@sammorrowdrums/mcpi-ext 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/.dockerignore +6 -0
- package/.github/dependabot.yml +25 -0
- package/.github/workflows/ci.yml +50 -0
- package/.hadolint.yaml +6 -0
- package/.prettierignore +4 -0
- package/.prettierrc +7 -0
- package/AGENTS.md +118 -0
- package/DECISIONS.md +99 -0
- package/Dockerfile +39 -0
- package/README.md +278 -0
- package/aube-lock.yaml +4146 -0
- package/dist/code-mode/code-mode-integration.test.d.ts +2 -0
- package/dist/code-mode/code-mode-integration.test.d.ts.map +1 -0
- package/dist/code-mode/code-mode-integration.test.js +116 -0
- package/dist/code-mode/code-mode-integration.test.js.map +1 -0
- package/dist/code-mode/eligibility.d.ts +10 -0
- package/dist/code-mode/eligibility.d.ts.map +1 -0
- package/dist/code-mode/eligibility.js +14 -0
- package/dist/code-mode/eligibility.js.map +1 -0
- package/dist/code-mode/eligibility.test.d.ts +2 -0
- package/dist/code-mode/eligibility.test.d.ts.map +1 -0
- package/dist/code-mode/eligibility.test.js +53 -0
- package/dist/code-mode/eligibility.test.js.map +1 -0
- package/dist/code-mode/executor.d.ts +35 -0
- package/dist/code-mode/executor.d.ts.map +1 -0
- package/dist/code-mode/executor.js +120 -0
- package/dist/code-mode/executor.js.map +1 -0
- package/dist/code-mode/executor.test.d.ts +2 -0
- package/dist/code-mode/executor.test.d.ts.map +1 -0
- package/dist/code-mode/executor.test.js +130 -0
- package/dist/code-mode/executor.test.js.map +1 -0
- package/dist/code-mode/index.d.ts +65 -0
- package/dist/code-mode/index.d.ts.map +1 -0
- package/dist/code-mode/index.js +175 -0
- package/dist/code-mode/index.js.map +1 -0
- package/dist/code-mode/tools.d.ts +45 -0
- package/dist/code-mode/tools.d.ts.map +1 -0
- package/dist/code-mode/tools.js +70 -0
- package/dist/code-mode/tools.js.map +1 -0
- package/dist/code-mode/type-hints.d.ts +20 -0
- package/dist/code-mode/type-hints.d.ts.map +1 -0
- package/dist/code-mode/type-hints.js +189 -0
- package/dist/code-mode/type-hints.js.map +1 -0
- package/dist/code-mode/type-hints.test.d.ts +2 -0
- package/dist/code-mode/type-hints.test.d.ts.map +1 -0
- package/dist/code-mode/type-hints.test.js +157 -0
- package/dist/code-mode/type-hints.test.js.map +1 -0
- package/dist/docker-e2e.d.ts +55 -0
- package/dist/docker-e2e.d.ts.map +1 -0
- package/dist/docker-e2e.js +235 -0
- package/dist/docker-e2e.js.map +1 -0
- package/dist/docker-e2e.test.d.ts +2 -0
- package/dist/docker-e2e.test.d.ts.map +1 -0
- package/dist/docker-e2e.test.js +176 -0
- package/dist/docker-e2e.test.js.map +1 -0
- package/dist/index.d.ts +3 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +128 -0
- package/dist/index.js.map +1 -0
- package/dist/index.test.d.ts +2 -0
- package/dist/index.test.d.ts.map +1 -0
- package/dist/index.test.js +8 -0
- package/dist/index.test.js.map +1 -0
- package/dist/mcp/client-manager.d.ts +46 -0
- package/dist/mcp/client-manager.d.ts.map +1 -0
- package/dist/mcp/client-manager.js +148 -0
- package/dist/mcp/client-manager.js.map +1 -0
- package/dist/mcp/client-manager.test.d.ts +2 -0
- package/dist/mcp/client-manager.test.d.ts.map +1 -0
- package/dist/mcp/client-manager.test.js +204 -0
- package/dist/mcp/client-manager.test.js.map +1 -0
- package/dist/mcp/config-loader.d.ts +8 -0
- package/dist/mcp/config-loader.d.ts.map +1 -0
- package/dist/mcp/config-loader.js +43 -0
- package/dist/mcp/config-loader.js.map +1 -0
- package/dist/mcp/config-loader.test.d.ts +2 -0
- package/dist/mcp/config-loader.test.d.ts.map +1 -0
- package/dist/mcp/config-loader.test.js +93 -0
- package/dist/mcp/config-loader.test.js.map +1 -0
- package/dist/mcp/config.d.ts +46 -0
- package/dist/mcp/config.d.ts.map +1 -0
- package/dist/mcp/config.js +28 -0
- package/dist/mcp/config.js.map +1 -0
- package/dist/mcp/index.d.ts +5 -0
- package/dist/mcp/index.d.ts.map +1 -0
- package/dist/mcp/index.js +4 -0
- package/dist/mcp/index.js.map +1 -0
- package/dist/skills/discover.d.ts +10 -0
- package/dist/skills/discover.d.ts.map +1 -0
- package/dist/skills/discover.js +60 -0
- package/dist/skills/discover.js.map +1 -0
- package/dist/skills/format.d.ts +10 -0
- package/dist/skills/format.d.ts.map +1 -0
- package/dist/skills/format.js +38 -0
- package/dist/skills/format.js.map +1 -0
- package/dist/skills/format.test.d.ts +2 -0
- package/dist/skills/format.test.d.ts.map +1 -0
- package/dist/skills/format.test.js +63 -0
- package/dist/skills/format.test.js.map +1 -0
- package/dist/skills/index.d.ts +6 -0
- package/dist/skills/index.d.ts.map +1 -0
- package/dist/skills/index.js +6 -0
- package/dist/skills/index.js.map +1 -0
- package/dist/skills/load-skill-tool.d.ts +40 -0
- package/dist/skills/load-skill-tool.d.ts.map +1 -0
- package/dist/skills/load-skill-tool.js +104 -0
- package/dist/skills/load-skill-tool.js.map +1 -0
- package/dist/skills/mcp-tool-proxy.d.ts +11 -0
- package/dist/skills/mcp-tool-proxy.d.ts.map +1 -0
- package/dist/skills/mcp-tool-proxy.js +116 -0
- package/dist/skills/mcp-tool-proxy.js.map +1 -0
- package/dist/skills/skill-integration.test.d.ts +2 -0
- package/dist/skills/skill-integration.test.d.ts.map +1 -0
- package/dist/skills/skill-integration.test.js +119 -0
- package/dist/skills/skill-integration.test.js.map +1 -0
- package/dist/skills/skill-registry.d.ts +39 -0
- package/dist/skills/skill-registry.d.ts.map +1 -0
- package/dist/skills/skill-registry.js +48 -0
- package/dist/skills/skill-registry.js.map +1 -0
- package/dist/skills/skill-registry.test.d.ts +2 -0
- package/dist/skills/skill-registry.test.d.ts.map +1 -0
- package/dist/skills/skill-registry.test.js +67 -0
- package/dist/skills/skill-registry.test.js.map +1 -0
- package/dist/test-servers/weather-server.d.ts +12 -0
- package/dist/test-servers/weather-server.d.ts.map +1 -0
- package/dist/test-servers/weather-server.js +139 -0
- package/dist/test-servers/weather-server.js.map +1 -0
- package/dist/test-servers/weather-stdio.d.ts +2 -0
- package/dist/test-servers/weather-stdio.d.ts.map +1 -0
- package/dist/test-servers/weather-stdio.js +11 -0
- package/dist/test-servers/weather-stdio.js.map +1 -0
