@huggingface/tasks 0.21.4 → 0.21.6
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/dist/commonjs/agent-harnesses.d.ts +196 -0
- package/dist/commonjs/agent-harnesses.d.ts.map +1 -0
- package/dist/commonjs/agent-harnesses.js +128 -0
- package/dist/commonjs/eval.d.ts +10 -0
- package/dist/commonjs/eval.d.ts.map +1 -1
- package/dist/commonjs/eval.js +10 -0
- package/dist/commonjs/index.d.ts +2 -0
- package/dist/commonjs/index.d.ts.map +1 -1
- package/dist/commonjs/index.js +4 -1
- package/dist/esm/agent-harnesses.d.ts +196 -0
- package/dist/esm/agent-harnesses.d.ts.map +1 -0
- package/dist/esm/agent-harnesses.js +125 -0
- package/dist/esm/eval.d.ts +10 -0
- package/dist/esm/eval.d.ts.map +1 -1
- package/dist/esm/eval.js +10 -0
- package/dist/esm/index.d.ts +2 -0
- package/dist/esm/index.d.ts.map +1 -1
- package/dist/esm/index.js +1 -0
- package/package.json +1 -1
- package/src/agent-harnesses.ts +167 -0
- package/src/eval.ts +12 -0
- package/src/index.ts +3 -0
|
@@ -0,0 +1,196 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Registry of AI coding agents / harnesses known to use the Hugging Face Hub.
|
|
3
|
+
*
|
|
4
|
+
* To add your harness, append an entry below keyed by its `id` (the name used
|
|
5
|
+
* when reporting Hub activity), and list the environment variable(s) that
|
|
6
|
+
* identify it.
|
|
7
|
+
*/
|
|
8
|
+
export interface AgentHarness {
|
|
9
|
+
/**
|
|
10
|
+
* Human-readable name of the harness, e.g. displayed in a leaderboard.
|
|
11
|
+
*/
|
|
12
|
+
prettyLabel: string;
|
|
13
|
+
/**
|
|
14
|
+
* URL to the harness's code repository (usually on GitHub).
|
|
15
|
+
*/
|
|
16
|
+
repoUrl?: string;
|
|
17
|
+
/**
|
|
18
|
+
* URL to the harness's documentation or website.
|
|
19
|
+
*/
|
|
20
|
+
docsUrl?: string;
|
|
21
|
+
/**
|
|
22
|
+
* Short description of the harness.
|
|
23
|
+
*/
|
|
24
|
+
description?: string;
|
|
25
|
+
/**
|
|
26
|
+
* Environment variable(s) that identify this harness, mapped to the value
|
|
27
|
+
* pattern they must match. Detection matches if ANY entry matches.
|
|
28
|
+
*
|
|
29
|
+
* The value pattern is one of:
|
|
30
|
+
* - `"*"`: the variable is set to any (non-empty) value
|
|
31
|
+
* - `"<value>"`: the variable equals this exact value
|
|
32
|
+
* - `"<prefix>*"`: the variable value starts with `<prefix>` (fuzzy match, resolved client-side)
|
|
33
|
+
*
|
|
34
|
+
* If not provided, the harness is detected through the standard AI_AGENT / AGENT variables only.
|
|
35
|
+
*/
|
|
36
|
+
envVars?: Record<string, string>;
|
|
37
|
+
}
|
|
38
|
+
/**
|
|
39
|
+
* Standard environment variables that any tool can set to identify itself.
|
|
40
|
+
* When one of these is set, its value is used directly as the agent `id`
|
|
41
|
+
* (matched against the keys of `AGENT_HARNESSES`); unrecognized values are
|
|
42
|
+
* reported as `"unknown"`.
|
|
43
|
+
*/
|
|
44
|
+
export declare const STANDARD_AGENT_ENV_VARS: readonly ["AI_AGENT", "AGENT"];
|
|
45
|
+
/**
|
|
46
|
+
* Add your new agent harness here.
|
|
47
|
+
*
|
|
48
|
+
* /!\ IMPORTANT
|
|
49
|
+
*
|
|
50
|
+
* Insertion order matters for detection priority: harnesses are checked from
|
|
51
|
+
* top to bottom and the first match wins. In particular, `cowork` must stay
|
|
52
|
+
* before `claude-code` so the more specific signal takes priority when both
|
|
53
|
+
* `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
54
|
+
*/
|
|
55
|
+
export declare const AGENT_HARNESSES: {
|
|
56
|
+
antigravity: {
|
|
57
|
+
prettyLabel: string;
|
|
58
|
+
docsUrl: string;
|
|
59
|
+
description: string;
|
|
60
|
+
envVars: {
|
|
61
|
+
ANTIGRAVITY_AGENT: string;
|
|
62
|
+
};
|
|
63
|
+
};
|
|
64
|
+
"augment-cli": {
|
|
65
|
+
prettyLabel: string;
|
|
66
|
+
repoUrl: string;
|
|
67
|
+
docsUrl: string;
|
|
68
|
+
description: string;
|
|
69
|
+
envVars: {
|
|
70
|
+
AUGMENT_AGENT: string;
|
|
71
|
+
};
|
|
72
|
+
};
|
|
73
|
+
cline: {
|
|
74
|
+
prettyLabel: string;
|
|
75
|
+
repoUrl: string;
|
|
76
|
+
docsUrl: string;
|
|
77
|
+
description: string;
|
|
78
|
+
envVars: {
|
|
79
|
+
CLINE_ACTIVE: string;
|
|
80
|
+
};
|
|
81
|
+
};
|
|
82
|
+
cowork: {
|
|
83
|
+
prettyLabel: string;
|
|
84
|
+
docsUrl: string;
|
|
85
|
