@cogitator-ai/self-modifying 0.1.0
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.
- package/LICENSE +21 -0
- package/README.md +714 -0
- package/dist/architecture-evolution/capability-analyzer.d.ts +32 -0
- package/dist/architecture-evolution/capability-analyzer.d.ts.map +1 -0
- package/dist/architecture-evolution/capability-analyzer.js +264 -0
- package/dist/architecture-evolution/capability-analyzer.js.map +1 -0
- package/dist/architecture-evolution/evolution-strategy.d.ts +29 -0
- package/dist/architecture-evolution/evolution-strategy.d.ts.map +1 -0
- package/dist/architecture-evolution/evolution-strategy.js +176 -0
- package/dist/architecture-evolution/evolution-strategy.js.map +1 -0
- package/dist/architecture-evolution/index.d.ts +5 -0
- package/dist/architecture-evolution/index.d.ts.map +1 -0
- package/dist/architecture-evolution/index.js +5 -0
- package/dist/architecture-evolution/index.js.map +1 -0
- package/dist/architecture-evolution/parameter-optimizer.d.ts +67 -0
- package/dist/architecture-evolution/parameter-optimizer.d.ts.map +1 -0
- package/dist/architecture-evolution/parameter-optimizer.js +341 -0
- package/dist/architecture-evolution/parameter-optimizer.js.map +1 -0
- package/dist/architecture-evolution/prompts.d.ts +33 -0
- package/dist/architecture-evolution/prompts.d.ts.map +1 -0
- package/dist/architecture-evolution/prompts.js +169 -0
- package/dist/architecture-evolution/prompts.js.map +1 -0
- package/dist/constraints/index.d.ts +4 -0
- package/dist/constraints/index.d.ts.map +1 -0
- package/dist/constraints/index.js +4 -0
- package/dist/constraints/index.js.map +1 -0
- package/dist/constraints/modification-validator.d.ts +26 -0
- package/dist/constraints/modification-validator.d.ts.map +1 -0
- package/dist/constraints/modification-validator.js +313 -0
- package/dist/constraints/modification-validator.js.map +1 -0
- package/dist/constraints/rollback-manager.d.ts +52 -0
- package/dist/constraints/rollback-manager.d.ts.map +1 -0
- package/dist/constraints/rollback-manager.js +113 -0
- package/dist/constraints/rollback-manager.js.map +1 -0
- package/dist/constraints/safety-constraints.d.ts +11 -0
- package/dist/constraints/safety-constraints.d.ts.map +1 -0
- package/dist/constraints/safety-constraints.js +78 -0
- package/dist/constraints/safety-constraints.js.map +1 -0
- package/dist/events/event-emitter.d.ts +12 -0
- package/dist/events/event-emitter.d.ts.map +1 -0
- package/dist/events/event-emitter.js +43 -0
- package/dist/events/event-emitter.js.map +1 -0
- package/dist/events/index.d.ts +2 -0
- package/dist/events/index.d.ts.map +1 -0
- package/dist/events/index.js +2 -0
- package/dist/events/index.js.map +1 -0
- package/dist/index.d.ts +8 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +7 -0
- package/dist/index.js.map +1 -0
- package/dist/meta-reasoning/index.d.ts +5 -0
- package/dist/meta-reasoning/index.d.ts.map +1 -0
- package/dist/meta-reasoning/index.js +5 -0
- package/dist/meta-reasoning/index.js.map +1 -0
- package/dist/meta-reasoning/meta-reasoner.d.ts +53 -0
- package/dist/meta-reasoning/meta-reasoner.d.ts.map +1 -0
- package/dist/meta-reasoning/meta-reasoner.js +261 -0
- package/dist/meta-reasoning/meta-reasoner.js.map +1 -0
- package/dist/meta-reasoning/observation-collector.d.ts +37 -0
- package/dist/meta-reasoning/observation-collector.d.ts.map +1 -0
- package/dist/meta-reasoning/observation-collector.js +123 -0
- package/dist/meta-reasoning/observation-collector.js.map +1 -0
- package/dist/meta-reasoning/prompts.d.ts +31 -0
- package/dist/meta-reasoning/prompts.d.ts.map +1 -0
- package/dist/meta-reasoning/prompts.js +96 -0
- package/dist/meta-reasoning/prompts.js.map +1 -0
- package/dist/meta-reasoning/strategy-selector.d.ts +27 -0
- package/dist/meta-reasoning/strategy-selector.d.ts.map +1 -0
- package/dist/meta-reasoning/strategy-selector.js +138 -0
- package/dist/meta-reasoning/strategy-selector.js.map +1 -0
- package/dist/self-modifying-agent.d.ts +61 -0
- package/dist/self-modifying-agent.d.ts.map +1 -0
- package/dist/self-modifying-agent.js +449 -0
- package/dist/self-modifying-agent.js.map +1 -0
- package/dist/tool-generation/gap-analyzer.d.ts +25 -0
- package/dist/tool-generation/gap-analyzer.d.ts.map +1 -0
- package/dist/tool-generation/gap-analyzer.js +153 -0
- package/dist/tool-generation/gap-analyzer.js.map +1 -0
- package/dist/tool-generation/generated-tool-store.d.ts +51 -0
- package/dist/tool-generation/generated-tool-store.d.ts.map +1 -0
- package/dist/tool-generation/generated-tool-store.js +195 -0
- package/dist/tool-generation/generated-tool-store.js.map +1 -0
- package/dist/tool-generation/index.d.ts +7 -0
- package/dist/tool-generation/index.d.ts.map +1 -0
- package/dist/tool-generation/index.js +7 -0
- package/dist/tool-generation/index.js.map +1 -0
- package/dist/tool-generation/prompts.d.ts +28 -0