- package/dist/tool-cli/cli.d.ts +3 -0
- package/dist/tool-cli/cli.d.ts.map +1 -0
- package/dist/tool-cli/cli.js +179 -0
- package/dist/tool-cli/cli.js.map +1 -0
- package/dist/tool-cli/constants.d.ts +7 -0
- package/dist/tool-cli/constants.d.ts.map +1 -0
- package/dist/tool-cli/constants.js +15 -0
- package/dist/tool-cli/constants.js.map +1 -0
- package/dist/tool-cli/format.d.ts +8 -0
- package/dist/tool-cli/format.d.ts.map +1 -0
- package/dist/tool-cli/format.js +57 -0
- package/dist/tool-cli/format.js.map +1 -0
- package/dist/tool-cli/format.test.d.ts +2 -0
- package/dist/tool-cli/format.test.d.ts.map +1 -0
- package/dist/tool-cli/format.test.js +30 -0
- package/dist/tool-cli/format.test.js.map +1 -0
- package/dist/tool-cli/index.d.ts +4 -0
- package/dist/tool-cli/index.d.ts.map +1 -0
- package/dist/tool-cli/index.js +4 -0
- package/dist/tool-cli/index.js.map +1 -0
- package/dist/tool-cli/rpc-client.d.ts +6 -0
- package/dist/tool-cli/rpc-client.d.ts.map +1 -0
- package/dist/tool-cli/rpc-client.js +25 -0
- package/dist/tool-cli/rpc-client.js.map +1 -0
- package/dist/tool-cli/rpc-server.d.ts +48 -0
- package/dist/tool-cli/rpc-server.d.ts.map +1 -0
- package/dist/tool-cli/rpc-server.js +220 -0
- package/dist/tool-cli/rpc-server.js.map +1 -0
- package/dist/tool-cli/rpc-server.test.d.ts +2 -0
- package/dist/tool-cli/rpc-server.test.d.ts.map +1 -0
- package/dist/tool-cli/rpc-server.test.js +154 -0
- package/dist/tool-cli/rpc-server.test.js.map +1 -0
- package/eslint.config.js +27 -0
- package/images/banner.webp +0 -0
- package/images/code-c-maude.webp +0 -0
- package/images/nuclear-mcp-football.webp +0 -0
- package/images/the-skill-dealer.webp +0 -0
- package/mise.toml +65 -0
- package/package.json +46 -0
- package/scripts/docker-entrypoint.sh +29 -0
- package/scripts/test-echo-server.mjs +28 -0
- package/scripts/test-mcp-integration.mjs +59 -0
- package/scripts/test-tool-cli-smoke.mjs +62 -0
- package/src/code-mode/code-mode-integration.test.ts +135 -0
- package/src/code-mode/eligibility.test.ts +60 -0
- package/src/code-mode/eligibility.ts +16 -0
- package/src/code-mode/executor.test.ts +168 -0
- package/src/code-mode/executor.ts +163 -0
- package/src/code-mode/index.ts +208 -0
- package/src/code-mode/tools.ts +110 -0
- package/src/code-mode/type-hints.test.ts +187 -0
- package/src/code-mode/type-hints.ts +249 -0
- package/src/docker-e2e.test.ts +195 -0
- package/src/docker-e2e.ts +277 -0
- package/src/index.test.ts +8 -0
- package/src/index.ts +158 -0
- package/src/mcp/client-manager.test.ts +250 -0
- package/src/mcp/client-manager.ts +199 -0
- package/src/mcp/config-loader.test.ts +115 -0
- package/src/mcp/config-loader.ts +46 -0
- package/src/mcp/config.ts +40 -0
- package/src/mcp/index.ts +4 -0
- package/src/skills/discover.ts +81 -0
- package/src/skills/format.test.ts +70 -0
- package/src/skills/format.ts +42 -0
- package/src/skills/index.ts +9 -0
- package/src/skills/load-skill-tool.ts +136 -0
- package/src/skills/mcp-tool-proxy.ts +143 -0
- package/src/skills/skill-integration.test.ts +139 -0
- package/src/skills/skill-registry.test.ts +83 -0
- package/src/skills/skill-registry.ts +69 -0
- package/src/test-servers/weather-server.ts +171 -0
- package/src/test-servers/weather-stdio.ts +11 -0
- package/src/tool-cli/cli.ts +232 -0
- package/src/tool-cli/constants.ts +15 -0
- package/src/tool-cli/format.test.ts +33 -0
- package/src/tool-cli/format.ts +56 -0
- package/src/tool-cli/index.ts +3 -0
- package/src/tool-cli/rpc-client.ts +39 -0
- package/src/tool-cli/rpc-server.test.ts +193 -0
- package/src/tool-cli/rpc-server.ts +287 -0
- package/tsconfig.json +21 -0
|
@@ -0,0 +1,157 @@
|
|
|
1
|
+
import { describe, expect, it } from "vitest";
|
|
2
|
+
import { generateTypeHints, jsonSchemaToTypeString, sanitizeToolName } from "./type-hints.js";
|
|
3
|
+
describe("jsonSchemaToTypeString", () => {
|
|
4
|
+
it("converts string type", () => {
|
|
5
|
+
expect(jsonSchemaToTypeString({ type: "string" })).toBe("string");
|
|
6
|
+
});
|
|
7
|
+
it("converts number type", () => {
|
|
8
|
+
expect(jsonSchemaToTypeString({ type: "number" })).toBe("number");
|
|
9
|
+
});
|
|
10
|
+
it("converts integer type", () => {
|
|
11
|
+
expect(jsonSchemaToTypeString({ type: "integer" })).toBe("number");
|
|
12
|
+
});
|
|
13
|
+
it("converts boolean type", () => {
|
|
14
|
+
expect(jsonSchemaToTypeString({ type: "boolean" })).toBe("boolean");
|
|
15
|
+
});
|
|
16
|
+
it("converts null type", () => {
|
|
17
|
+
expect(jsonSchemaToTypeString({ type: "null" })).toBe("null");
|
|
18
|
+
});
|
|
19
|
+
it("converts type array", () => {
|
|
20
|
+
expect(jsonSchemaToTypeString({ type: ["string", "null"] })).toBe("string | null");
|
|
21
|
+
});
|
|
22
|
+
it("converts enum to union", () => {
|
|
23
|
+
expect(jsonSchemaToTypeString({ enum: ["a", "b", "c"] })).toBe('"a" | "b" | "c"');
|
|
24
|
+
});
|
|
25
|
+
it("converts const", () => {
|
|
26
|
+
expect(jsonSchemaToTypeString({ const: "fixed" })).toBe('"fixed"');
|
|
27
|
+
});
|
|
28
|
+
it("converts simple object", () => {
|
|
29
|
+
const result = jsonSchemaToTypeString({
|
|
30
|
+
type: "object",
|
|
31
|
+
properties: {
|
|
32
|
+
name: { type: "string" },
|
|
33
|
+
age: { type: "number" },
|
|
34
|
+
},
|
|
35
|
+
required: ["name"],
|
|
36
|
+
});
|
|
37
|
+
expect(result).toContain("name: string;");
|
|
38
|
+
expect(result).toContain("age?: number;");
|
|
39
|
+
});
|
|
40
|
+
it("converts empty object to Record", () => {
|
|
41
|
+
expect(jsonSchemaToTypeString({ type: "object" })).toBe("Record<string, unknown>");
|
|
42
|
+
});
|
|
43
|
+
it("converts array with items", () => {
|
|
44
|
+
expect(jsonSchemaToTypeString({ type: "array", items: { type: "string" } })).toBe("string[]");
|
|
45
|
+
});
|
|
46
|
+
it("converts array without items", () => {
|
|
47
|
+
expect(jsonSchemaToTypeString({ type: "array" })).toBe("unknown[]");
|
|
48
|
+
});
|
|
49
|
+
it("handles anyOf", () => {
|
|
50
|
+
const result = jsonSchemaToTypeString({
|
|
51
|
+
anyOf: [{ type: "string" }, { type: "number" }],
|
|
52
|
+
});