+
description: string;
|
|
86
|
+
envVars: {
|
|
87
|
+
CLAUDE_CODE_IS_COWORK: string;
|
|
88
|
+
};
|
|
89
|
+
};
|
|
90
|
+
"claude-code": {
|
|
91
|
+
prettyLabel: string;
|
|
92
|
+
repoUrl: string;
|
|
93
|
+
docsUrl: string;
|
|
94
|
+
description: string;
|
|
95
|
+
envVars: {
|
|
96
|
+
CLAUDECODE: string;
|
|
97
|
+
CLAUDE_CODE: string;
|
|
98
|
+
};
|
|
99
|
+
};
|
|
100
|
+
codex: {
|
|
101
|
+
prettyLabel: string;
|
|
102
|
+
repoUrl: string;
|
|
103
|
+
docsUrl: string;
|
|
104
|
+
description: string;
|
|
105
|
+
envVars: {
|
|
106
|
+
CODEX_SANDBOX: string;
|
|
107
|
+
CODEX_CI: string;
|
|
108
|
+
CODEX_THREAD_ID: string;
|
|
109
|
+
};
|
|
110
|
+
};
|
|
111
|
+
"cursor-cli": {
|
|
112
|
+
prettyLabel: string;
|
|
113
|
+
docsUrl: string;
|
|
114
|
+
description: string;
|
|
115
|
+
envVars: {
|
|
116
|
+
CURSOR_AGENT: string;
|
|
117
|
+
};
|
|
118
|
+
};
|
|
119
|
+
cursor: {
|
|
120
|
+
prettyLabel: string;
|
|
121
|
+
docsUrl: string;
|
|
122
|
+
description: string;
|
|
123
|
+
envVars: {
|
|
124
|
+
CURSOR_TRACE_ID: string;
|
|
125
|
+
};
|
|
126
|
+
};
|
|
127
|
+
"github-copilot": {
|
|
128
|
+
prettyLabel: string;
|
|
129
|
+
docsUrl: string;
|
|
130
|
+
description: string;
|
|
131
|
+
envVars: {
|
|
132
|
+
COPILOT_MODEL: string;
|
|
133
|
+
COPILOT_ALLOW_ALL: string;
|
|
134
|
+
COPILOT_GITHUB_TOKEN: string;
|
|
135
|
+
};
|
|
136
|
+
};
|
|
137
|
+
goose: {
|
|
138
|
+
prettyLabel: string;
|
|
139
|
+
repoUrl: string;
|
|
140
|
+
docsUrl: string;
|
|
141
|
+
description: string;
|
|
142
|
+
envVars: {
|
|
143
|
+
GOOSE_TERMINAL: string;
|
|
144
|
+
};
|
|
145
|
+
};
|
|
146
|
+
openclaw: {
|
|
147
|
+
prettyLabel: string;
|
|
148
|
+
repoUrl: string;
|
|
149
|
+
docsUrl: string;
|
|
150
|
+
description: string;
|
|
151
|
+
envVars: {
|
|
152
|
+
OPENCLAW_SHELL: string;
|
|
153
|
+
};
|
|
154
|
+
};
|
|
155
|
+
opencode: {
|
|
156
|
+
prettyLabel: string;
|
|
157
|
+
repoUrl: string;
|
|
158
|
+
docsUrl: string;
|
|
159
|
+
description: string;
|
|
160
|
+
envVars: {
|
|
161
|
+
OPENCODE_CLIENT: string;
|
|
162
|
+
};
|
|
163
|
+
};
|
|
164
|
+
pi: {
|
|
165
|
+
prettyLabel: string;
|
|
166
|
+
repoUrl: string;
|
|
167
|
+
docsUrl: string;
|
|
168
|
+
description: string;
|
|
169
|
+
envVars: {
|
|
170
|
+
PI_CODING_AGENT: string;
|
|
171
|
+
};
|
|
172
|
+
};
|
|
173
|
+
replit: {
|
|
174
|
+
prettyLabel: string;
|
|
175
|
+
docsUrl: string;
|
|
176
|
+
description: string;
|
|
177
|
+
envVars: {
|
|
178
|
+
REPL_ID: string;
|
|
179
|
+
};
|
|
180
|
+
};
|
|
181
|
+
trae: {
|
|
182
|
+
prettyLabel: string;
|
|
183
|
+
docsUrl: string;
|
|
184
|
+
description: string;
|
|
185
|
+
envVars: {
|
|
186
|
+
TRAE_AI_SHELL_ID: string;
|
|
187
|
+
};
|
|
188
|
+
};
|
|
189
|
+
devin: {
|
|
190
|
+
prettyLabel: string;
|
|
191
|
+
docsUrl: string;
|
|
192
|
+
description: string;
|
|
193
|
+
};
|
|
194
|
+
};
|
|
195
|
+
export type AgentHarnessKey = keyof typeof AGENT_HARNESSES;
|
|
196
|
+
//# sourceMappingURL=agent-harnesses.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-harnesses.d.ts","sourceRoot":"","sources":["../../src/agent-harnesses.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,MAAM,WAAW,YAAY;IAC5B;;OAEG;IACH,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;OAEG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;OAEG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;;;;;;;OAUG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjC;AAED;;;;;GAKG;AACH,eAAO,MAAM,uBAAuB,gCAAiC,CAAC;AAEtE;;;;;;;;;GASG;AACH,eAAO,MAAM,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA2GY,CAAC;AAGzC,MAAM,MAAM,eAAe,GAAG,MAAM,OAAO,eAAe,CAAC"}
|
|
@@ -0,0 +1,128 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.AGENT_HARNESSES = exports.STANDARD_AGENT_ENV_VARS = void 0;
|
|
4
|
+
/**
|
|
5
|
+
* Standard environment variables that any tool can set to identify itself.
|
|
6
|
+
* When one of these is set, its value is used directly as the agent `id`
|
|
7
|
+
* (matched against the keys of `AGENT_HARNESSES`); unrecognized values are
|
|
8
|
+
* reported as `"unknown"`.
|
|
9
|
+
*/
|
|
10
|
+
exports.STANDARD_AGENT_ENV_VARS = ["AI_AGENT", "AGENT"];
|
|
11
|
+
/**
|
|
12
|
+
* Add your new agent harness here.