- package/dist/tool-generation/prompts.d.ts.map +1 -0
- package/dist/tool-generation/prompts.js +269 -0
- package/dist/tool-generation/prompts.js.map +1 -0
- package/dist/tool-generation/tool-generator.d.ts +29 -0
- package/dist/tool-generation/tool-generator.d.ts.map +1 -0
- package/dist/tool-generation/tool-generator.js +169 -0
- package/dist/tool-generation/tool-generator.js.map +1 -0
- package/dist/tool-generation/tool-sandbox.d.ts +31 -0
- package/dist/tool-generation/tool-sandbox.d.ts.map +1 -0
- package/dist/tool-generation/tool-sandbox.js +240 -0
- package/dist/tool-generation/tool-sandbox.js.map +1 -0
- package/dist/tool-generation/tool-validator.d.ts +32 -0
- package/dist/tool-generation/tool-validator.d.ts.map +1 -0
- package/dist/tool-generation/tool-validator.js +304 -0
- package/dist/tool-generation/tool-validator.js.map +1 -0
- package/dist/utils/index.d.ts +2 -0
- package/dist/utils/index.d.ts.map +1 -0
- package/dist/utils/index.js +2 -0
- package/dist/utils/index.js.map +1 -0
- package/dist/utils/llm-helper.d.ts +6 -0
- package/dist/utils/llm-helper.d.ts.map +1 -0
- package/dist/utils/llm-helper.js +18 -0
- package/dist/utils/llm-helper.js.map +1 -0
- package/package.json +61 -0
- package/src/__tests__/architecture-evolution.test.ts +368 -0
- package/src/__tests__/constraints.test.ts +266 -0
- package/src/__tests__/index.test.ts +99 -0
- package/src/__tests__/meta-reasoning.test.ts +343 -0
- package/src/__tests__/tool-generation.test.ts +455 -0
- package/src/architecture-evolution/capability-analyzer.ts +337 -0
- package/src/architecture-evolution/evolution-strategy.ts +224 -0
- package/src/architecture-evolution/index.ts +26 -0
- package/src/architecture-evolution/parameter-optimizer.ts +489 -0
- package/src/architecture-evolution/prompts.ts +216 -0
- package/src/constraints/index.ts +23 -0
- package/src/constraints/modification-validator.ts +402 -0
- package/src/constraints/rollback-manager.ts +173 -0
- package/src/constraints/safety-constraints.ts +103 -0
- package/src/events/event-emitter.ts +62 -0
- package/src/events/index.ts +1 -0
- package/src/index.ts +112 -0
- package/src/meta-reasoning/index.ts +24 -0
- package/src/meta-reasoning/meta-reasoner.ts +381 -0
- package/src/meta-reasoning/observation-collector.ts +161 -0
- package/src/meta-reasoning/prompts.ts +131 -0
- package/src/meta-reasoning/strategy-selector.ts +179 -0
- package/src/self-modifying-agent.ts +585 -0
- package/src/tool-generation/gap-analyzer.ts +234 -0
- package/src/tool-generation/generated-tool-store.ts +268 -0
- package/src/tool-generation/index.ts +19 -0
- package/src/tool-generation/prompts.ts +308 -0
- package/src/tool-generation/tool-generator.ts +243 -0
- package/src/tool-generation/tool-sandbox.ts +332 -0
- package/src/tool-generation/tool-validator.ts +365 -0
- package/src/utils/index.ts +1 -0
- package/src/utils/llm-helper.ts +24 -0
|
@@ -0,0 +1,169 @@
|
|
|
1
|
+
import { ToolValidator } from './tool-validator';
|
|
2
|
+
import { TOOL_GENERATION_SYSTEM_PROMPT, buildToolGenerationPrompt, buildToolImprovementPrompt, parseToolGenerationResponse, } from './prompts';
|
|
3
|
+
export class ToolGenerator {
|
|
4
|
+
llm;
|
|
5
|
+
config;
|
|
6
|
+
validator;
|
|
7
|
+
constructor(options) {
|
|
8
|
+
this.llm = options.llm;
|
|
9
|
+
this.config = options.config;
|
|
10
|
+
this.validator = new ToolValidator({
|
|
11
|
+
llm: options.llm,
|
|
12
|
+
config: options.config,
|
|
13
|
+
});
|
|
14
|
+
}
|
|
15
|
+
async generate(gap, existingTools, testCases) {
|
|
16
|
+
let currentTool = null;
|
|
17
|
+
let validationResult = null;
|
|
18
|
+
let iterations = 0;
|
|
19
|
+
const maxIterations = this.config.maxIterationsPerTool || 3;
|
|
20
|
+
while (iterations < maxIterations) {
|
|
21
|
+
iterations++;
|
|
22
|
+
try {
|
|
23
|
+
if (currentTool === null) {
|
|
24
|
+
currentTool = await this.generateInitial(gap, existingTools);
|
|
25
|
+
}
|
|
26
|
+
else if (validationResult) {
|
|
27
|
+
currentTool = await this.improve(currentTool, validationResult, iterations);
|
|
28
|
+
}
|
|
29
|
+
if (!currentTool) {
|
|
30
|
+
return {
|
|
31
|
+
tool: null,
|
|
32
|
+
validationResult: null,
|
|
33
|
+
iterations,
|
|
34
|
+
success: false,
|
|
35
|
+
error: 'Failed to generate tool implementation',
|
|
36
|
+
};
|
|
37
|
+
}
|
|
38
|
+
validationResult = await this.validator.validate(currentTool, testCases);
|
|
39
|
+
if (validationResult.isValid) {
|
|
40
|
+
currentTool.status = 'validated';
|
|
41
|
+
currentTool.validationScore = validationResult.overallScore;
|
|
42
|
+
return {
|
|
43
|
+
tool: currentTool,
|
|
44
|
+
validationResult,
|
|
45
|
+
iterations,
|
|
46
|
+
success: true,
|
|
47
|
+
};
|
|
48
|
+
}
|
|
49
|
+
if (validationResult.securityIssues.length > 0 && iterations >= 2) {
|
|
50
|
+
return {
|
|
51
|
+
tool: currentTool,