|
|
53
|
+
expect(result).toBe("string | number");
|
|
54
|
+
});
|
|
55
|
+
it("handles oneOf", () => {
|
|
56
|
+
const result = jsonSchemaToTypeString({
|
|
57
|
+
oneOf: [{ type: "boolean" }, { type: "null" }],
|
|
58
|
+
});
|
|
59
|
+
expect(result).toBe("boolean | null");
|
|
60
|
+
});
|
|
61
|
+
it("handles allOf", () => {
|
|
62
|
+
const result = jsonSchemaToTypeString({
|
|
63
|
+
allOf: [
|
|
64
|
+
{ type: "object", properties: { a: { type: "string" } }, required: ["a"] },
|
|
65
|
+
{ type: "object", properties: { b: { type: "number" } }, required: ["b"] },
|
|
66
|
+
],
|
|
67
|
+
});
|
|
68
|
+
expect(result).toContain("a: string;");
|
|
69
|
+
expect(result).toContain("b: number;");
|
|
70
|
+
expect(result).toContain("&");
|
|
71
|
+
});
|
|
72
|
+
it("resolves $ref", () => {
|
|
73
|
+
const result = jsonSchemaToTypeString({ $ref: "#/definitions/Foo" }, { Foo: { type: "string" } });
|
|
74
|
+
expect(result).toBe("string");
|
|
75
|
+
});
|
|
76
|
+
it("handles unknown $ref gracefully", () => {
|
|
77
|
+
expect(jsonSchemaToTypeString({ $ref: "#/definitions/Missing" }, {})).toBe("unknown");
|
|
78
|
+
});
|
|
79
|
+
it("guards against deep recursion", () => {
|
|
80
|
+
// Create a deeply nested schema
|
|
81
|
+
let schema = { type: "string" };
|
|
82
|
+
for (let i = 0; i < 25; i++) {
|
|
83
|
+
schema = { type: "object", properties: { nested: schema }, required: ["nested"] };
|
|
84
|
+
}
|
|
85
|
+
const result = jsonSchemaToTypeString(schema);
|
|
86
|
+
expect(result).toContain("unknown"); // Should hit depth guard
|
|
87
|
+
});
|
|
88
|
+
it("includes description as JSDoc in object properties", () => {
|
|
89
|
+
const result = jsonSchemaToTypeString({
|
|
90
|
+
type: "object",
|
|
91
|
+
properties: {
|
|
92
|
+
query: { type: "string", description: "Search term" },
|
|
93
|
+
},
|
|
94
|
+
required: ["query"],
|
|
95
|
+
});
|
|
96
|
+
expect(result).toContain("/** Search term */");
|
|
97
|
+
});
|
|
98
|
+
});
|
|
99
|
+
describe("sanitizeToolName", () => {
|
|
100
|
+
it("replaces hyphens with underscores", () => {
|
|
101
|
+
expect(sanitizeToolName("list-issues")).toBe("list_issues");
|
|
102
|
+
});
|
|
103
|
+
it("replaces dots with underscores", () => {
|
|
104
|
+
expect(sanitizeToolName("github.search")).toBe("github_search");
|
|
105
|
+
});
|
|
106
|
+
it("keeps valid identifiers unchanged", () => {
|
|
107
|
+
expect(sanitizeToolName("searchDocs")).toBe("searchDocs");
|
|
108
|
+
});
|
|
109
|
+
});
|
|
110
|
+
describe("generateTypeHints", () => {
|
|
111
|
+
it("returns empty string for no tools", () => {
|
|
112
|
+
expect(generateTypeHints([])).toBe("");
|
|
113
|
+
});
|
|
114
|
+
it("generates declaration for a simple tool", () => {
|
|
115
|
+
const tools = [
|
|
116
|
+
{
|
|
117
|
+
name: "search_docs",
|
|
118
|
+
description: "Search documentation",
|
|
119
|
+
inputSchema: {
|
|
120
|
+
type: "object",
|
|
121
|
+
properties: {
|
|
122
|
+
query: { type: "string", description: "Search term" },
|
|
123
|
+
},
|
|
124
|
+
required: ["query"],
|
|
125
|
+
},
|
|
126
|
+
outputSchema: {
|
|
127
|
+
type: "object",
|
|
128
|
+
properties: {
|
|
129
|
+
results: { type: "array", items: { type: "string" } },
|
|
130
|
+
},
|
|
131
|
+
required: ["results"],
|
|
132
|
+
},
|
|
133
|
+
serverName: "docs",
|
|
134
|
+
},
|
|
135
|
+
];
|
|
136
|
+
const result = generateTypeHints(tools);
|
|
137
|
+
expect(result).toContain("declare const codemode");
|
|
138
|
+
expect(result).toContain("search_docs:");
|
|
139
|
+
expect(result).toContain("Promise<");
|
|
140
|
+
expect(result).toContain("listTools");
|
|
141
|
+
expect(result).toContain("describeTools");
|
|
142
|
+
});
|
|
143
|
+
it("sanitizes tool names with special characters", () => {
|
|
144
|
+
const tools = [
|
|
145
|
+
{
|
|
146
|
+
name: "github.list-repos",
|
|
147
|
+
description: "List repos",
|
|
148
|
+
inputSchema: { type: "object", properties: {} },
|
|
149
|
+
outputSchema: { type: "object", properties: {} },
|
|
150
|
+
serverName: "github",
|
|
151
|
+
},
|
|
152
|
+
];
|
|
153
|
+
const result = generateTypeHints(tools);
|
|
154
|
+
expect(result).toContain("github_list_repos:");
|
|
155
|
+
});
|
|
156
|
+
});
|
|
157
|
+
//# sourceMappingURL=type-hints.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"type-hints.test.js","sourceRoot":"","sources":["../../src/code-mode/type-hints.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAE9C,OAAO,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAE9F,QAAQ,CAAC,wBAAwB,EAAE,GAAG,EAAE;IACtC,EAAE,CAAC,sBAAsB,EAAE,GAAG,EAAE;QAC9B,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACpE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,sBAAsB,EAAE,GAAG,EAAE;QAC9B,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACpE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uBAAuB,EAAE,GAAG,EAAE;QAC/B,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACrE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uBAAuB,EAAE,GAAG,EAAE;QAC/B,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACtE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAC5B,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAChE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qBAAqB,EAAE,GAAG,EAAE;QAC7B,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,CAAC,QAAQ,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IACrF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wBAAwB,EAAE,GAAG,EAAE;QAChC,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC;IACpF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gBAAgB,EAAE,GAAG,EAAE;QACxB,MAAM,CAAC,sBAAsB,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACrE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wBAAwB,EAAE,GAAG,EAAE;QAChC,MAAM,MAAM,GAAG,sBAAsB,CAAC;YACpC,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;gBACxB,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;aACxB;YACD,QAAQ,EAAE,CAAC,MAAM,CAAC;SACnB,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;QAC1C,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iCAAiC,EAAE,GAAG,EAAE;QACzC,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IACrF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2BAA2B,EAAE,GAAG,EAAE;QACnC,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IAChG,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8BAA8B,EAAE,GAAG,EAAE;QACtC,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IACtE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,EAAE,GAAG,EAAE;QACvB,MAAM,MAAM,GAAG,sBAAsB,CAAC;YACpC,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;SAChD,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,EAAE,GAAG,EAAE;QACvB,MAAM,MAAM,GAAG,sBAAsB,CAAC;YACpC,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;SAC/C,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACxC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,EAAE,GAAG,EAAE;QACvB,MAAM,MAAM,GAAG,sBAAsB,CAAC;YACpC,KAAK,EAAE;gBACL,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,CAAC,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,QAAQ,EAAE,CAAC,GAAG,CAAC,EAAE;gBAC1E,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,CAAC,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,QAAQ,EAAE,CAAC,GAAG,CAAC,EAAE;aAC3E;SACF,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,eAAe,EAAE,GAAG,EAAE;QACvB,MAAM,MAAM,GAAG,sBAAsB,CACnC,EAAE,IAAI,EAAE,mBAAmB,EAAE,EAC7B,EAAE,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,CAC5B,CAAC;QACF,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iCAAiC,EAAE,GAAG,EAAE;QACzC,MAAM,CAAC,sBAAsB,CAAC,EAAE,IAAI,EAAE,uBAAuB,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACxF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+BAA+B,EAAE,GAAG,EAAE;QACvC,gCAAgC;QAChC,IAAI,MAAM,GAA4B,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;QACzD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,MAAM,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpF,CAAC;QACD,MAAM,MAAM,GAAG,sBAAsB,CAAC,MAAM,CAAC,CAAC;QAC9C,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC,CAAC,yBAAyB;IAChE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oDAAoD,EAAE,GAAG,EAAE;QAC5D,MAAM,MAAM,GAAG,sBAAsB,CAAC;YACpC,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,aAAa,EAAE;aACtD;YACD,QAAQ,EAAE,CAAC,OAAO,CAAC;SACpB,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,oBAAoB,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,kBAAkB,EAAE,GAAG,EAAE;IAChC,EAAE,CAAC,mCAAmC,EAAE,GAAG,EAAE;QAC3C,MAAM,CAAC,gBAAgB,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC9D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gCAAgC,EAAE,GAAG,EAAE;QACxC,MAAM,CAAC,gBAAgB,CAAC,eAAe,CAAC,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAClE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mCAAmC,EAAE,GAAG,EAAE;QAC3C,MAAM,CAAC,gBAAgB,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IAC5D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;IACjC,EAAE,CAAC,mCAAmC,EAAE,GAAG,EAAE;QAC3C,MAAM,CAAC,iBAAiB,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,GAAG,EAAE;QACjD,MAAM,KAAK,GAAc;YACvB;gBACE,IAAI,EAAE,aAAa;gBACnB,WAAW,EAAE,sBAAsB;gBACnC,WAAW,EAAE;oBACX,IAAI,EAAE,QAAQ;oBACd,UAAU,EAAE;wBACV,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,aAAa,EAAE;qBACtD;oBACD,QAAQ,EAAE,CAAC,OAAO,CAAC;iBACpB;gBACD,YAAY,EAAE;oBACZ,IAAI,EAAE,QAAQ;oBACd,UAAU,EAAE;wBACV,OAAO,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;qBACtD;oBACD,QAAQ,EAAE,CAAC,SAAS,CAAC;iBACtB;gBACD,UAAU,EAAE,MAAM;aACnB;SACF,CAAC;QAEF,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;QACxC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,wBAAwB,CAAC,CAAC;QACnD,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,cAAc,CAAC,CAAC;QACzC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC;QACtC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,MAAM,KAAK,GAAc;YACvB;gBACE,IAAI,EAAE,mBAAmB;gBACzB,WAAW,EAAE,YAAY;gBACzB,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE;gBAC/C,YAAY,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE;gBAChD,UAAU,EAAE,QAAQ;aACrB;SACF,CAAC;QAEF,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;QACxC,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,oBAAoB,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,55 @@
|
|
|
1
|
+
import type { AgentToolResult, AgentToolUpdateCallback, ExtensionContext } from "@sammorrowdrums/mcpi";
|
|
2
|
+
import { Type, type Static } from "typebox";
|
|
3
|
+
declare const DockerE2EParams: Type.TObject<{
|
|
4
|
+
task: Type.TString;
|
|
5
|
+
token: Type.TString;
|
|
6
|
+
model: Type.TOptional<Type.TString>;
|
|
7
|
+
timeout: Type.TOptional<Type.TNumber>;
|
|
8
|
+
}>;
|
|
9
|
+
type DockerE2EInput = Static<typeof DockerE2EParams>;
|
|
10
|
+
interface RunLogEntry {
|
|
11
|
+
type: string;
|
|
12
|
+
timestamp: number;
|
|
13
|
+
raw: string;
|
|
14
|
+
}
|
|
15
|
+
export interface DockerE2EDetails {
|
|
16
|
+
exitCode: number | null;
|
|
17
|
+
log: RunLogEntry[];
|
|
18
|
+
duration: number;
|
|
19
|
+
model?: string;
|
|
20
|
+
usage?: {
|
|
21
|
+
input: number;
|
|
22
|
+
output: number;
|
|
23
|
+
cacheRead: number;
|
|
24
|
+
cacheWrite: number;
|
|
25
|
+
cost: number;
|
|
26
|
+
turns: number;
|
|
27
|
+
};
|
|
28
|
+
}
|
|
29
|
+
export declare function parseJsonlLine(line: string): RunLogEntry | null;
|
|
30
|
+
export declare function extractAssistantText(log: RunLogEntry[]): string;
|
|
31
|
+
export declare function extractUsage(log: RunLogEntry[]): {
|
|
32
|
+
input: number;
|
|
33
|
+
output: number;
|
|
34
|
+
cacheRead: number;
|
|
35
|
+
cacheWrite: number;
|
|
36
|
+
cost: number;
|
|
37
|
+
turns: number;
|
|
38
|
+
};
|
|
39
|
+
export declare function extractModel(log: RunLogEntry[]): string | undefined;
|
|
40
|
+
export declare function buildDockerArgs(image: string, params: DockerE2EInput): string[];
|
|
41
|
+
export declare const dockerE2ETool: {
|
|
42
|
+
name: string;
|
|
43
|
+
label: string;
|
|
44
|
+
description: string;
|
|
45
|
+
promptSnippet: string;
|
|
46
|
+
parameters: Type.TObject<{
|
|
47
|
+
task: Type.TString;
|
|
48
|
+
token: Type.TString;
|
|
49
|
+
model: Type.TOptional<Type.TString>;
|
|
50
|
+
timeout: Type.TOptional<Type.TNumber>;