|
|
13
|
+
*
|
|
14
|
+
* /!\ IMPORTANT
|
|
15
|
+
*
|
|
16
|
+
* Insertion order matters for detection priority: harnesses are checked from
|
|
17
|
+
* top to bottom and the first match wins. In particular, `cowork` must stay
|
|
18
|
+
* before `claude-code` so the more specific signal takes priority when both
|
|
19
|
+
* `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
20
|
+
*/
|
|
21
|
+
exports.AGENT_HARNESSES = {
|
|
22
|
+
antigravity: {
|
|
23
|
+
prettyLabel: "Antigravity",
|
|
24
|
+
docsUrl: "https://antigravity.google",
|
|
25
|
+
description: "Agentic development platform from Google built around Gemini.",
|
|
26
|
+
envVars: { ANTIGRAVITY_AGENT: "*" },
|
|
27
|
+
},
|
|
28
|
+
"augment-cli": {
|
|
29
|
+
prettyLabel: "Augment CLI",
|
|
30
|
+
repoUrl: "https://github.com/augmentcode/auggie",
|
|
31
|
+
docsUrl: "https://www.augmentcode.com",
|
|
32
|
+
description: "Auggie, the command-line coding agent from Augment Code.",
|
|
33
|
+
envVars: { AUGMENT_AGENT: "*" },
|
|
34
|
+
},
|
|
35
|
+
cline: {
|
|
36
|
+
prettyLabel: "Cline",
|
|
37
|
+
repoUrl: "https://github.com/cline/cline",
|
|
38
|
+
docsUrl: "https://cline.bot",
|
|
39
|
+
description: "Open-source autonomous coding agent for VS Code.",
|
|
40
|
+
envVars: { CLINE_ACTIVE: "*" },
|
|
41
|
+
},
|
|
42
|
+
cowork: {
|
|
43
|
+
// must stay before `claude-code` so the more specific signal takes priority when both `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
44
|
+
prettyLabel: "Cowork",
|
|
45
|
+
docsUrl: "https://claude.com/product/cowork",
|
|
46
|
+
description: "Anthropic's agent for autonomous knowledge work, built on top of Claude Code.",
|
|
47
|
+
envVars: { CLAUDE_CODE_IS_COWORK: "*" },
|
|
48
|
+
},
|
|
49
|
+
"claude-code": {
|
|
50
|
+
prettyLabel: "Claude Code",
|
|
51
|
+
repoUrl: "https://github.com/anthropics/claude-code",
|
|
52
|
+
docsUrl: "https://code.claude.com/docs",
|
|
53
|
+
description: "Anthropic's agentic coding tool that lives in your terminal.",
|
|
54
|
+
envVars: { CLAUDECODE: "*", CLAUDE_CODE: "*" },
|
|
55
|
+
},
|
|
56
|
+
codex: {
|
|
57
|
+
prettyLabel: "Codex",
|
|
58
|
+
repoUrl: "https://github.com/openai/codex",
|
|
59
|
+
docsUrl: "https://developers.openai.com/codex",
|
|
60
|
+
description: "OpenAI's lightweight coding agent that runs in your terminal.",
|
|
61
|
+
envVars: { CODEX_SANDBOX: "*", CODEX_CI: "*", CODEX_THREAD_ID: "*" },
|
|
62
|
+
},
|
|
63
|
+
"cursor-cli": {
|
|
64
|
+
// must stay before `cursor` so the more specific signal takes priority: when the CLI runs inside
|
|
65
|
+
// the Cursor editor's terminal, child processes inherit `CURSOR_TRACE_ID` and the CLI sets `CURSOR_AGENT`.
|
|
66
|
+
prettyLabel: "Cursor CLI",
|
|
67
|
+
docsUrl: "https://cursor.com/docs/cli/overview",
|
|
68
|
+
description: "Cursor's coding agent for the command line.",
|
|
69
|
+
envVars: { CURSOR_AGENT: "*" },
|
|
70
|
+
},
|
|
71
|
+
cursor: {
|
|
72
|
+
prettyLabel: "Cursor",
|
|
73
|
+
docsUrl: "https://cursor.com",
|
|
74
|
+
description: "AI-powered code editor.",
|
|
75
|
+
envVars: { CURSOR_TRACE_ID: "*" },
|
|
76
|
+
},
|
|
77
|
+
"github-copilot": {
|
|
78
|
+
prettyLabel: "GitHub Copilot",
|
|
79
|
+
docsUrl: "https://docs.github.com/copilot",
|
|
80
|
+
description: "GitHub's AI coding assistant.",
|
|
81
|
+
envVars: { COPILOT_MODEL: "*", COPILOT_ALLOW_ALL: "*", COPILOT_GITHUB_TOKEN: "*" },
|
|
82
|
+
},
|
|
83
|
+
goose: {
|
|
84
|
+
prettyLabel: "Goose",
|
|
85
|
+
repoUrl: "https://github.com/aaif-goose/goose",
|
|
86
|
+
docsUrl: "https://goose-docs.ai/",
|
|
87
|
+
description: "Open-source, extensible AI agent, originally from Block and now part of the Agentic AI Foundation.",
|
|
88
|
+
envVars: { GOOSE_TERMINAL: "*" },
|
|
89
|
+
},
|
|
90
|
+
openclaw: {
|
|
91
|
+
prettyLabel: "OpenClaw",
|
|
92
|
+
repoUrl: "https://github.com/openclaw/openclaw",
|
|
93
|
+
docsUrl: "https://openclaw.ai",
|
|
94
|
+
description: "Open-source, self-hosted personal AI assistant that runs on your own devices.",
|
|
95
|
+
envVars: { OPENCLAW_SHELL: "*" },
|
|
96
|
+
},
|
|
97
|
+
opencode: {
|
|
98
|
+
prettyLabel: "opencode",
|
|
99
|
+
repoUrl: "https://github.com/anomalyco/opencode",
|
|
100
|
+
docsUrl: "https://opencode.ai",
|
|
101
|
+
description: "Open-source AI coding agent built for the terminal.",
|
|
102
|
+
envVars: { OPENCODE_CLIENT: "*" },
|
|
103
|
+
},
|
|
104
|
+
pi: {
|
|
105
|
+
prettyLabel: "Pi",
|
|
106
|
+
repoUrl: "https://github.com/earendil-works/pi",
|
|
107
|
+
docsUrl: "https://pi.dev",
|
|
108
|
+
description: "Minimal, self-extensible terminal coding agent with a unified multi-provider LLM API.",
|
|
109
|
+
envVars: { PI_CODING_AGENT: "*" },
|
|
110
|
+
},
|
|
111
|
+
replit: {
|
|
112
|
+
prettyLabel: "Replit",
|
|
113
|
+
docsUrl: "https://replit.com",
|
|
114
|
+
description: "Cloud development environment with an AI coding agent.",
|
|
115
|
+
envVars: { REPL_ID: "*" },
|
|
116
|
+
},
|
|
117
|
+
trae: {
|
|
118
|
+
prettyLabel: "Trae",
|
|
119
|
+
docsUrl: "https://trae.ai",
|
|
120
|
+
description: "AI-powered IDE from ByteDance.",
|
|
121
|
+
envVars: { TRAE_AI_SHELL_ID: "*" },
|
|
122
|
+
},
|
|
123
|
+
devin: {
|
|
124
|
+