|
|
52
|
+
validationResult,
|
|
53
|
+
iterations,
|
|
54
|
+
success: false,
|
|
55
|
+
error: `Security issues persist after ${iterations} iterations: ${validationResult.securityIssues.join(', ')}`,
|
|
56
|
+
};
|
|
57
|
+
}
|
|
58
|
+
}
|
|
59
|
+
catch (error) {
|
|
60
|
+
return {
|
|
61
|
+
tool: currentTool,
|
|
62
|
+
validationResult,
|
|
63
|
+
iterations,
|
|
64
|
+
success: false,
|
|
65
|
+
error: error instanceof Error ? error.message : String(error),
|
|
66
|
+
};
|
|
67
|
+
}
|
|
68
|
+
}
|
|
69
|
+
return {
|
|
70
|
+
tool: currentTool,
|
|
71
|
+
validationResult,
|
|
72
|
+
iterations,
|
|
73
|
+
success: false,
|
|
74
|
+
error: `Failed to generate valid tool after ${maxIterations} iterations`,
|
|
75
|
+
};
|
|
76
|
+
}
|
|
77
|
+
async generateQuick(description, name, parameters) {
|
|
78
|
+
const gap = {
|
|
79
|
+
id: `quick_${Date.now()}`,
|
|
80
|
+
description,
|
|
81
|
+
requiredCapability: description,
|
|
82
|
+
suggestedToolName: name,
|
|
83
|
+
complexity: 'simple',
|
|
84
|
+
confidence: 1,
|
|
85
|
+
reasoning: 'User-requested quick generation',
|
|
86
|
+
};
|
|
87
|
+
const result = await this.generate(gap, []);
|
|
88
|
+
return result.tool;
|
|
89
|
+
}
|
|
90
|
+
async generateInitial(gap, existingTools) {
|
|
91
|
+
const toolSummaries = existingTools.map((t) => ({
|
|
92
|
+
name: t.name,
|
|
93
|
+
description: t.description,
|
|
94
|
+
}));
|
|
95
|
+
const prompt = buildToolGenerationPrompt(gap, toolSummaries, {
|
|
96
|
+
maxLines: 100,
|
|
97
|
+
securityLevel: 'strict',
|
|
98
|
+
allowedModules: this.config.sandboxConfig?.allowedModules,
|
|
99
|
+
});
|
|
100
|
+
const response = await this.callLLM([
|
|
101
|
+
{ role: 'system', content: TOOL_GENERATION_SYSTEM_PROMPT },
|
|
102
|
+
{ role: 'user', content: prompt },
|
|
103
|
+
], 0.4);
|
|
104
|
+
const tool = parseToolGenerationResponse(response.content);
|
|
105
|
+
if (tool) {
|
|
106
|
+
tool.metadata = {
|
|
107
|
+
...tool.metadata,
|
|
108
|
+
gapId: gap.id,
|
|
109
|
+
complexity: gap.complexity,
|
|
110
|
+
};
|
|
111
|
+
}
|
|
112
|
+
return tool;
|
|
113
|
+
}
|
|
114
|
+
async improve(tool, validationResult, iteration) {
|
|
115
|
+
const prompt = buildToolImprovementPrompt(tool, validationResult, iteration);
|
|
116
|
+
const response = await this.callLLM([
|
|
117
|
+
{ role: 'system', content: TOOL_GENERATION_SYSTEM_PROMPT },
|
|
118
|
+
{ role: 'user', content: prompt },
|
|
119
|
+
], 0.3);
|
|
120
|
+
const improved = parseToolGenerationResponse(response.content);
|
|
121
|
+
if (improved) {
|
|
122
|
+
improved.id = tool.id;
|
|
123
|
+
improved.version = tool.version + 1;
|
|
124
|
+
improved.metadata = {
|
|
125
|
+
...tool.metadata,
|
|
126
|
+
...improved.metadata,
|
|
127
|
+
previousVersion: tool.version,
|
|
128
|
+
improvementIteration: iteration,
|
|
129
|
+
};
|
|
130
|
+
}
|
|
131
|
+
return improved;
|
|
132
|
+
}
|
|
133
|
+
createExecutableTool(generated) {
|
|
134
|
+
const execute = this.compileImplementation(generated.implementation);
|
|
135
|
+
const { z } = require('zod');
|
|
136
|
+
return {
|
|
137
|
+
name: generated.name,
|
|
138
|
+
description: generated.description,
|
|
139
|
+
parameters: z.record(z.unknown()),
|
|
140
|
+
execute,
|
|
141
|
+
toJSON: () => ({
|
|
142
|
+
name: generated.name,
|
|
143
|
+
description: generated.description,
|
|
144
|
+
parameters: {
|
|
145
|
+
type: 'object',
|
|
146
|
+
properties: generated.parameters,
|
|
147
|
+
},
|
|
148
|
+
}),
|
|
149
|
+
};
|
|
150
|
+
}
|
|
151
|
+
compileImplementation(implementation) {
|
|
152
|
+
return async (params) => {
|
|
153
|
+
const factory = new Function(`
|
|
154
|
+
"use strict";
|
|
155
|
+
${implementation}
|
|
156
|
+
return execute;
|
|
157
|
+
`);
|
|
158
|
+
const execute = factory();
|
|
159
|
+
return execute(params);
|
|
160
|
+
};
|
|
161
|
+
}
|
|
162
|
+
async callLLM(messages, temperature) {
|
|
163
|
+
if (this.llm.complete) {
|
|
164
|
+
return this.llm.complete({ messages, temperature });
|
|
165
|
+
}
|
|
166
|
+
return this.llm.chat({ model: 'default', messages, temperature });
|
|
167
|
+
}
|
|
168
|
+
}
|
|
169
|
+