|
|
51
|
+
}>;
|
|
52
|
+
execute(_toolCallId: string, params: DockerE2EInput, signal: AbortSignal | undefined, onUpdate: AgentToolUpdateCallback<DockerE2EDetails> | undefined, _ctx: ExtensionContext): Promise<AgentToolResult<DockerE2EDetails>>;
|
|
53
|
+
};
|
|
54
|
+
export {};
|
|
55
|
+
//# sourceMappingURL=docker-e2e.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"docker-e2e.d.ts","sourceRoot":"","sources":["../src/docker-e2e.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,eAAe,EACf,uBAAuB,EACvB,gBAAgB,EACjB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,IAAI,EAAE,KAAK,MAAM,EAAE,MAAM,SAAS,CAAC;AAI5C,QAAA,MAAM,eAAe;;;;;EAenB,CAAC;AAEH,KAAK,cAAc,GAAG,MAAM,CAAC,OAAO,eAAe,CAAC,CAAC;AAErD,UAAU,WAAW;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,GAAG,EAAE,MAAM,CAAC;CACb;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,GAAG,EAAE,WAAW,EAAE,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE;QACN,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,IAAI,EAAE,MAAM,CAAC;QACb,KAAK,EAAE,MAAM,CAAC;KACf,CAAC;CACH;AAED,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,IAAI,CAS/D;AAED,wBAAgB,oBAAoB,CAAC,GAAG,EAAE,WAAW,EAAE,GAAG,MAAM,CAgB/D;AAED,wBAAgB,YAAY,CAAC,GAAG,EAAE,WAAW,EAAE;;;;;;;EAsB9C;AAED,wBAAgB,YAAY,CAAC,GAAG,EAAE,WAAW,EAAE,GAAG,MAAM,GAAG,SAAS,CAWnE;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,cAAc,GAAG,MAAM,EAAE,CAS/E;AAED,eAAO,MAAM,aAAa;;;;;;;;;;;yBAST,MAAM,UACX,cAAc,UACd,WAAW,GAAG,SAAS,YACrB,uBAAuB,CAAC,gBAAgB,CAAC,GAAG,SAAS,QACzD,gBAAgB,GACrB,OAAO,CAAC,eAAe,CAAC,gBAAgB,CAAC,CAAC;CAuI9C,CAAC"}
|
|
@@ -0,0 +1,235 @@
|
|
|
1
|
+
import { spawn } from "node:child_process";
|
|
2
|
+
import { Type } from "typebox";
|
|
3
|
+
const DOCKER_IMAGE = "mcpi-ext-e2e";
|
|
4
|
+
const DockerE2EParams = Type.Object({
|
|
5
|
+
task: Type.String({ description: "The task/prompt to run inside the sandbox" }),
|
|
6
|
+
token: Type.String({
|
|
7
|
+
description: "GitHub token for Copilot model access",
|
|
8
|
+
}),
|
|
9
|
+
model: Type.Optional(Type.String({
|
|
10
|
+
description: "Model to use (e.g. claude-sonnet-4-5). Defaults to the provider default.",
|
|
11
|
+
})),
|
|
12
|
+
timeout: Type.Optional(Type.Number({
|
|
13
|
+
description: "Timeout in seconds. Defaults to 300 (5 minutes).",
|
|
14
|
+
})),
|
|
15
|
+
});
|
|
16
|
+
export function parseJsonlLine(line) {
|
|
17
|
+
const trimmed = line.trim();
|
|
18
|
+
if (!trimmed)
|
|
19
|
+
return null;
|
|
20
|
+
try {
|
|
21
|
+
const parsed = JSON.parse(trimmed);
|
|
22
|
+
return { type: parsed.type ?? "unknown", timestamp: Date.now(), raw: trimmed };
|
|
23
|
+
}
|
|
24
|
+
catch {
|
|
25
|
+
return { type: "raw", timestamp: Date.now(), raw: trimmed };
|
|
26
|
+
}
|
|
27
|
+
}
|
|
28
|
+
export function extractAssistantText(log) {
|
|
29
|
+
const parts = [];
|
|
30
|
+
for (const entry of log) {
|
|
31
|
+
if (entry.type !== "message_end")
|
|
32
|
+
continue;
|
|
33
|
+
try {
|
|
34
|
+
const event = JSON.parse(entry.raw);
|
|
35
|
+
const msg = event.message;
|
|
36
|
+
if (msg?.role !== "assistant")
|
|
37
|
+
continue;
|
|
38
|
+
for (const part of msg.content ?? []) {
|
|
39
|
+
if (part.type === "text")
|
|
40
|
+
parts.push(part.text);
|
|
41
|
+
}
|
|
42
|
+
}
|
|
43
|
+
catch {
|
|
44
|
+
// skip malformed
|
|
45
|
+
}
|
|
46
|
+
}
|
|
47
|
+
return parts.join("\n");
|
|
48
|
+
}
|
|
49
|
+
export function extractUsage(log) {
|
|
50
|
+
const usage = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, cost: 0, turns: 0 };
|
|
51
|
+
for (const entry of log) {
|
|
52
|
+
if (entry.type !== "message_end")
|
|
53
|
+
continue;
|
|
54
|
+
try {
|
|
55
|
+
const event = JSON.parse(entry.raw);
|
|
56
|
+
const msg = event.message;
|
|
57
|
+
if (msg?.role !== "assistant")
|
|
58
|
+
continue;
|
|
59
|
+
usage.turns++;
|
|
60
|
+
const u = msg.usage;
|
|
61
|
+
if (u) {
|
|
62
|
+
usage.input += u.input ?? 0;
|
|
63
|
+
usage.output += u.output ?? 0;
|
|
64
|
+
usage.cacheRead += u.cacheRead ?? 0;
|
|
65
|
+
usage.cacheWrite += u.cacheWrite ?? 0;
|
|
66
|
+
usage.cost += u.cost?.total ?? 0;
|
|
67
|
+
}
|
|
68
|
+
}
|
|
69
|
+
catch {
|
|
70
|
+
// skip
|
|
71
|
+
}
|
|
72
|
+
}
|
|
73
|
+
return usage;
|
|
74
|
+
}
|
|
75
|
+
export function extractModel(log) {
|
|
76
|
+
for (const entry of log) {
|
|
77
|
+
if (entry.type !== "message_end")
|
|
78
|
+
continue;
|
|
79
|
+
try {
|
|
80
|
+
const event = JSON.parse(entry.raw);
|
|
81
|
+
if (event.message?.model)
|
|
82
|
+
return event.message.model;
|
|
83
|
+
}
|
|
84
|
+
catch {
|
|
85
|
+
// skip
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
return undefined;
|
|
89
|
+
}
|
|
90
|
+
export function buildDockerArgs(image, params) {
|
|
91
|
+
const args = ["run", "--rm", "-e", `GITHUB_TOKEN=${params.token}`];
|
|
92
|
+
if (params.model) {
|
|
93
|
+
args.push("-e", `PI_MODEL=${params.model}`);
|
|
94
|
+
}
|
|
95
|
+
args.push(image, params.task);
|
|
96
|
+
return args;
|
|
97
|
+
}
|
|
98
|
+
export const dockerE2ETool = {
|
|
99
|
+
name: "docker_e2e",
|
|
100
|
+
label: "Docker E2E",
|
|
101
|
+
description: "Run a task end-to-end inside an isolated Docker container with pi and this project's extensions. Returns the full run log for evaluation.",
|
|
102
|
+
promptSnippet: "Run an isolated e2e task in Docker. Requires a GitHub token and task prompt.",
|
|
103
|
+
parameters: DockerE2EParams,
|
|
104
|
+
async execute(_toolCallId, params, signal, onUpdate, _ctx) {
|
|
105
|
+
const startTime = Date.now();
|
|
106
|
+
const timeoutMs = (params.timeout ?? 300) * 1000;
|
|
107
|
+
const log = [];
|
|
108
|
+
let lastAssistantText = "";
|
|
109
|
+
const dockerArgs = buildDockerArgs(DOCKER_IMAGE, params);
|
|
110
|
+
const exitCode = await new Promise((resolve) => {
|
|
111
|
+
const proc = spawn("docker", dockerArgs, {
|
|
112
|
+
stdio: ["ignore", "pipe", "pipe"],
|
|
113
|
+
});
|
|
114
|
+
let buffer = "";
|
|
115
|
+
let timedOut = false;
|
|
116
|
+
let timeoutHandle;
|
|
117
|
+
if (timeoutMs > 0) {
|
|
118
|
+
timeoutHandle = setTimeout(() => {
|
|
119
|
+
timedOut = true;
|
|
120