prettyLabel: "Devin",
|
|
125
|
+
docsUrl: "https://devin.ai",
|
|
126
|
+
description: "Autonomous AI software engineer from Cognition.",
|
|
127
|
+
},
|
|
128
|
+
};
|
package/dist/commonjs/eval.d.ts
CHANGED
|
@@ -32,6 +32,11 @@ export declare const EVALUATION_FRAMEWORKS: {
|
|
|
32
32
|
readonly description: "Harbor is a framework for evaluating and optimizing agents and language models.";
|
|
33
33
|
readonly url: "https://github.com/laude-institute/harbor";
|
|
34
34
|
};
|
|
35
|
+
readonly pier: {
|
|
36
|
+
readonly name: "pier";
|
|
37
|
+
readonly description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.";
|
|
38
|
+
readonly url: "https://github.com/datacurve-ai/pier";
|
|
39
|
+
};
|
|
35
40
|
readonly archipelago: {
|
|
36
41
|
readonly name: "archipelago";
|
|
37
42
|
readonly description: "Archipelago is a system for running and evaluating AI agents against MCP applications.";
|
|
@@ -112,5 +117,10 @@ export declare const EVALUATION_FRAMEWORKS: {
|
|
|
112
117
|
readonly description: "WBench is a comprehensive multi-turn benchmark for interactive video world model evaluation, assessing models across 5 dimensions (video quality, setting adherence, interaction adherence, consistency, physics compliance) and 22 metrics over 289 multi-turn interaction cases.";
|
|
113
118
|
readonly url: "https://github.com/meituan-longcat/WBench";
|
|
114
119
|
};
|
|
120
|
+
readonly nanofold: {
|
|
121
|
+
readonly name: "nanofold";
|
|
122
|
+
readonly description: "nanoFold is a data-efficiency benchmark for protein structure prediction. Its goal is to evaluate models on scenarios with scarce data.";
|
|
123
|
+
readonly url: "https://github.com/ChrisHayduk/nanoFold-Competition";
|
|
124
|
+
};
|
|
115
125
|
};
|
|
116
126
|
//# sourceMappingURL=eval.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB
|
|
1
|
+
{"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAwIxB,CAAC"}
|
package/dist/commonjs/eval.js
CHANGED
|
@@ -35,6 +35,11 @@ exports.EVALUATION_FRAMEWORKS = {
|
|
|
35
35
|
description: "Harbor is a framework for evaluating and optimizing agents and language models.",
|
|
36
36
|
url: "https://github.com/laude-institute/harbor",
|
|
37
37
|
},
|
|
38
|
+
pier: {
|
|
39
|
+
name: "pier",
|
|
40
|
+
description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.",
|
|
41
|
+
url: "https://github.com/datacurve-ai/pier",
|
|
42
|
+
},
|
|
38
43
|
archipelago: {
|
|
39
44
|
name: "archipelago",
|
|
40
45
|
description: "Archipelago is a system for running and evaluating AI agents against MCP applications.",
|
|
@@ -115,4 +120,9 @@ exports.EVALUATION_FRAMEWORKS = {
|
|
|
115
120
|
description: "WBench is a comprehensive multi-turn benchmark for interactive video world model evaluation, assessing models across 5 dimensions (video quality, setting adherence, interaction adherence, consistency, physics compliance) and 22 metrics over 289 multi-turn interaction cases.",
|
|
116
121
|
url: "https://github.com/meituan-longcat/WBench",
|
|
117
122
|
},
|
|
123
|
+
nanofold: {
|
|
124
|
+
name: "nanofold",
|
|
125
|
+
description: "nanoFold is a data-efficiency benchmark for protein structure prediction. Its goal is to evaluate models on scenarios with scarce data.",
|
|
126
|
+
url: "https://github.com/ChrisHayduk/nanoFold-Competition",
|
|
127
|
+
},
|
|
118
128
|
};
|
package/dist/commonjs/index.d.ts
CHANGED
|
@@ -23,4 +23,6 @@ export { KERNEL_LIBRARIES_UI_ELEMENTS } from "./kernel-libraries.js";
|
|
|
23
23
|
export type { KernelLibraryKey, KernelLibraryUiElement } from "./kernel-libraries.js";
|
|
24
24
|
export * from "./inference-providers.js";
|
|
25
25
|
export { EVALUATION_FRAMEWORKS } from "./eval.js";
|
|
26
|
+
export { AGENT_HARNESSES, STANDARD_AGENT_ENV_VARS } from "./agent-harnesses.js";
|
|
27
|
+
export type { AgentHarness, AgentHarnessKey } from "./agent-harnesses.js";
|
|
26
28
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAC7D,OAAO,EAAE,sBAAsB,EAAE,MAAM,4BAA4B,CAAC;AACpE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AACvF,cAAc,kBAAkB,CAAC;AACjC,OAAO,EACN,aAAa,EACb,cAAc,EACd,KAAK,UAAU,EACf,KAAK,YAAY,EACjB,KAAK,YAAY,EACjB,KAAK,QAAQ,EACb,UAAU,EACV,eAAe,EACf,aAAa,EACb,kBAAkB,GAClB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACN,8BAA8B,EAC9B,sBAAsB,EACtB,2BAA2B,GAC3B,MAAM,sBAAsB,CAAC;AAC9B,YAAY,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AAC9E,YAAY,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,YAAY,EAAE,UAAU,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACzF,YAAY,EACX,aAAa,EACb,sBAAsB,EACtB,gCAAgC,EAChC,8BAA8B,EAC9B,kCAAkC,EAClC,uBAAuB,EACvB,sBAAsB,EACtB,oCAAoC,EACpC,gCAAgC,EAChC,2BAA2B,EAC3B,gCAAgC,EAChC,8BAA8B,EAC9B,sBAAsB,EACtB,8BAA8B,EAC9B,mBAAmB,EACnB,sBAAsB,EACtB,yBAAyB,EACzB,8BAA8B,EAC9B,uBAAuB,GACvB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAEhE,cAAc,WAAW,CAAC;AAE1B,OAAO,EACN,KAAK,gBAAgB,EACrB,KAAK,wBAAwB,EAC7B,KAAK,gBAAgB,EACrB,yBAAyB,EACzB,yBAAyB,EACzB,iBAAiB,EACjB,oBAAoB,GACpB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,IAAI,EAAE,sBAAsB,EAAE,MAAM,eAAe,CAAC;AAC7D,YAAY,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,eAAe,CAAC;AAC3D,YAAY,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAC;AAC5