//# sourceMappingURL=tool-generator.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tool-generator.js","sourceRoot":"","sources":["../../src/tool-generation/tool-generator.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACjD,OAAO,EACL,6BAA6B,EAC7B,yBAAyB,EACzB,0BAA0B,EAC1B,2BAA2B,GAC5B,MAAM,WAAW,CAAC;AAenB,MAAM,OAAO,aAAa;IACP,GAAG,CAAa;IAChB,MAAM,CAA2B;IACjC,SAAS,CAAgB;IAE1C,YAAY,OAA6B;QACvC,IAAI,CAAC,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC;QACvB,IAAI,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;QAC7B,IAAI,CAAC,SAAS,GAAG,IAAI,aAAa,CAAC;YACjC,GAAG,EAAE,OAAO,CAAC,GAAG;YAChB,MAAM,EAAE,OAAO,CAAC,MAAM;SACvB,CAAC,CAAC;IACL,CAAC;IAED,KAAK,CAAC,QAAQ,CACZ,GAAkB,EAClB,aAAqB,EACrB,SAA+D;QAE/D,IAAI,WAAW,GAAyB,IAAI,CAAC;QAC7C,IAAI,gBAAgB,GAAgC,IAAI,CAAC;QACzD,IAAI,UAAU,GAAG,CAAC,CAAC;QAEnB,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,oBAAoB,IAAI,CAAC,CAAC;QAE5D,OAAO,UAAU,GAAG,aAAa,EAAE,CAAC;YAClC,UAAU,EAAE,CAAC;YAEb,IAAI,CAAC;gBACH,IAAI,WAAW,KAAK,IAAI,EAAE,CAAC;oBACzB,WAAW,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;gBAC/D,CAAC;qBAAM,IAAI,gBAAgB,EAAE,CAAC;oBAC5B,WAAW,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE,gBAAgB,EAAE,UAAU,CAAC,CAAC;gBAC9E,CAAC;gBAED,IAAI,CAAC,WAAW,EAAE,CAAC;oBACjB,OAAO;wBACL,IAAI,EAAE,IAAI;wBACV,gBAAgB,EAAE,IAAI;wBACtB,UAAU;wBACV,OAAO,EAAE,KAAK;wBACd,KAAK,EAAE,wCAAwC;qBAChD,CAAC;gBACJ,CAAC;gBAED,gBAAgB,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;gBAEzE,IAAI,gBAAgB,CAAC,OAAO,EAAE,CAAC;oBAC7B,WAAW,CAAC,MAAM,GAAG,WAAW,CAAC;oBACjC,WAAW,CAAC,eAAe,GAAG,gBAAgB,CAAC,YAAY,CAAC;oBAE5D,OAAO;wBACL,IAAI,EAAE,WAAW;wBACjB,gBAAgB;wBAChB,UAAU;wBACV,OAAO,EAAE,IAAI;qBACd,CAAC;gBACJ,CAAC;gBAED,IAAI,gBAAgB,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,IAAI,UAAU,IAAI,CAAC,EAAE,CAAC;oBAClE,OAAO;wBACL,IAAI,EAAE,WAAW;wBACjB,gBAAgB;wBAChB,UAAU;wBACV,OAAO,EAAE,KAAK;wBACd,KAAK,EAAE,iCAAiC,UAAU,gBAAgB,gBAAgB,CAAC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;qBAC/G,CAAC;gBACJ,CAAC;YACH,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,OAAO;oBACL,IAAI,EAAE,WAAW;oBACjB,gBAAgB;oBAChB,UAAU;oBACV,OAAO,EAAE,KAAK;oBACd,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;iBAC9D,CAAC;YACJ,CAAC;QACH,CAAC;QAED,OAAO;YACL,IAAI,EAAE,WAAW;YACjB,gBAAgB;YAChB,UAAU;YACV,OAAO,EAAE,KAAK;YACd,KAAK,EAAE,uCAAuC,aAAa,aAAa;SACzE,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,aAAa,CACjB,WAAmB,EACnB,IAAY,EACZ,UAAmC;QAEnC,MAAM,GAAG,GAAkB;YACzB,EAAE,EAAE,SAAS,IAAI,CAAC,GAAG,EAAE,EAAE;YACzB,WAAW;YACX,kBAAkB,EAAE,WAAW;YAC/B,iBAAiB,EAAE,IAAI;YACvB,UAAU,EAAE,QAAQ;YACpB,UAAU,EAAE,CAAC;YACb,SAAS,EAAE,iCAAiC;SAC7C,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;QAC5C,OAAO,MAAM,CAAC,IAAI,CAAC;IACrB,CAAC;IAEO,KAAK,CAAC,eAAe,CAC3B,GAAkB,EAClB,aAAqB;QAErB,MAAM,aAAa,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC9C,IAAI,EAAE,CAAC,CAAC,IAAI;YACZ,WAAW,EAAE,CAAC,CAAC,WAAW;SAC3B,CAAC,CAAC,CAAC;QAEJ,MAAM,MAAM,GAAG,yBAAyB,CAAC,GAAG,EAAE,aAAa,EAAE;YAC3D,QAAQ,EAAE,GAAG;YACb,aAAa,EAAE,QAAQ;YACvB,cAAc,EAAE,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,cAAc;SAC1D,CAAC,CAAC;QAEH,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC;YAClC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,6BAA6B,EAAE;YAC1D,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE;SAClC,EAAE,GAAG,CAAC,CAAC;QAER,MAAM,IAAI,GAAG,2BAA2B,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAE3D,IAAI,IAAI,EAAE,CAAC;YACT,IAAI,CAAC,QAAQ,GAAG;gBACd,GAAG,IAAI,CAAC,QAAQ;gBAChB,KAAK,EAAE,GAAG,CAAC,EAAE;gBACb,UAAU,EAAE,GAAG,CAAC,UAAU;aAC3B,CAAC;QACJ,CAAC;QAED,OAAO,IAAI,CAAC;IACd,CAAC;IAEO,KAAK,CAAC,OAAO,CACnB,IAAmB,EACnB,gBAAsC,EACtC,SAAiB;QAEjB,MAAM,MAAM,GAAG,0BAA0B,CAAC,IAAI,EAAE,gBAAgB,EAAE,SAAS,CAAC,CAAC;QAE7E,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC;YAClC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,6BAA6B,EAAE;YAC1D,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE;SAClC,EAAE,GAAG,CAAC,CAAC;QAER,MAAM,QAAQ,GAAG,2BAA2B,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAE/D,IAAI,QAAQ,EAAE,CAAC;YACb,QAAQ,CAAC,EAAE,GAAG,IAAI,CAAC,EAAE,CAAC;YACtB,QAAQ,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC;YACpC,QAAQ,CAAC,QAAQ,GAAG;gBAClB,GAAG,IAAI,CAAC,QAAQ;gBAChB,GAAG,QAAQ,CAAC,QAAQ;gBACpB,eAAe,EAAE,IAAI,CAAC,OAAO;gBAC7B,oBAAoB,EAAE,SAAS;aAChC,CAAC;QACJ,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED,oBAAoB,CAAC,SAAwB;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,qBAAqB,CAAC,SAAS,CAAC,cAAc,CAAC,CAAC;QACrE,MAAM,EAAE,CAAC,EAAE,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC;QAE7B,OAAO;YACL,IAAI,EAAE,SAAS,CAAC,IAAI;YACpB,WAAW,EAAE,SAAS,CAAC,WAAW;YAClC,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,EAAE,CAAqB;YACrD,OAAO;YACP,MAAM,EAAE,GAAG,EAAE,CAAC,CAAC;gBACb,IAAI,EAAE,SAAS,CAAC,IAAI;gBACpB,WAAW,EAAE,SAAS,CAAC,WAAW;gBAClC,UAAU,EAAE;oBACV,IAAI,EAAE,QAAiB;oBACvB,UAAU,EAAE,SAAS,CAAC,UAAqC;iBAC5D;aACF,CAAC;SACH,CAAC;IACJ,CAAC;IAEO,qBAAqB,CAC3B,cAAsB;QAEtB,OAAO,KAAK,EAAE,MAAe,EAAoB,EAAE;YACjD,MAAM,OAAO,GAAG,IAAI,QAAQ,CAAC;;UAEzB,cAAc;;OAEjB,CAAC,CAAC;YAEH,MAAM,OAAO,GAAG,OAAO,EAAE,CAAC;YAC1B,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC;QACzB,CAAC,CAAC;IACJ,CAAC;IAEO,KAAK,CAAC,OAAO,CACnB,QAA2E,EAC3E,WAAmB;QAEnB,IAAI,IAAI,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC;YACtB,OAAO,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,CAAC,CAAC;QACtD,CAAC;QACD,OAAO,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,WAAW,EAAE,CAAC,CAAC;IACpE,CAAC;CACF"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
import type { ToolSandboxConfig, ToolSandboxResult, GeneratedTool } from '@cogitator-ai/types';