|
+
proc.kill("SIGTERM");
|
|
121
|
+
setTimeout(() => {
|
|
122
|
+
if (!proc.killed)
|
|
123
|
+
proc.kill("SIGKILL");
|
|
124
|
+
}, 5000);
|
|
125
|
+
}, timeoutMs);
|
|
126
|
+
}
|
|
127
|
+
const processLine = (line) => {
|
|
128
|
+
const entry = parseJsonlLine(line);
|
|
129
|
+
if (!entry)
|
|
130
|
+
return;
|
|
131
|
+
log.push(entry);
|
|
132
|
+
if (entry.type === "message_end") {
|
|
133
|
+
const text = extractAssistantText(log);
|
|
134
|
+
if (text)
|
|
135
|
+
lastAssistantText = text;
|
|
136
|
+
}
|
|
137
|
+
onUpdate?.({
|
|
138
|
+
content: [
|
|
139
|
+
{
|
|
140
|
+
type: "text",
|
|
141
|
+
text: lastAssistantText || `(running… ${log.length} events)`,
|
|
142
|
+
},
|
|
143
|
+
],
|
|
144
|
+
details: {
|
|
145
|
+
exitCode: null,
|
|
146
|
+
log,
|
|
147
|
+
duration: Date.now() - startTime,
|
|
148
|
+
model: extractModel(log),
|
|
149
|
+
usage: extractUsage(log),
|
|
150
|
+
},
|
|
151
|
+
});
|
|
152
|
+
};
|
|
153
|
+
proc.stdout.on("data", (data) => {
|
|
154
|
+
buffer += data.toString();
|
|
155
|
+
const lines = buffer.split("\n");
|
|
156
|
+
buffer = lines.pop() ?? "";
|
|
157
|
+
for (const line of lines)
|
|
158
|
+
processLine(line);
|
|
159
|
+
});
|
|
160
|
+
proc.stderr.on("data", (data) => {
|
|
161
|
+
const text = data.toString().trim();
|
|
162
|
+
if (text) {
|
|
163
|
+
log.push({ type: "stderr", timestamp: Date.now(), raw: text });
|
|
164
|
+
}
|
|
165
|
+
});
|
|
166
|
+
proc.on("close", (code) => {
|
|
167
|
+
if (timeoutHandle)
|
|
168
|
+
clearTimeout(timeoutHandle);
|
|
169
|
+
if (buffer.trim())
|
|
170
|
+
processLine(buffer);
|
|
171
|
+
if (timedOut) {
|
|
172
|
+
log.push({
|
|
173
|
+
type: "timeout",
|
|
174
|
+
timestamp: Date.now(),
|
|
175
|
+
raw: `Timed out after ${params.timeout ?? 300}s`,
|
|
176
|
+
});
|
|
177
|
+
}
|
|
178
|
+
resolve(timedOut ? null : (code ?? 1));
|
|
179
|
+
});
|
|
180
|
+
proc.on("error", (err) => {
|
|
181
|
+
if (timeoutHandle)
|
|
182
|
+
clearTimeout(timeoutHandle);
|
|
183
|
+
log.push({
|
|
184
|
+
type: "error",
|
|
185
|
+
timestamp: Date.now(),
|
|
186
|
+
raw: err.message,
|
|
187
|
+
});
|
|
188
|
+
resolve(null);
|
|
189
|
+
});
|
|
190
|
+
if (signal) {
|
|
191
|
+
const onAbort = () => {
|
|
192
|
+
proc.kill("SIGTERM");
|
|
193
|
+
setTimeout(() => {
|
|
194
|
+
if (!proc.killed)
|
|
195
|
+
proc.kill("SIGKILL");
|
|
196
|
+
}, 5000);
|
|
197
|
+
};
|
|
198
|
+
if (signal.aborted)
|
|
199
|
+
onAbort();
|
|
200
|
+
else
|
|
201
|
+
signal.addEventListener("abort", onAbort, { once: true });
|
|
202
|
+
}
|
|
203
|
+
});
|
|
204
|
+
const duration = Date.now() - startTime;
|
|
205
|
+
const assistantText = extractAssistantText(log);
|
|
206
|
+
const usage = extractUsage(log);
|
|
207
|
+
const model = extractModel(log);
|
|
208
|
+
const fullLog = log.map((e) => e.raw).join("\n");
|
|
209
|
+
const summary = [
|
|
210
|
+
`## Docker E2E Run Complete`,
|
|
211
|
+
``,
|
|
212
|
+
`**Exit code:** ${exitCode ?? "killed"}`,
|
|
213
|
+
`**Duration:** ${(duration / 1000).toFixed(1)}s`,
|
|
214
|
+
model ? `**Model:** ${model}` : null,
|
|
215
|
+
usage.turns > 0
|
|
216
|
+
? `**Usage:** ${usage.turns} turns, ↑${usage.input} ↓${usage.output} tokens, $${usage.cost.toFixed(4)}`
|
|
217
|
+
: null,
|
|
218
|
+
``,
|
|
219
|
+
`### Output`,
|
|
220
|
+
assistantText || "(no assistant output)",
|
|
221
|
+
``,
|
|
222
|
+
`### Full Log (${log.length} events)`,
|
|
223
|
+
"```",
|
|
224
|
+
fullLog,
|
|
225
|
+
"```",
|
|
226
|
+
]
|
|
227
|
+
.filter((l) => l !== null)
|
|
228
|
+
.join("\n");
|
|
229
|
+
return {
|
|
230
|
+
content: [{ type: "text", text: summary }],
|
|
231
|
+
details: { exitCode, log, duration, model, usage },
|
|
232
|
+
};
|
|
233
|
+
},
|
|
234
|
+
};
|
|
235
|
+
//# sourceMappingURL=docker-e2e.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"docker-e2e.js","sourceRoot":"","sources":["../src/docker-e2e.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAM3C,OAAO,EAAE,IAAI,EAAe,MAAM,SAAS,CAAC;AAE5C,MAAM,YAAY,GAAG,cAAc,CAAC;AAEpC,MAAM,eAAe,GAAG,IAAI,CAAC,MAAM,CAAC;IAClC,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,2CAA2C,EAAE,CAAC;IAC/E,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC;QACjB,WAAW,EAAE,uCAAuC;KACrD,CAAC;IACF,KAAK,EAAE,IAAI,CAAC,QAAQ,CAClB,IAAI,CAAC,MAAM,CAAC;QACV,WAAW,EAAE,0EAA0E;KACxF,CAAC,CACH;IACD,OAAO,EAAE,IAAI,CAAC,QAAQ,CACpB,IAAI,CAAC,MAAM,CAAC;QACV,WAAW,EAAE,kDAAkD;KAChE,CAAC,CACH;CACF,CAAC,CAAC;AAyBH,MAAM,UAAU,cAAc,CAAC,IAAY;IACzC,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,IAAI,CAAC,OAAO;QAAE,OAAO,IAAI,CAAC;IAC1B,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QACnC,OAAO,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,SAAS,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC;IACjF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC;IAC9D,CAAC;AACH,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,GAAkB;IACrD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,KAAK,IAAI,GAAG,EAAE,CAAC;QACxB,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa;YAAE,SAAS;QAC3C,IAAI,CAAC;YACH,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACpC,MAAM,GAAG,GAAG,KAAK,CAAC,OAAO,CAAC;YAC1B,IAAI,GAAG,EAAE,IAAI,KAAK,WAAW;gBAAE,SAAS;YACxC,KAAK,MAAM,IAAI,IAAI,GAAG,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;gBACrC,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM;oBAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClD,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,iBAAiB;QACnB,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,GAAkB;IAC7C,MAAM,KAAK,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;IACtF,KAAK,MAAM,KAAK,IAAI,GAAG,EAAE,CAAC;QACxB,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa;YAAE,SAAS;QAC3C,IAAI,CAAC;YACH,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACpC,MAAM,GAAG,GAAG,KAAK,CAAC,OAAO,CAAC;YAC1B,IAAI,GAAG,EAAE,IAAI,KAAK,WAAW;gBAAE,SAAS;YACxC,KAAK,CAAC,KAAK,EAAE,CAAC;YACd,MAAM,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC;YACpB,IAAI,CAAC,EAAE,CAAC;gBACN,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC;gBAC5B,