D,YAAY,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAC7C,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAE9E,OAAO,EAAE,6BAA6B,EAAE,MAAM,wBAAwB,CAAC;AACvE,YAAY,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAEzF,OAAO,EAAE,4BAA4B,EAAE,MAAM,uBAAuB,CAAC;AACrE,YAAY,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAEtF,cAAc,0BAA0B,CAAC;AAEzC,OAAO,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAC7D,OAAO,EAAE,sBAAsB,EAAE,MAAM,4BAA4B,CAAC;AACpE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AACvF,cAAc,kBAAkB,CAAC;AACjC,OAAO,EACN,aAAa,EACb,cAAc,EACd,KAAK,UAAU,EACf,KAAK,YAAY,EACjB,KAAK,YAAY,EACjB,KAAK,QAAQ,EACb,UAAU,EACV,eAAe,EACf,aAAa,EACb,kBAAkB,GAClB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACN,8BAA8B,EAC9B,sBAAsB,EACtB,2BAA2B,GAC3B,MAAM,sBAAsB,CAAC;AAC9B,YAAY,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AAC9E,YAAY,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,YAAY,EAAE,UAAU,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACzF,YAAY,EACX,aAAa,EACb,sBAAsB,EACtB,gCAAgC,EAChC,8BAA8B,EAC9B,kCAAkC,EAClC,uBAAuB,EACvB,sBAAsB,EACtB,oCAAoC,EACpC,gCAAgC,EAChC,2BAA2B,EAC3B,gCAAgC,EAChC,8BAA8B,EAC9B,sBAAsB,EACtB,8BAA8B,EAC9B,mBAAmB,EACnB,sBAAsB,EACtB,yBAAyB,EACzB,8BAA8B,EAC9B,uBAAuB,GACvB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAEhE,cAAc,WAAW,CAAC;AAE1B,OAAO,EACN,KAAK,gBAAgB,EACrB,KAAK,wBAAwB,EAC7B,KAAK,gBAAgB,EACrB,yBAAyB,EACzB,yBAAyB,EACzB,iBAAiB,EACjB,oBAAoB,GACpB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,IAAI,EAAE,sBAAsB,EAAE,MAAM,eAAe,CAAC;AAC7D,YAAY,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,eAAe,CAAC;AAC3D,YAAY,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAC;AAC5D,YAAY,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAC7C,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAE9E,OAAO,EAAE,6BAA6B,EAAE,MAAM,wBAAwB,CAAC;AACvE,YAAY,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAEzF,OAAO,EAAE,4BAA4B,EAAE,MAAM,uBAAuB,CAAC;AACrE,YAAY,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAEtF,cAAc,0BAA0B,CAAC;AAEzC,OAAO,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC;AAElD,OAAO,EAAE,eAAe,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AAChF,YAAY,EAAE,YAAY,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC"}
|
package/dist/commonjs/index.js
CHANGED
|
@@ -14,7 +14,7 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
|
|
|
14
14
|
for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
|
|
15
15
|
};
|
|
16
16
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
17
|
-
exports.EVALUATION_FRAMEWORKS = exports.KERNEL_LIBRARIES_UI_ELEMENTS = exports.DATASET_LIBRARIES_UI_ELEMENTS = exports.LOCAL_APPS = exports.DEFAULT_MEMORY_OPTIONS = exports.SKUS = exports.getModelInputSnippet = exports.stringifyMessages = exports.stringifyGenerationConfig = exports.inferenceSnippetLanguages = exports.SPECIAL_TOKENS_ATTRIBUTES = exports.MODEL_LIBRARIES_UI_ELEMENTS = exports.ALL_MODEL_LIBRARY_KEYS = exports.ALL_DISPLAY_MODEL_LIBRARY_KEYS = exports.PIPELINE_TYPES_SET = exports.SUBTASK_TYPES = exports.MODALITY_LABELS = exports.MODALITIES = exports.PIPELINE_TYPES = exports.PIPELINE_DATA = exports.MAPPING_DEFAULT_WIDGET = exports.LIBRARY_TASK_MAPPING = void 0;
|
|
17
|
+
exports.STANDARD_AGENT_ENV_VARS = exports.AGENT_HARNESSES = exports.EVALUATION_FRAMEWORKS = exports.KERNEL_LIBRARIES_UI_ELEMENTS = exports.DATASET_LIBRARIES_UI_ELEMENTS = exports.LOCAL_APPS = exports.DEFAULT_MEMORY_OPTIONS = exports.SKUS = exports.getModelInputSnippet = exports.stringifyMessages = exports.stringifyGenerationConfig = exports.inferenceSnippetLanguages = exports.SPECIAL_TOKENS_ATTRIBUTES = exports.MODEL_LIBRARIES_UI_ELEMENTS = exports.ALL_MODEL_LIBRARY_KEYS = exports.ALL_DISPLAY_MODEL_LIBRARY_KEYS = exports.PIPELINE_TYPES_SET = exports.SUBTASK_TYPES = exports.MODALITY_LABELS = exports.MODALITIES = exports.PIPELINE_TYPES = exports.PIPELINE_DATA = exports.MAPPING_DEFAULT_WIDGET = exports.LIBRARY_TASK_MAPPING = void 0;
|
|
18
18
|
var library_to_tasks_js_1 = require("./library-to-tasks.js");
|
|
19
19
|
Object.defineProperty(exports, "LIBRARY_TASK_MAPPING", { enumerable: true, get: function () { return library_to_tasks_js_1.LIBRARY_TASK_MAPPING; } });
|
|
20
20
|
var default_widget_inputs_js_1 = require("./default-widget-inputs.js");
|
|
@@ -51,3 +51,6 @@ Object.defineProperty(exports, "KERNEL_LIBRARIES_UI_ELEMENTS", { enumerable: tru
|
|
|
51
51
|
__exportStar(require("./inference-providers.js"), exports);
|
|
52
52
|
var eval_js_1 = require("./eval.js");
|
|
53
53
|
Object.defineProperty(exports, "EVALUATION_FRAMEWORKS", { enumerable: true, get: function () { return eval_js_1.EVALUATION_FRAMEWORKS; } });
|
|
54
|
+
var agent_harnesses_js_1 = require("./agent-harnesses.js");
|
|
55
|
+
Object.defineProperty(exports, "AGENT_HARNESSES", { enumerable: true, get: function () { return agent_harnesses_js_1.AGENT_HARNESSES; } });
|
|
56
|
+
Object.defineProperty(exports, "STANDARD_AGENT_ENV_VARS", { enumerable: true, get: function () { return agent_harnesses_js_1.STANDARD_AGENT_ENV_VARS; } });
|
|
@@ -0,0 +1,196 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Registry of AI coding agents / harnesses known to use the Hugging Face Hub.