|
|
2
|
+
export declare const DEFAULT_SANDBOX_CONFIG: ToolSandboxConfig;
|
|
3
|
+
export declare class ToolSandbox {
|
|
4
|
+
private readonly config;
|
|
5
|
+
private readonly logs;
|
|
6
|
+
constructor(config?: Partial<ToolSandboxConfig>);
|
|
7
|
+
execute(tool: GeneratedTool, params: unknown): Promise<ToolSandboxResult>;
|
|
8
|
+
testWithCases(tool: GeneratedTool, testCases: Array<{
|
|
9
|
+
input: unknown;
|
|
10
|
+
expectedOutput?: unknown;
|
|
11
|
+
shouldThrow?: boolean;
|
|
12
|
+
}>): Promise<{
|
|
13
|
+
passed: number;
|
|
14
|
+
failed: number;
|
|
15
|
+
results: Array<{
|
|
16
|
+
input: unknown;
|
|
17
|
+
output?: unknown;
|
|
18
|
+
error?: string;
|
|
19
|
+
passed: boolean;
|
|
20
|
+
executionTime: number;
|
|
21
|
+
}>;
|
|
22
|
+
}>;
|
|
23
|
+
private validateImplementation;
|
|
24
|
+
private createContext;
|
|
25
|
+
private wrapImplementation;
|
|
26
|
+
private executeWithTimeout;
|
|
27
|
+
private executeUnsandboxed;
|
|
28
|
+
private estimateMemoryUsage;
|
|
29
|
+
private deepEqual;
|
|
30
|
+
}
|
|
31
|
+
//# sourceMappingURL=tool-sandbox.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tool-sandbox.d.ts","sourceRoot":"","sources":["../../src/tool-generation/tool-sandbox.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,iBAAiB,EACjB,iBAAiB,EACjB,aAAa,EACd,MAAM,qBAAqB,CAAC;AAE7B,eAAO,MAAM,sBAAsB,EAAE,iBAMpC,CAAC;AAyBF,qBAAa,WAAW;IACtB,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAoB;IAC3C,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAgB;gBAEzB,MAAM,GAAE,OAAO,CAAC,iBAAiB,CAAM;IAI7C,OAAO,CACX,IAAI,EAAE,aAAa,EACnB,MAAM,EAAE,OAAO,GACd,OAAO,CAAC,iBAAiB,CAAC;IAuCvB,aAAa,CACjB,IAAI,EAAE,aAAa,EACnB,SAAS,EAAE,KAAK,CAAC;QAAE,KAAK,EAAE,OAAO,CAAC;QAAC,cAAc,CAAC,EAAE,OAAO,CAAC;QAAC,WAAW,CAAC,EAAE,OAAO,CAAA;KAAE,CAAC,GACpF,OAAO,CAAC;QACT,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,KAAK,CAAC;YACb,KAAK,EAAE,OAAO,CAAC;YACf,MAAM,CAAC,EAAE,OAAO,CAAC;YACjB,KAAK,CAAC,EAAE,MAAM,CAAC;YACf,MAAM,EAAE,OAAO,CAAC;YAChB,aAAa,EAAE,MAAM,CAAC;SACvB,CAAC,CAAC;KACJ,CAAC;IAuCF,OAAO,CAAC,sBAAsB;IAiD9B,OAAO,CAAC,aAAa;IAiCrB,OAAO,CAAC,kBAAkB;YAgBZ,kBAAkB;YA+BlB,kBAAkB;IAgChC,OAAO,CAAC,mBAAmB;IAS3B,OAAO,CAAC,SAAS;CAsBlB"}
|
|
@@ -0,0 +1,240 @@
|
|
|
1
|
+
export const DEFAULT_SANDBOX_CONFIG = {
|
|
2
|
+
enabled: true,
|
|
3
|
+
maxExecutionTime: 5000,
|
|
4
|
+
maxMemory: 50 * 1024 * 1024,
|
|
5
|
+
allowedModules: [],
|
|
6
|
+
isolationLevel: 'strict',
|
|
7
|
+
};
|
|
8
|
+
export class ToolSandbox {
|
|
9
|
+
config;
|
|
10
|
+
logs = [];
|
|
11
|
+
constructor(config = {}) {
|
|
12
|
+
this.config = { ...DEFAULT_SANDBOX_CONFIG, ...config };
|
|
13
|
+
}
|
|
14
|
+
async execute(tool, params) {
|
|
15
|
+
const startTime = Date.now();
|
|
16
|
+
this.logs.length = 0;
|
|
17
|
+
if (!this.config.enabled) {
|
|
18
|
+
return this.executeUnsandboxed(tool, params, startTime);
|
|
19
|
+
}
|
|
20
|
+
try {
|
|
21
|
+
this.validateImplementation(tool.implementation);
|
|
22
|
+
const context = this.createContext();
|
|
23
|
+
const wrappedCode = this.wrapImplementation(tool.implementation);
|
|
24
|
+
const result = await this.executeWithTimeout(wrappedCode, context, params, this.config.maxExecutionTime);
|
|
25
|
+
return {
|
|
26
|
+
success: true,
|
|
27
|
+
result,
|
|
28
|
+
executionTime: Date.now() - startTime,
|
|
29
|
+
memoryUsed: this.estimateMemoryUsage(result),
|
|
30
|
+
logs: [...this.logs],
|
|
31
|
+
};
|
|
32
|
+
}
|
|
33
|
+
catch (error) {
|
|
34
|
+
return {
|
|
35
|
+
success: false,
|
|
36
|
+
error: error instanceof Error ? error.message : String(error),
|
|
37
|
+
executionTime: Date.now() - startTime,
|
|
38
|
+
memoryUsed: 0,
|
|
39
|
+
logs: [...this.logs],
|
|
40
|
+
};
|
|
41
|
+
}
|
|
42
|
+
}
|
|
43
|
+
async testWithCases(tool, testCases) {
|
|
44
|
+
const results = [];
|
|
45
|
+
for (const testCase of testCases) {
|
|
46
|
+
const execResult = await this.execute(tool, testCase.input);
|
|
47
|
+
let passed = false;
|
|
48
|
+
if (testCase.shouldThrow) {
|
|
49
|
+
passed = !execResult.success;
|
|
50
|
+
}
|
|
51
|
+
else if (testCase.expectedOutput !== undefined) {
|
|
52
|
+
passed =
|
|
53
|
+
execResult.success &&
|
|
54
|
+
this.deepEqual(execResult.result, testCase.expectedOutput);
|
|
55
|
+
}
|
|
56
|
+
else {
|
|
57
|
+
passed = execResult.success;
|
|
58
|
+
}
|
|
59
|
+
results.push({
|
|
60
|
+
input: testCase.input,
|
|
61
|
+
output: execResult.result,
|
|
62
|
+
error: execResult.error,
|
|
63
|
+
passed,
|
|
64
|
+
executionTime: execResult.executionTime,
|
|
65
|
+
});
|
|
66
|
+
}
|
|
67
|
+
return {
|
|
68
|
+
passed: results.filter((r) => r.passed).length,
|
|
69
|
+
failed: results.filter((r) => !r.passed).length,
|
|
70
|
+
results,
|
|
71
|
+
};
|
|
72
|
+
}
|
|
73
|
+
validateImplementation(code) {
|
|
74
|
+
const forbidden = [
|
|
75
|
+