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC;gBAC9B,KAAK,CAAC,SAAS,IAAI,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC;gBACpC,KAAK,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC;gBACtC,KAAK,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,IAAI,CAAC,CAAC;YACnC,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,OAAO;QACT,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,GAAkB;IAC7C,KAAK,MAAM,KAAK,IAAI,GAAG,EAAE,CAAC;QACxB,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa;YAAE,SAAS;QAC3C,IAAI,CAAC;YACH,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACpC,IAAI,KAAK,CAAC,OAAO,EAAE,KAAK;gBAAE,OAAO,KAAK,CAAC,OAAO,CAAC,KAAe,CAAC;QACjE,CAAC;QAAC,MAAM,CAAC;YACP,OAAO;QACT,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,KAAa,EAAE,MAAsB;IACnE,MAAM,IAAI,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,gBAAgB,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IAEnE,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QACjB,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,YAAY,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IAC9C,CAAC;IAED,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC;IAC9B,OAAO,IAAI,CAAC;AACd,CAAC;AAED,MAAM,CAAC,MAAM,aAAa,GAAG;IAC3B,IAAI,EAAE,YAAY;IAClB,KAAK,EAAE,YAAY;IACnB,WAAW,EACT,2IAA2I;IAC7I,aAAa,EAAE,8EAA8E;IAC7F,UAAU,EAAE,eAAe;IAE3B,KAAK,CAAC,OAAO,CACX,WAAmB,EACnB,MAAsB,EACtB,MAA+B,EAC/B,QAA+D,EAC/D,IAAsB;QAEtB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,SAAS,GAAG,CAAC,MAAM,CAAC,OAAO,IAAI,GAAG,CAAC,GAAG,IAAI,CAAC;QACjD,MAAM,GAAG,GAAkB,EAAE,CAAC;QAC9B,IAAI,iBAAiB,GAAG,EAAE,CAAC;QAE3B,MAAM,UAAU,GAAG,eAAe,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;QAEzD,MAAM,QAAQ,GAAG,MAAM,IAAI,OAAO,CAAgB,CAAC,OAAO,EAAE,EAAE;YAC5D,MAAM,IAAI,GAAG,KAAK,CAAC,QAAQ,EAAE,UAAU,EAAE;gBACvC,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC;aAClC,CAAC,CAAC;YAEH,IAAI,MAAM,GAAG,EAAE,CAAC;YAChB,IAAI,QAAQ,GAAG,KAAK,CAAC;YACrB,IAAI,aAAyC,CAAC;YAE9C,IAAI,SAAS,GAAG,CAAC,EAAE,CAAC;gBAClB,aAAa,GAAG,UAAU,CAAC,GAAG,EAAE;oBAC9B,QAAQ,GAAG,IAAI,CAAC;oBAChB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACrB,UAAU,CAAC,GAAG,EAAE;wBACd,IAAI,CAAC,IAAI,CAAC,MAAM;4BAAE,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACzC,CAAC,EAAE,IAAI,CAAC,CAAC;gBACX,CAAC,EAAE,SAAS,CAAC,CAAC;YAChB,CAAC;YAED,MAAM,WAAW,GAAG,CAAC,IAAY,EAAE,EAAE;gBACnC,MAAM,KAAK,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;gBACnC,IAAI,CAAC,KAAK;oBAAE,OAAO;gBACnB,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBAEhB,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;oBACjC,MAAM,IAAI,GAAG,oBAAoB,CAAC,GAAG,CAAC,CAAC;oBACvC,IAAI,IAAI;wBAAE,iBAAiB,GAAG,IAAI,CAAC;gBACrC,CAAC;gBAED,QAAQ,EAAE,CAAC;oBACT,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EAAE,iBAAiB,IAAI,aAAa,GAAG,CAAC,MAAM,UAAU;yBAC7D;qBACF;oBACD,OAAO,EAAE;wBACP,QAAQ,EAAE,IAAI;wBACd,GAAG;wBACH,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,KAAK,EAAE,YAAY,CAAC,GAAG,CAAC;wBACxB,KAAK,EAAE,YAAY,CAAC,GAAG,CAAC;qBACzB;iBACF,CAAC,CAAC;YACL,CAAC,CAAC;YAEF,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAY,EAAE,EAAE;gBACtC,MAAM,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;gBAC1B,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBACjC,MAAM,GAAG,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC;gBAC3B,KAAK,MAAM,IAAI,IAAI,KAAK;oBAAE,WAAW,CAAC,IAAI,CAAC,CAAC;YAC9C,CAAC,CAAC,CAAC;YAEH,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAY,EAAE,EAAE;gBACtC,MAAM,IAAI,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,CAAC;gBACpC,IAAI,IAAI,EAAE,CAAC;oBACT,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC;gBACjE,CAAC;YACH,CAAC,CAAC,CAAC;YAEH,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE;gBACxB,IAAI,aAAa;oBAAE,YAAY,CAAC,aAAa,CAAC,CAAC;gBAC/C,IAAI,MAAM,CAAC,IAAI,EAAE;oBAAE,WAAW,CAAC,MAAM,CAAC,CAAC;gBACvC,IAAI,QAAQ,EAAE,CAAC;oBACb,GAAG,CAAC,IAAI,CAAC;wBACP,IAAI,EAAE,SAAS;wBACf,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;wBACrB,GAAG,EAAE,mBAAmB,MAAM,CAAC,OAAO,IAAI,GAAG,GAAG;qBACjD,CAAC,CAAC;gBACL,CAAC;gBACD,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;YACzC,CAAC,CAAC,CAAC;YAEH,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;gBACvB,IAAI,aAAa;oBAAE,YAAY,CAAC,aAAa,CAAC,CAAC;gBAC/C,GAAG,CAAC,IAAI,CAAC;oBACP,IAAI,EAAE,OAAO;oBACb,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;oBACrB,GAAG,EAAE,GAAG,CAAC,OAAO;iBACjB,CAAC,CAAC;gBACH,OAAO,CAAC,IAAI,CAAC,CAAC;YAChB,CAAC,CAAC,CAAC;YAEH,IAAI,MAAM,EAAE,CAAC;gBACX,MAAM,OAAO,GAAG,GAAG,EAAE;oBACnB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACrB,UAAU,CAAC,GAAG,EAAE;wBACd,IAAI,CAAC,IAAI,CAAC,MAAM;4BAAE,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACzC,CAAC,EAAE,IAAI,CAAC,CAAC;gBACX,CAAC,CAAC;gBACF,IAAI,MAAM,CAAC,OAAO;oBAAE,OAAO,EAAE,CAAC;;oBACzB,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YACjE,CAAC;QACH,CAAC,CAAC,CAAC;QAEH,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QACxC,MAAM,aAAa,GAAG,oBAAoB,CAAC,GAAG,CAAC,CAAC;QAChD,MAAM,KAAK,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC;QAChC,MAAM,KAAK,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC;QAChC,MAAM,OAAO,GAAG,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEjD,MAAM,OAAO,GAAG;YACd,4BAA4B;YAC5B,EAAE;YACF,kBAAkB,QAAQ,IAAI,QAAQ,EAAE;YACxC,iBAAiB,CAAC,QAAQ,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAChD,KAAK,CAAC,CAAC,CAAC,cAAc,KAAK,EAAE,CAAC,CAAC,CAAC,IAAI;YACpC,KAAK,CAAC,KAAK,GAAG,CAAC;gBACb,CAAC,CAAC,cAAc,KAAK,CAAC,KAAK,YAAY,KAAK,CAAC,KAAK,KAAK,KAAK,CAAC,MAAM,aAAa,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;gBACvG,CAAC,CAAC,IAAI;YACR,EAAE;YACF,YAAY;YACZ,aAAa,IAAI,uBAAuB;YACxC,EAAE;YACF,iBAAiB,GAAG,CAAC,MAAM,UAAU;YACrC,KAAK;YACL,OAAO;YACP,KAAK;SACN;aACE,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,IAAI,CAAC;aACzB,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,OAAO;YACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;YAC1C,OAAO,EAAE,EAAE,QAAQ,EAAE,GAAG,EAAE,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE;SACnD,CAAC;IACJ,CAAC;CACF,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"docker-e2e.test.d.ts","sourceRoot":"","sources":["../src/docker-e2e.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,176 @@