|
|
3
|
+
*
|
|
4
|
+
* To add your harness, append an entry below keyed by its `id` (the name used
|
|
5
|
+
* when reporting Hub activity), and list the environment variable(s) that
|
|
6
|
+
* identify it.
|
|
7
|
+
*/
|
|
8
|
+
export interface AgentHarness {
|
|
9
|
+
/**
|
|
10
|
+
* Human-readable name of the harness, e.g. displayed in a leaderboard.
|
|
11
|
+
*/
|
|
12
|
+
prettyLabel: string;
|
|
13
|
+
/**
|
|
14
|
+
* URL to the harness's code repository (usually on GitHub).
|
|
15
|
+
*/
|
|
16
|
+
repoUrl?: string;
|
|
17
|
+
/**
|
|
18
|
+
* URL to the harness's documentation or website.
|
|
19
|
+
*/
|
|
20
|
+
docsUrl?: string;
|
|
21
|
+
/**
|
|
22
|
+
* Short description of the harness.
|
|
23
|
+
*/
|
|
24
|
+
description?: string;
|
|
25
|
+
/**
|
|
26
|
+
* Environment variable(s) that identify this harness, mapped to the value
|
|
27
|
+
* pattern they must match. Detection matches if ANY entry matches.
|
|
28
|
+
*
|
|
29
|
+
* The value pattern is one of:
|
|
30
|
+
* - `"*"`: the variable is set to any (non-empty) value
|
|
31
|
+
* - `"<value>"`: the variable equals this exact value
|
|
32
|
+
* - `"<prefix>*"`: the variable value starts with `<prefix>` (fuzzy match, resolved client-side)
|
|
33
|
+
*
|
|
34
|
+
* If not provided, the harness is detected through the standard AI_AGENT / AGENT variables only.
|
|
35
|
+
*/
|
|
36
|
+
envVars?: Record<string, string>;
|
|
37
|
+
}
|
|
38
|
+
/**
|
|
39
|
+
* Standard environment variables that any tool can set to identify itself.
|
|
40
|
+
* When one of these is set, its value is used directly as the agent `id`
|
|
41
|
+
* (matched against the keys of `AGENT_HARNESSES`); unrecognized values are
|
|
42
|
+
* reported as `"unknown"`.
|
|
43
|
+
*/
|
|
44
|
+
export declare const STANDARD_AGENT_ENV_VARS: readonly ["AI_AGENT", "AGENT"];
|
|
45
|
+
/**
|
|
46
|
+
* Add your new agent harness here.
|
|
47
|
+
*
|
|
48
|
+
* /!\ IMPORTANT
|
|
49
|
+
*
|
|
50
|
+
* Insertion order matters for detection priority: harnesses are checked from
|
|
51
|
+
* top to bottom and the first match wins. In particular, `cowork` must stay
|
|
52
|
+
* before `claude-code` so the more specific signal takes priority when both
|
|
53
|
+
* `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
54
|
+
*/
|
|
55
|
+
export declare const AGENT_HARNESSES: {
|
|
56
|
+
antigravity: {
|
|
57
|
+
prettyLabel: string;
|
|
58
|
+
docsUrl: string;
|
|
59
|
+
description: string;
|
|
60
|
+
envVars: {
|
|
61
|
+
ANTIGRAVITY_AGENT: string;
|
|
62
|
+
};
|
|
63
|
+
};
|
|
64
|
+
"augment-cli": {
|
|
65
|
+
prettyLabel: string;
|
|
66
|
+
repoUrl: string;
|
|
67
|
+
docsUrl: string;
|
|
68
|
+
description: string;
|
|
69
|
+
envVars: {
|
|
70
|
+
AUGMENT_AGENT: string;
|
|
71
|
+
};
|
|
72
|
+
};
|
|
73
|
+
cline: {
|
|
74
|
+
prettyLabel: string;
|
|
75
|
+
repoUrl: string;
|
|
76
|
+
docsUrl: string;
|
|
77
|
+
description: string;
|
|
78
|
+
envVars: {
|
|
79
|
+
CLINE_ACTIVE: string;
|
|
80
|
+
};
|
|
81
|
+
};
|
|
82
|
+
cowork: {
|
|
83
|
+
prettyLabel: string;
|
|
84
|
+
docsUrl: string;
|
|
85
|
+
description: string;
|
|
86
|
+
envVars: {
|
|
87
|
+
CLAUDE_CODE_IS_COWORK: string;
|
|
88
|
+
};
|
|
89
|
+
};
|
|
90
|
+
"claude-code": {
|
|
91
|
+
prettyLabel: string;
|
|
92
|
+
repoUrl: string;
|
|
93
|
+
docsUrl: string;
|
|
94
|
+
description: string;
|
|
95
|
+
envVars: {
|
|
96
|
+
CLAUDECODE: string;
|
|
97
|
+
CLAUDE_CODE: string;
|
|
98
|
+
};
|
|
99
|
+
};
|
|
100
|
+
codex: {
|
|
101
|
+
prettyLabel: string;
|
|
102
|
+
repoUrl: string;
|
|
103
|
+
docsUrl: string;
|
|
104
|
+
description: string;