/\beval\s*\(/,
|
|
76
|
+
/\bFunction\s*\(/,
|
|
77
|
+
/\bnew\s+Function\s*\(/,
|
|
78
|
+
/\bimport\s*\(/,
|
|
79
|
+
/\brequire\s*\(/,
|
|
80
|
+
/\bprocess\./,
|
|
81
|
+
/\bglobal\./,
|
|
82
|
+
/\bglobalThis\./,
|
|
83
|
+
/\bwindow\./,
|
|
84
|
+
/\bdocument\./,
|
|
85
|
+
/\bchild_process/,
|
|
86
|
+
/\bfs\./,
|
|
87
|
+
/\bhttp\./,
|
|
88
|
+
/\bhttps\./,
|
|
89
|
+
/\bnet\./,
|
|
90
|
+
/\bdns\./,
|
|
91
|
+
/\bos\./,
|
|
92
|
+
/\bexec\s*\(/,
|
|
93
|
+
/\bspawn\s*\(/,
|
|
94
|
+
/__proto__/,
|
|
95
|
+
/\bconstructor\s*\[/,
|
|
96
|
+
];
|
|
97
|
+
if (this.config.isolationLevel === 'strict') {
|
|
98
|
+
forbidden.push(/\bfetch\s*\(/, /\bXMLHttpRequest/, /\bWebSocket/, /\bsetTimeout\s*\(/, /\bsetInterval\s*\(/);
|
|
99
|
+
}
|
|
100
|
+
for (const pattern of forbidden) {
|
|
101
|
+
if (pattern.test(code)) {
|
|
102
|
+
throw new Error(`Security violation: forbidden pattern detected - ${pattern.source}`);
|
|
103
|
+
}
|
|
104
|
+
}
|
|
105
|
+
const lines = code.split('\n').length;
|
|
106
|
+
if (lines > 200) {
|
|
107
|
+
throw new Error(`Implementation too large: ${lines} lines (max 200)`);
|
|
108
|
+
}
|
|
109
|
+
}
|
|
110
|
+
createContext() {
|
|
111
|
+
const sandbox = this;
|
|
112
|
+
return {
|
|
113
|
+
console: {
|
|
114
|
+
log: (...args) => {
|
|
115
|
+
sandbox.logs.push(`[LOG] ${args.map(String).join(' ')}`);
|
|
116
|
+
},
|
|
117
|
+
warn: (...args) => {
|
|
118
|
+
sandbox.logs.push(`[WARN] ${args.map(String).join(' ')}`);
|
|
119
|
+
},
|
|
120
|
+
error: (...args) => {
|
|
121
|
+
sandbox.logs.push(`[ERROR] ${args.map(String).join(' ')}`);
|
|
122
|
+
},
|
|
123
|
+
},
|
|
124
|
+
Math,
|
|
125
|
+
JSON,
|
|
126
|
+
Date,
|
|
127
|
+
Array,
|
|
128
|
+
Object,
|
|
129
|
+
String,
|
|
130
|
+
Number,
|
|
131
|
+
Boolean,
|
|
132
|
+
RegExp,
|
|
133
|
+
Map,
|
|
134
|
+
Set,
|
|
135
|
+
Promise,
|
|
136
|
+
Error,
|
|
137
|
+
TypeError,
|
|
138
|
+
RangeError,
|
|
139
|
+
};
|
|
140
|
+
}
|
|
141
|
+
wrapImplementation(code) {
|
|
142
|
+
return `
|
|
143
|
+
"use strict";
|
|
144
|
+
return (async function sandboxedExecution(params, context) {
|
|
145
|
+
const { console, Math, JSON, Date, Array, Object, String, Number, Boolean, RegExp, Map, Set, Promise, Error, TypeError, RangeError } = context;
|
|
146
|
+
|
|
147
|
+
${code}
|
|
148
|
+
|
|
149
|
+
if (typeof execute === 'function') {
|
|
150
|
+
return await execute(params);
|
|
151
|
+
}
|
|
152
|
+
throw new Error('Implementation must define an execute function');
|
|
153
|
+
});
|
|
154
|
+
`;
|
|
155
|
+
}
|
|
156
|
+
async executeWithTimeout(wrappedCode, context, params, timeout) {
|
|
157
|
+
return new Promise((resolve, reject) => {
|
|
158
|
+
const timer = setTimeout(() => {
|
|
159
|
+
reject(new Error(`Execution timeout: exceeded ${timeout}ms`));
|
|
160
|
+
}, timeout);
|
|
161
|
+
try {
|
|
162
|
+
const factory = new Function(wrappedCode);
|
|
163
|
+
const executor = factory();
|
|
164
|
+
Promise.resolve(executor(params, context))
|
|
165
|
+
.then((result) => {
|
|
166
|
+
clearTimeout(timer);
|
|
167
|
+
resolve(result);
|
|
168
|
+
})
|
|
169
|
+
.catch((error) => {
|
|
170
|
+
clearTimeout(timer);
|
|
171
|
+
reject(error);
|
|
172
|
+
});
|
|
173
|
+
}
|
|
174
|
+
catch (error) {
|
|
175
|
+
clearTimeout(timer);
|
|
176
|
+
reject(error);
|
|
177
|
+
}
|
|
178
|
+
});
|
|
179
|
+
}
|
|
180
|
+
async executeUnsandboxed(tool, params, startTime) {
|
|
181
|
+
try {
|
|
182
|
+
const factory = new Function(`
|
|
183
|
+
"use strict";
|
|
184
|
+
${tool.implementation}
|
|
185
|
+
return execute;
|
|
186
|
+
`);
|
|
187
|
+
const execute = factory();
|
|
188
|
+
const result = await execute(params);
|
|
189
|
+
return {
|
|
190
|
+
success: true,
|
|
191
|
+
result,
|
|
192
|
+
executionTime: Date.now() - startTime,
|
|
193
|
+
memoryUsed: this.estimateMemoryUsage(result),
|
|
194
|
+
logs: [],
|
|
195
|
+
};
|
|
196
|
+
}
|
|
197
|
+
catch (error) {
|
|
198
|
+
return {
|
|
199
|
+
success: false,
|
|
200
|
+
error: error instanceof Error ? error.message : String(error),
|
|
201
|
+
executionTime: Date.now() - startTime,
|
|
202
|
+
memoryUsed: 0,
|
|
203
|
+
logs: [],
|
|
204
|
+
};
|
|
205
|
+
}
|
|
206
|
+
}
|
|
207
|
+
estimateMemoryUsage(value) {
|
|
208
|
+
try {
|
|
209
|
+
const str = JSON.stringify(value);
|
|
210
|
+
return str.length * 2;
|
|
211
|
+
}
|
|
212
|
+
catch {
|
|
213
|
+
return 0;
|
|
214
|
+
}
|
|
215
|
+
}
|
|
216
|
+
deepEqual(a, b) {
|
|
217
|
+
if (a === b)
|
|
218
|
+
return true;
|
|
219
|
+
if (typeof a !== typeof b)
|
|
220
|
+
return false;
|
|
221
|
+
if (a === null || b === null)
|
|
222
|
+
return a === b;
|
|
223
|
+
if (Array.isArray(a) && Array.isArray(b)) {
|
|
224
|
+
if (a.length !== b.length)
|
|
225
|
+
return false;
|
|
226
|
+
return a.every((val, i) => this.deepEqual(val, b[i]));
|
|
227
|
+
}
|
|
228
|
+