|
|
|
1
|
+
import { describe, it, expect } from "vitest";
|
|
2
|
+
import { parseJsonlLine, extractAssistantText, extractUsage, extractModel, buildDockerArgs, } from "./docker-e2e.js";
|
|
3
|
+
describe("parseJsonlLine", () => {
|
|
4
|
+
it("parses valid JSON event", () => {
|
|
5
|
+
const line = '{"type":"message_start","message":{"role":"assistant"}}';
|
|
6
|
+
const result = parseJsonlLine(line);
|
|
7
|
+
expect(result).not.toBeNull();
|
|
8
|
+
expect(result?.type).toBe("message_start");
|
|
9
|
+
expect(result?.raw).toBe(line);
|
|
10
|
+
});
|
|
11
|
+
it("returns raw type for non-JSON", () => {
|
|
12
|
+
const result = parseJsonlLine("some plain text output");
|
|
13
|
+
expect(result).not.toBeNull();
|
|
14
|
+
expect(result?.type).toBe("raw");
|
|
15
|
+
expect(result?.raw).toBe("some plain text output");
|
|
16
|
+
});
|
|
17
|
+
it("returns null for empty lines", () => {
|
|
18
|
+
expect(parseJsonlLine("")).toBeNull();
|
|
19
|
+
expect(parseJsonlLine(" ")).toBeNull();
|
|
20
|
+
});
|
|
21
|
+
it("defaults to unknown type when type field missing", () => {
|
|
22
|
+
const result = parseJsonlLine('{"data":"hello"}');
|
|
23
|
+
expect(result?.type).toBe("unknown");
|
|
24
|
+
});
|
|
25
|
+
});
|
|
26
|
+
describe("extractAssistantText", () => {
|
|
27
|
+
it("extracts text from message_end events", () => {
|
|
28
|
+
const log = [
|
|
29
|
+
{
|
|
30
|
+
type: "message_end",
|
|
31
|
+
timestamp: 0,
|
|
32
|
+
raw: JSON.stringify({
|
|
33
|
+
type: "message_end",
|
|
34
|
+
message: {
|
|
35
|
+
role: "assistant",
|
|
36
|
+
content: [{ type: "text", text: "Hello world" }],
|
|
37
|
+
},
|
|
38
|
+
}),
|
|
39
|
+
},
|
|
40
|
+
];
|
|
41
|
+
expect(extractAssistantText(log)).toBe("Hello world");
|
|
42
|
+
});
|
|
43
|
+
it("concatenates multiple assistant messages", () => {
|
|
44
|
+
const makeEnd = (text) => ({
|
|
45
|
+
type: "message_end",
|
|
46
|
+
timestamp: 0,
|
|
47
|
+
raw: JSON.stringify({
|
|
48
|
+
type: "message_end",
|
|
49
|
+
message: {
|
|
50
|
+
role: "assistant",
|
|
51
|
+
content: [{ type: "text", text }],
|
|
52
|
+
},
|
|
53
|
+
}),
|
|
54
|
+
});
|
|
55
|
+
const log = [makeEnd("First"), makeEnd("Second")];
|
|
56
|
+
expect(extractAssistantText(log)).toBe("First\nSecond");
|
|
57
|
+
});
|
|
58
|
+
it("ignores non-assistant messages", () => {
|
|
59
|
+
const log = [
|
|
60
|
+
{
|
|
61
|
+
type: "message_end",
|
|
62
|
+
timestamp: 0,
|
|
63
|
+
raw: JSON.stringify({
|
|
64
|
+
type: "message_end",
|
|
65
|
+
message: {
|
|
66
|
+
role: "user",
|
|
67
|
+
content: [{ type: "text", text: "ignored" }],
|
|
68
|
+
},
|
|
69
|
+
}),
|
|
70
|
+
},
|
|
71
|
+
];
|
|
72
|
+
expect(extractAssistantText(log)).toBe("");
|
|
73
|
+
});
|
|
74
|
+
it("ignores non-message_end events", () => {
|
|
75
|
+
const log = [{ type: "message_start", timestamp: 0, raw: "{}" }];
|
|
76
|
+
expect(extractAssistantText(log)).toBe("");
|
|
77
|
+
});
|
|
78
|
+
});
|
|
79
|
+
describe("extractUsage", () => {
|
|
80
|
+
it("accumulates usage from assistant messages", () => {
|
|
81
|
+
const log = [
|
|
82
|
+
{
|
|
83
|
+
type: "message_end",
|
|
84
|
+
timestamp: 0,
|
|
85
|
+
raw: JSON.stringify({
|
|
86
|
+
type: "message_end",
|
|
87
|
+
message: {
|
|
88
|
+
role: "assistant",
|
|
89
|
+
content: [],
|
|
90
|
+
usage: {
|
|
91
|
+
input: 100,
|
|
92
|
+
output: 50,
|
|
93
|
+
cacheRead: 10,
|
|
94
|
+
cacheWrite: 5,
|
|
95
|
+
cost: { total: 0.01 },
|
|
96
|
+
},
|
|
97
|
+
},
|
|
98
|
+
}),
|
|
99
|
+
},
|
|
100
|
+
{
|
|
101
|
+
type: "message_end",
|
|
102
|
+
timestamp: 0,
|
|
103
|
+
raw: JSON.stringify({
|
|
104
|
+
type: "message_end",
|
|
105
|
+
message: {
|
|
106
|
+
role: "assistant",
|
|
107
|
+
content: [],
|
|
108
|
+
usage: {
|
|
109
|
+
input: 200,
|
|
110
|
+
output: 100,
|
|
111
|
+
cacheRead: 20,
|
|
112
|
+
cacheWrite: 10,
|
|
113
|
+
cost: { total: 0.02 },
|
|
114
|
+
},
|
|
115
|
+
},
|
|
116
|
+
}),
|
|
117
|
+
},
|
|
118
|
+
];
|
|
119
|
+
const usage = extractUsage(log);
|
|
120
|
+
expect(usage.turns).toBe(2);
|
|
121
|
+
expect(usage.input).toBe(300);
|
|
122
|
+
expect(usage.output).toBe(150);
|
|
123
|
+
expect(usage.cacheRead).toBe(30);
|
|
124
|
+
expect(usage.cacheWrite).toBe(15);
|
|
125
|
+
expect(usage.cost).toBeCloseTo(0.03);
|
|
126
|
+
});
|
|
127
|
+
it("returns zeroes for empty log", () => {
|
|
128
|
+
const usage = extractUsage([]);
|
|
129
|
+
expect(usage.turns).toBe(0);
|
|
130
|
+
expect(usage.input).toBe(0);
|
|
131
|
+
});
|
|
132
|
+
});
|
|
133
|
+
describe("extractModel", () => {
|
|
134
|
+
it("extracts model from message_end", () => {
|
|
135
|
+
const log = [
|
|
136
|
+
{
|
|
137
|
+
type: "message_end",
|
|
138
|
+
timestamp: 0,
|
|
139
|
+
raw: JSON.stringify({
|
|
140
|
+
type: "message_end",
|
|
141
|
+
message: { model: "claude-sonnet-4-5", role: "assistant", content: [] },
|
|
142
|
+
}),
|
|
143
|
+
},
|
|
144
|
+
];
|
|
145
|
+
expect(extractModel(log)).toBe("claude-sonnet-4-5");
|
|
146
|
+
});
|
|
147
|
+
it("returns undefined for empty log", () => {
|
|
148
|
+
expect(extractModel([])).toBeUndefined();
|
|
149
|
+
});
|
|
150
|
+
});
|
|
151
|
+
describe("buildDockerArgs", () => {
|
|
152
|
+
it("builds basic args", () => {
|
|
153
|
+
const args = buildDockerArgs("my-image", {
|
|
154
|
+
task: "Fix the bug",
|
|
155
|
+
token: "ghp_abc123",
|
|
156
|
+
});
|
|
157
|
+
expect(args).toEqual([
|
|
158
|
+
"run",
|
|
159
|
+
"--rm",
|
|
160
|
+
"-e",
|
|
161
|
+
"GITHUB_TOKEN=ghp_abc123",
|
|
162
|
+
"my-image",
|
|
163
|
+
"Fix the bug",
|
|
164
|
+
]);
|
|
165
|
+
});
|
|
166
|
+
it("includes model env var when specified", () => {
|
|
167
|
+
const args = buildDockerArgs("my-image", {
|
|
168
|
+
task: "Fix it",
|
|
169
|
+
token: "ghp_abc",
|
|
170
|
+
model: "claude-opus-4-5",
|
|
171
|
+
});
|
|
172
|
+
expect(args).toContain("-e");
|
|
173
|
+
expect(args).toContain("PI_MODEL=claude-opus-4-5");
|
|
174
|
+
});
|
|
175
|
+
});
|
|
176
|
+
//# sourceMappingURL=docker-e2e.test.js.map
|