|
|
105
|
+
envVars: {
|
|
106
|
+
CODEX_SANDBOX: string;
|
|
107
|
+
CODEX_CI: string;
|
|
108
|
+
CODEX_THREAD_ID: string;
|
|
109
|
+
};
|
|
110
|
+
};
|
|
111
|
+
"cursor-cli": {
|
|
112
|
+
prettyLabel: string;
|
|
113
|
+
docsUrl: string;
|
|
114
|
+
description: string;
|
|
115
|
+
envVars: {
|
|
116
|
+
CURSOR_AGENT: string;
|
|
117
|
+
};
|
|
118
|
+
};
|
|
119
|
+
cursor: {
|
|
120
|
+
prettyLabel: string;
|
|
121
|
+
docsUrl: string;
|
|
122
|
+
description: string;
|
|
123
|
+
envVars: {
|
|
124
|
+
CURSOR_TRACE_ID: string;
|
|
125
|
+
};
|
|
126
|
+
};
|
|
127
|
+
"github-copilot": {
|
|
128
|
+
prettyLabel: string;
|
|
129
|
+
docsUrl: string;
|
|
130
|
+
description: string;
|
|
131
|
+
envVars: {
|
|
132
|
+
COPILOT_MODEL: string;
|
|
133
|
+
COPILOT_ALLOW_ALL: string;
|
|
134
|
+
COPILOT_GITHUB_TOKEN: string;
|
|
135
|
+
};
|
|
136
|
+
};
|
|
137
|
+
goose: {
|
|
138
|
+
prettyLabel: string;
|
|
139
|
+
repoUrl: string;
|
|
140
|
+
docsUrl: string;
|
|
141
|
+
description: string;
|
|
142
|
+
envVars: {
|
|
143
|
+
GOOSE_TERMINAL: string;
|
|
144
|
+
};
|
|
145
|
+
};
|
|
146
|
+
openclaw: {
|
|
147
|
+
prettyLabel: string;
|
|
148
|
+
repoUrl: string;
|
|
149
|
+
docsUrl: string;
|
|
150
|
+
description: string;
|
|
151
|
+
envVars: {
|
|
152
|
+
OPENCLAW_SHELL: string;
|
|
153
|
+
};
|
|
154
|
+
};
|
|
155
|
+
opencode: {
|
|
156
|
+
prettyLabel: string;
|
|
157
|
+
repoUrl: string;
|
|
158
|
+
docsUrl: string;
|
|
159
|
+
description: string;
|
|
160
|
+
envVars: {
|
|
161
|
+
OPENCODE_CLIENT: string;
|
|
162
|
+
};
|
|
163
|
+
};
|
|
164
|
+
pi: {
|
|
165
|
+
prettyLabel: string;
|
|
166
|
+
repoUrl: string;
|
|
167
|
+
docsUrl: string;
|
|
168
|
+
description: string;
|
|
169
|
+
envVars: {
|
|
170
|
+
PI_CODING_AGENT: string;
|
|
171
|
+
};
|
|
172
|
+
};
|
|
173
|
+
replit: {
|
|
174
|
+
prettyLabel: string;
|
|
175
|
+
docsUrl: string;
|
|
176
|
+
description: string;
|
|
177
|
+
envVars: {
|
|
178
|
+
REPL_ID: string;
|
|
179
|
+
};
|
|
180
|
+
};
|
|
181
|
+
trae: {
|
|
182
|
+
prettyLabel: string;
|
|
183
|
+
docsUrl: string;
|
|
184
|
+
description: string;
|
|
185
|
+
envVars: {
|
|
186
|
+
TRAE_AI_SHELL_ID: string;
|
|
187
|
+
};
|
|
188
|
+
};
|
|
189
|
+
devin: {
|
|
190
|
+
prettyLabel: string;
|
|
191
|
+
docsUrl: string;
|
|
192
|
+
description: string;
|
|
193
|
+
};
|
|
194
|
+
};
|
|
195
|
+
export type AgentHarnessKey = keyof typeof AGENT_HARNESSES;
|
|
196
|
+
//# sourceMappingURL=agent-harnesses.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-harnesses.d.ts","sourceRoot":"","sources":["../../src/agent-harnesses.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,MAAM,WAAW,YAAY;IAC5B;;OAEG;IACH,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;OAEG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;OAEG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;;;;;;;OAUG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjC;AAED;;;;;GAKG;AACH,eAAO,MAAM,uBAAuB,gCAAiC,CAAC;AAEtE;;;;;;;;;GASG;AACH,eAAO,MAAM,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA2GY,CAAC;AAGzC,MAAM,MAAM,eAAe,GAAG,MAAM,OAAO,eAAe,CAAC"}
|
|
@@ -0,0 +1,125 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Standard environment variables that any tool can set to identify itself.
|
|
3
|
+
* When one of these is set, its value is used directly as the agent `id`
|
|
4
|
+
* (matched against the keys of `AGENT_HARNESSES`); unrecognized values are
|
|
5
|
+
* reported as `"unknown"`.
|
|
6
|
+
*/
|
|
7
|
+
export const STANDARD_AGENT_ENV_VARS = ["AI_AGENT", "AGENT"];
|
|
8
|
+
/**
|
|
9
|
+
* Add your new agent harness here.