if (typeof a === 'object' && typeof b === 'object') {
|
|
229
|
+
const aObj = a;
|
|
230
|
+
const bObj = b;
|
|
231
|
+
const keysA = Object.keys(aObj);
|
|
232
|
+
const keysB = Object.keys(bObj);
|
|
233
|
+
if (keysA.length !== keysB.length)
|
|
234
|
+
return false;
|
|
235
|
+
return keysA.every((key) => this.deepEqual(aObj[key], bObj[key]));
|
|
236
|
+
}
|
|
237
|
+
return false;
|
|
238
|
+
}
|
|
239
|
+
}
|
|
240
|
+
//# sourceMappingURL=tool-sandbox.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tool-sandbox.js","sourceRoot":"","sources":["../../src/tool-generation/tool-sandbox.ts"],"names":[],"mappings":"AAMA,MAAM,CAAC,MAAM,sBAAsB,GAAsB;IACvD,OAAO,EAAE,IAAI;IACb,gBAAgB,EAAE,IAAI;IACtB,SAAS,EAAE,EAAE,GAAG,IAAI,GAAG,IAAI;IAC3B,cAAc,EAAE,EAAE;IAClB,cAAc,EAAE,QAAQ;CACzB,CAAC;AAyBF,MAAM,OAAO,WAAW;IACL,MAAM,CAAoB;IAC1B,IAAI,GAAa,EAAE,CAAC;IAErC,YAAY,SAAqC,EAAE;QACjD,IAAI,CAAC,MAAM,GAAG,EAAE,GAAG,sBAAsB,EAAE,GAAG,MAAM,EAAE,CAAC;IACzD,CAAC;IAED,KAAK,CAAC,OAAO,CACX,IAAmB,EACnB,MAAe;QAEf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC;QAErB,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACzB,OAAO,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;QAC1D,CAAC;QAED,IAAI,CAAC;YACH,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YAEjD,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;YACrC,MAAM,WAAW,GAAG,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YAEjE,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,kBAAkB,CAC1C,WAAW,EACX,OAAO,EACP,MAAM,EACN,IAAI,CAAC,MAAM,CAAC,gBAAgB,CAC7B,CAAC;YAEF,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,MAAM;gBACN,aAAa,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACrC,UAAU,EAAE,IAAI,CAAC,mBAAmB,CAAC,MAAM,CAAC;gBAC5C,IAAI,EAAE,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;aACrB,CAAC;QACJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;gBAC7D,aAAa,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACrC,UAAU,EAAE,CAAC;gBACb,IAAI,EAAE,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;aACrB,CAAC;QACJ,CAAC;IACH,CAAC;IAED,KAAK,CAAC,aAAa,CACjB,IAAmB,EACnB,SAAqF;QAYrF,MAAM,OAAO,GAMR,EAAE,CAAC;QAER,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;YAE5D,IAAI,MAAM,GAAG,KAAK,CAAC;YACnB,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC;gBACzB,MAAM,GAAG,CAAC,UAAU,CAAC,OAAO,CAAC;YAC/B,CAAC;iBAAM,IAAI,QAAQ,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;gBACjD,MAAM;oBACJ,UAAU,CAAC,OAAO;wBAClB,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,MAAM,EAAE,QAAQ,CAAC,cAAc,CAAC,CAAC;YAC/D,CAAC;iBAAM,CAAC;gBACN,MAAM,GAAG,UAAU,CAAC,OAAO,CAAC;YAC9B,CAAC;YAED,OAAO,CAAC,IAAI,CAAC;gBACX,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,MAAM,EAAE,UAAU,CAAC,MAAM;gBACzB,KAAK,EAAE,UAAU,CAAC,KAAK;gBACvB,MAAM;gBACN,aAAa,EAAE,UAAU,CAAC,aAAa;aACxC,CAAC,CAAC;QACL,CAAC;QAED,OAAO;YACL,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM;YAC9C,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM;YAC/C,OAAO;SACR,CAAC;IACJ,CAAC;IAEO,sBAAsB,CAAC,IAAY;QACzC,MAAM,SAAS,GAAG;YAChB,aAAa;YACb,iBAAiB;YACjB,uBAAuB;YACvB,eAAe;YACf,gBAAgB;YAChB,aAAa;YACb,YAAY;YACZ,gBAAgB;YAChB,YAAY;YACZ,cAAc;YACd,iBAAiB;YACjB,QAAQ;YACR,UAAU;YACV,WAAW;YACX,SAAS;YACT,SAAS;YACT,QAAQ;YACR,aAAa;YACb,cAAc;YACd,WAAW;YACX,oBAAoB;SACrB,CAAC;QAEF,IAAI,IAAI,CAAC,MAAM,CAAC,cAAc,KAAK,QAAQ,EAAE,CAAC;YAC5C,SAAS,CAAC,IAAI,CACZ,cAAc,EACd,kBAAkB,EAClB,aAAa,EACb,mBAAmB,EACnB,oBAAoB,CACrB,CAAC;QACJ,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,SAAS,EAAE,CAAC;YAChC,IAAI,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBACvB,MAAM,IAAI,KAAK,CACb,oDAAoD,OAAO,CAAC,MAAM,EAAE,CACrE,CAAC;YACJ,CAAC;QACH,CAAC;QAED,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;QACtC,IAAI,KAAK,GAAG,GAAG,EAAE,CAAC;YAChB,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,kBAAkB,CAAC,CAAC;QACxE,CAAC;IACH,CAAC;IAEO,aAAa;QACnB,MAAM,OAAO,GAAG,IAAI,CAAC;QAErB,OAAO;YACL,OAAO,EAAE;gBACP,GAAG,EAAE,CAAC,GAAG,IAAe,EAAE,EAAE;oBAC1B,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBAC3D,CAAC;gBACD,IAAI,EAAE,CAAC,GAAG,IAAe,EAAE,EAAE;oBAC3B,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBAC5D,CAAC;gBACD,KAAK,EAAE,CAAC,GAAG,IAAe,EAAE,EAAE;oBAC5B,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBAC7D,CAAC;aACF;YACD,IAAI;YACJ,IAAI;YACJ,IAAI;YACJ,KAAK;YACL,MAAM;YACN,MAAM;YACN,MAAM;YACN,OAAO;YACP,MAAM;YACN,GAAG;YACH,GAAG;YACH,OAAO;YACP,KAAK;YACL,SAAS;YACT,UAAU;SACX,CAAC;IACJ,CAAC;IAEO,kBAAkB,CAAC,IAAY;QACrC,OAAO;;;;;UAKD,IAAI;;;;;;;KAOT,CAAC;IACJ,CAAC;IAEO,KAAK,CAAC,kBAAkB,CAC9B,WAAmB,EACnB,OAAuB,EACvB,MAAe,EACf,OAAe;QAEf,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YACrC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;gBAC5B,MAAM,CAAC,IAAI,KAAK,CAAC,+BAA+B,OAAO,IAAI,CAAC,CAAC,CAAC;YAChE,CAAC,EAAE,OAAO,CAAC,CAAC;YAEZ,IAAI,CAAC;gBACH,MAAM,OAAO,GAAG,IAAI,QAAQ,CAAC,WAAW,CAAC,CAAC;gBAC1C,MAAM,QAAQ,GAAG,OAAO,EAAE,CAAC;gBAE3B,