE;YAClG,qBAAqB,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,MAAM,CAAC,SAAS,eAAe;YACjF,EAAE;YACF,mBAAmB;YACnB,eAAe,OAAO,CAAC,SAAS,CAAC,OAAO,MAAM;YAC9C,eAAe,OAAO,CAAC,SAAS,CAAC,OAAO,MAAM;YAC9C,eAAe,OAAO,CAAC,SAAS,CAAC,QAAQ,MAAM;YAC/C,iBAAiB,OAAO,CAAC,SAAS,CAAC,gBAAgB,MAAM;YACzD,EAAE;YACF,eAAe;YACf,UAAU,KAAK,CAAC,eAAe,gBAAgB,KAAK,CAAC,cAAc,iBAAiB,KAAK,CAAC,WAAW,EAAE;SACxG,CAAC;QAEF,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChC,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,qBAAqB,CAAC,CAAC;YACtC,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,QAAQ;gBAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACzD,CAAC;QAED,IAAI,OAAO,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,qBAAqB,CAAC,CAAC;YACtC,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,SAAS;gBAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAC1D,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;CACF;AA/bD,0BA+bC"}
@@ -1,102 +0,0 @@
1
- import { PreExecMetrics, RuntimeMetrics, PostExecMetrics } from '../types';
2
- import { SchemaGate } from '../guard/schema-gate';
3
- import { RiskGate } from '../guard/risk-gate';
4
- import { WorkingMemory } from '../memory/working';
5
- /**
6
- * PreExecEvaluator — captures metrics before tool execution.
7
- *
8
- * Watches the Guard layer output and WorkingMemory context
9
- * to score parameter quality, context utilization, and risk.
10
- */
11
- export declare class PreExecEvaluator {
12
- private schemaGate;
13
- private riskGate;
14
- private workingMemory;
15
- constructor(schemaGate: SchemaGate, riskGate: RiskGate, workingMemory: WorkingMemory);
16
- /**
17
- * Evaluate a tool call before execution.
18
- */
19
- evaluate(toolName: string, parameters: Record<string, unknown>): PreExecMetrics;
20
- /**
21
- * Score parameter quality based on contextual awareness.
22
- *
23
- * High quality: path contains session-relevant project paths,
24
- * content references open files, etc.
25
- * Low quality: bare strings, random-looking paths, missing files.
26
- */
27
- private evaluateParamQuality;
28
- /**
29
- * Score how well the agent uses stored context.
30
- */
31
- private evaluateContextUtilization;
32
- }
33
- /**
34
- * RuntimeEvaluator — captures metrics during execution.
35
- *
36
- * Tracks retries, self-corrections, timeouts, and
37
- * whether the agent selected the right tool for the job.
38
- */
39
- export declare class RuntimeEvaluator {
40
- /** Historical tool selection patterns — toolName -> successful scenario count */
41
- private toolHistory;
42
- /**
43
- * Evaluate a completed tool execution.
44
- */
45
- evaluate(options: {
46
- toolName: string;
47
- startTime: number;
48
- endTime: number;
49
- retryCount: number;
50
- wasSelfCorrected: boolean;
51
- hadTimeout: boolean;
52
- expectedTool?: string;
53
- toolResult: unknown;
54
- }): RuntimeMetrics;
55
- /** Record a tool call in the history tracker */
56
- private recordToolCall;
57
- /** Get tool selection accuracy statistics */
58
- getToolAccuracy(): Record<string, {
59
- calls: number;
60
- successRate: number;
61
- }>;
62
- }
63
- /**
64
- * PostExecEvaluator — captures metrics after execution.
65
- *
66
- * Scores verify results, user acceptance patterns,
67
- * and checks if the agent actually used its own result later.
68
- */
69
- export declare class PostExecEvaluator {
70
- /** Track result references for utilization scoring */
71
- private resultReferenceTracker;
72
- /**
73
- * Evaluate post-execution outcomes.
74
- */
75
- evaluate(options: {
76
- verifyPassed: boolean;
77
- verifyChecks: number;
78
- verifyFailures: number;
79
- userAccepted: boolean;
80
- userProvidedEdit: boolean;
81
- resultWasUsed: boolean;
82
- diffLinesChanged?: number;
83
- }): PostExecMetrics;
84
- /**
85
- * Track a tool result for later utilization detection.
86
- * Call this after each tool execution.
87
- */
88
- trackResult(operationId: string, result: unknown): void;
89
- /**
90
- * Mark a previously-tracked result as referenced (used by the agent later).
91
- */
92
- markResultReferenced(operationId: string): void;
93
- /**
94
- * Check if a result has been utilized by the agent.
95
- */
96
- isResultReferenced(operationId: string): boolean;
97
- /**
98
- * Get overall result utilization rate.
99
- */
100
- getUtilizationRate(): number;
101
- }
102
- //# sourceMappingURL=exec-evaluator.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"exec-evaluator.d.ts","sourceRoot":"","sources":["../../src/evaluator/exec-evaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,cAAc,EACd,cAAc,EACd,eAAe,EAGhB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAElD;;;;;GAKG;AACH,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,UAAU,CAAa;IAC/B,OAAO,CAAC,QAAQ,CAAW;IAC3B,OAAO,CAAC,aAAa,CAAgB;gBAGnC,UAAU,EAAE,UAAU,EACtB,QAAQ,EAAE,QAAQ,EAClB,aAAa,EAAE,aAAa;IAO9B;;OAEG;IACH,QAAQ,CACN,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAClC,cAAc;IAuBjB;;;;;;OAMG;IACH,OAAO,CAAC,oBAAoB;IA+C5B;;OAEG;IACH,OAAO,CAAC,0BAA0B;CAqCnC;AAED;;;;;GAKG;AACH,qBAAa,gBAAgB;IAC3B,iFAAiF;IACjF,OAAO,CAAC,WAAW,CAAgE;IAEnF;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,MAAM,CAAC;QACnB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,UAAU,EAAE,OAAO,CAAC;QACpB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,UAAU,EAAE,OAAO,CAAC;KACrB,GAAG,cAAc;IA0ClB,gDAAgD;IAChD,OAAO,CAAC,cAAc;IAUtB,6CAA6C;IAC7C,eAAe,IAAI,MAAM,CAAC,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;CAY1E;AAED;;;;;GAKG;AACH,qBAAa,iBAAiB;IAC5B,sDAAsD;IACtD,OAAO,CAAC,sBAAsB,CAAoE;IAElG;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE;QAChB,YAAY,EAAE,OAAO,CAAC;QACtB,YAAY,EAAE,MAAM,CAAC;QACrB,cAAc,EAAE,MAAM,CAAC;QACvB,YAAY,EAAE,OAAO,CAAC;QACtB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,aAAa,EAAE,OAAO,CAAC;QACvB,gBAAgB,CAAC,EAAE,MAAM,CAAC;KAC3B,GAAG,eAAe;IA+BnB;;;OAGG;IACH,WAAW,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,GAAG,IAAI;IAIvD;;OAEG;IACH,oBAAoB,CAAC,WAAW,EAAE,MAAM,GAAG,IAAI;IAK/C;;OAEG;IACH,kBAAkB,CAAC,WAAW,EAAE,MAAM,GAAG,OAAO;IAIhD;;OAEG;IACH,kBAAkB,IAAI,MAAM;CAM7B"}