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;qBACvC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE;oBACf,YAAY,CAAC,KAAK,CAAC,CAAC;oBACpB,OAAO,CAAC,MAAM,CAAC,CAAC;gBAClB,CAAC,CAAC;qBACD,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE;oBACf,YAAY,CAAC,KAAK,CAAC,CAAC;oBACpB,MAAM,CAAC,KAAK,CAAC,CAAC;gBAChB,CAAC,CAAC,CAAC;YACP,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,YAAY,CAAC,KAAK,CAAC,CAAC;gBACpB,MAAM,CAAC,KAAK,CAAC,CAAC;YAChB,CAAC;QACH,CAAC,CAAC,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,kBAAkB,CAC9B,IAAmB,EACnB,MAAe,EACf,SAAiB;QAEjB,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,IAAI,QAAQ,CAAC;;UAEzB,IAAI,CAAC,cAAc;;OAEtB,CAAC,CAAC;YACH,MAAM,OAAO,GAAG,OAAO,EAAE,CAAC;YAC1B,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC;YAErC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,MAAM;gBACN,aAAa,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACrC,UAAU,EAAE,IAAI,CAAC,mBAAmB,CAAC,MAAM,CAAC;gBAC5C,IAAI,EAAE,EAAE;aACT,CAAC;QACJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;gBAC7D,aAAa,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBACrC,UAAU,EAAE,CAAC;gBACb,IAAI,EAAE,EAAE;aACT,CAAC;QACJ,CAAC;IACH,CAAC;IAEO,mBAAmB,CAAC,KAAc;QACxC,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YAClC,OAAO,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC;QACxB,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,CAAC,CAAC;QACX,CAAC;IACH,CAAC;IAEO,SAAS,CAAC,CAAU,EAAE,CAAU;QACtC,IAAI,CAAC,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QACzB,IAAI,OAAO,CAAC,KAAK,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;QACxC,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,KAAK,IAAI;YAAE,OAAO,CAAC,KAAK,CAAC,CAAC;QAE7C,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;YACzC,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;gBAAE,OAAO,KAAK,CAAC;YACxC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACxD,CAAC;QAED,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,QAAQ,EAAE,CAAC;YACnD,MAAM,IAAI,GAAG,CAA4B,CAAC;YAC1C,MAAM,IAAI,GAAG,CAA4B,CAAC;YAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAChC,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAEhC,IAAI,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM;gBAAE,OAAO,KAAK,CAAC;YAChD,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACpE,CAAC;QAED,OAAO,KAAK,CAAC;IACf,CAAC;CACF"}
|
|
@@ -0,0 +1,32 @@
|
|
|
1
|
+
import type { GeneratedTool, ToolValidationResult, LLMBackend, ToolSelfGenerationConfig } from '@cogitator-ai/types';
|
|
2
|
+
export interface ToolValidatorOptions {
|
|
3
|
+
llm?: LLMBackend;
|
|
4
|
+
config: ToolSelfGenerationConfig;
|
|
5
|
+
}
|
|
6
|
+
interface ValidationRule {
|
|
7
|
+
id: string;
|
|
8
|
+
name: string;
|
|
9
|
+
severity: 'error' | 'warning' | 'info';
|
|
10
|
+
check: (code: string, tool: GeneratedTool) => string | null;
|
|
11
|
+
}
|
|
12
|
+
export declare class ToolValidator {
|
|
13
|
+
private readonly llm?;
|
|
14
|
+
private readonly config;
|
|
15
|
+
private readonly sandbox;
|
|
16
|
+
private readonly customRules;
|
|
17
|
+
constructor(options: ToolValidatorOptions);
|
|
18
|
+
validate(tool: GeneratedTool, testCases?: Array<{
|
|
19
|
+
input: unknown;
|
|
20
|
+
expectedOutput?: unknown;
|
|
21
|
+
shouldThrow?: boolean;
|
|
22
|
+
}>): Promise<ToolValidationResult>;
|
|
23
|
+
addRule(rule: ValidationRule): void;
|
|
24
|
+
private runStaticAnalysis;
|
|
25
|
+
private runLLMValidation;
|
|
26
|
+
private generateBasicTestCases;
|
|
27
|
+
private generateSampleValue;
|
|
28
|
+
private calculateScore;
|
|
29
|
+
private callLLM;
|
|
30
|
+
}
|
|
31
|
+
export {};
|
|
32
|
+
//# sourceMappingURL=tool-validator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tool-validator.d.ts","sourceRoot":"","sources":["../../src/tool-generation/tool-validator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,aAAa,EACb,oBAAoB,EACpB,UAAU,EACV,wBAAwB,EACzB,MAAM,qBAAqB,CAAC;AAI7B,MAAM,WAAW,oBAAoB;IACnC,GAAG,CAAC,EAAE,UAAU,CAAC;IACjB,MAAM,EAAE,wBAAwB,CAAC;CAClC;AAED,UAAU,cAAc;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,OAAO,GAAG,SAAS,GAAG,MAAM,CAAC;IACvC,KAAK,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,aAAa,KAAK,MAAM,GAAG,IAAI,CAAC;CAC7D;AAyHD,qBAAa,aAAa;IACxB,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAa;IAClC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAA2B;IAClD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAc;IACtC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;gBAExC,OAAO,EAAE,oBAAoB;IAMnC,QAAQ,CACZ,IAAI,EAAE,aAAa,EACnB,SAAS,CAAC,EAAE,KAAK,CAAC;QAAE,KAAK,EAAE,OAAO,CAAC;QAAC,cAAc,CAAC,EAAE,OAAO,CAAC;QAAC,WAAW,CAAC,EAAE,OAAO,CAAA;KAAE,CAAC,GACrF,OAAO,CAAC,oBAAoB,CAAC;IAoEhC,OAAO,CAAC,IAAI,EAAE,cAAc,GAAG,IAAI;IAInC,OAAO,CAAC,iBAAiB;YAsBX,gBAAgB;IAmC9B,OAAO,CAAC,sBAAsB;IA2B9B,OAAO,CAAC,mBAAmB;IAoB3B,OAAO,CAAC,cAAc;YAuBR,OAAO;CAUtB"}
|