|
|
10
|
+
*
|
|
11
|
+
* /!\ IMPORTANT
|
|
12
|
+
*
|
|
13
|
+
* Insertion order matters for detection priority: harnesses are checked from
|
|
14
|
+
* top to bottom and the first match wins. In particular, `cowork` must stay
|
|
15
|
+
* before `claude-code` so the more specific signal takes priority when both
|
|
16
|
+
* `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
17
|
+
*/
|
|
18
|
+
export const AGENT_HARNESSES = {
|
|
19
|
+
antigravity: {
|
|
20
|
+
prettyLabel: "Antigravity",
|
|
21
|
+
docsUrl: "https://antigravity.google",
|
|
22
|
+
description: "Agentic development platform from Google built around Gemini.",
|
|
23
|
+
envVars: { ANTIGRAVITY_AGENT: "*" },
|
|
24
|
+
},
|
|
25
|
+
"augment-cli": {
|
|
26
|
+
prettyLabel: "Augment CLI",
|
|
27
|
+
repoUrl: "https://github.com/augmentcode/auggie",
|
|
28
|
+
docsUrl: "https://www.augmentcode.com",
|
|
29
|
+
description: "Auggie, the command-line coding agent from Augment Code.",
|
|
30
|
+
envVars: { AUGMENT_AGENT: "*" },
|
|
31
|
+
},
|
|
32
|
+
cline: {
|
|
33
|
+
prettyLabel: "Cline",
|
|
34
|
+
repoUrl: "https://github.com/cline/cline",
|
|
35
|
+
docsUrl: "https://cline.bot",
|
|
36
|
+
description: "Open-source autonomous coding agent for VS Code.",
|
|
37
|
+
envVars: { CLINE_ACTIVE: "*" },
|
|
38
|
+
},
|
|
39
|
+
cowork: {
|
|
40
|
+
// must stay before `claude-code` so the more specific signal takes priority when both `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
41
|
+
prettyLabel: "Cowork",
|
|
42
|
+
docsUrl: "https://claude.com/product/cowork",
|
|
43
|
+
description: "Anthropic's agent for autonomous knowledge work, built on top of Claude Code.",
|
|
44
|
+
envVars: { CLAUDE_CODE_IS_COWORK: "*" },
|
|
45
|
+
},
|
|
46
|
+
"claude-code": {
|
|
47
|
+
prettyLabel: "Claude Code",
|
|
48
|
+
repoUrl: "https://github.com/anthropics/claude-code",
|
|
49
|
+
docsUrl: "https://code.claude.com/docs",
|
|
50
|
+
description: "Anthropic's agentic coding tool that lives in your terminal.",
|
|
51
|
+
envVars: { CLAUDECODE: "*", CLAUDE_CODE: "*" },
|
|
52
|
+
},
|
|
53
|
+
codex: {
|
|
54
|
+
prettyLabel: "Codex",
|
|
55
|
+
repoUrl: "https://github.com/openai/codex",
|
|
56
|
+
docsUrl: "https://developers.openai.com/codex",
|
|
57
|
+
description: "OpenAI's lightweight coding agent that runs in your terminal.",
|
|
58
|
+
envVars: { CODEX_SANDBOX: "*", CODEX_CI: "*", CODEX_THREAD_ID: "*" },
|
|
59
|
+
},
|
|
60
|
+
"cursor-cli": {
|
|
61
|
+
// must stay before `cursor` so the more specific signal takes priority: when the CLI runs inside
|
|
62
|
+
// the Cursor editor's terminal, child processes inherit `CURSOR_TRACE_ID` and the CLI sets `CURSOR_AGENT`.
|
|
63
|
+
prettyLabel: "Cursor CLI",
|
|
64
|
+
docsUrl: "https://cursor.com/docs/cli/overview",
|
|
65
|
+
description: "Cursor's coding agent for the command line.",
|
|
66
|
+
envVars: { CURSOR_AGENT: "*" },
|
|
67
|
+
},
|
|
68
|
+
cursor: {
|
|
69
|
+
prettyLabel: "Cursor",
|
|
70
|
+
docsUrl: "https://cursor.com",
|
|
71
|
+
description: "AI-powered code editor.",
|
|
72
|
+
envVars: { CURSOR_TRACE_ID: "*" },
|
|
73
|
+
},
|
|
74
|
+
"github-copilot": {
|
|
75
|
+
prettyLabel: "GitHub Copilot",
|
|
76
|
+
docsUrl: "https://docs.github.com/copilot",
|
|
77
|
+
description: "GitHub's AI coding assistant.",
|
|
78
|
+
envVars: { COPILOT_MODEL: "*", COPILOT_ALLOW_ALL: "*", COPILOT_GITHUB_TOKEN: "*" },
|
|
79
|
+
},
|
|
80
|
+
goose: {
|
|
81
|
+
prettyLabel: "Goose",
|
|
82
|
+
repoUrl: "https://github.com/aaif-goose/goose",
|
|
83
|
+
docsUrl: "https://goose-docs.ai/",
|
|
84
|
+
description: "Open-source, extensible AI agent, originally from Block and now part of the Agentic AI Foundation.",
|
|
85
|
+
envVars: { GOOSE_TERMINAL: "*" },
|
|
86
|
+
},
|
|
87
|
+
openclaw: {
|
|
88
|
+
prettyLabel: "OpenClaw",
|
|
89
|
+
repoUrl: "https://github.com/openclaw/openclaw",
|
|
90
|
+
docsUrl: "https://openclaw.ai",
|
|
91
|
+
description: "Open-source, self-hosted personal AI assistant that runs on your own devices.",
|
|
92
|
+
envVars: { OPENCLAW_SHELL: "*" },
|
|
93
|
+
},
|
|
94
|
+
opencode: {
|
|
95
|
+
prettyLabel: "opencode",
|
|
96
|
+
repoUrl: "https://github.com/anomalyco/opencode",
|
|
97
|
+
docsUrl: "https://opencode.ai",
|
|
98
|
+
description: "Open-source AI coding agent built for the terminal.",
|
|
99
|
+
envVars: { OPENCODE_CLIENT: "*" },
|
|
100
|
+
},
|
|
101
|
+
pi: {
|
|
102
|
+
prettyLabel: "Pi",
|
|
103
|
+
repoUrl: "https://github.com/earendil-works/pi",
|
|
104
|
+
docsUrl: "https://pi.dev",
|
|
105
|
+
description: "Minimal, self-extensible terminal coding agent with a unified multi-provider LLM API.",
|
|
106
|
+
envVars: { PI_CODING_AGENT: "*" },
|
|
107
|
+
},
|
|
108
|
+
replit: {
|
|
109
|
+
prettyLabel: "Replit",
|
|
110
|
+
docsUrl: "https://replit.com",
|
|
111
|
+
description: "Cloud development environment with an AI coding agent.",
|
|
112
|
+
envVars: { REPL_ID: "*" },
|
|
113
|
+
},
|
|
114
|
+
trae: {
|
|
115
|
+
prettyLabel: "Trae",
|
|
116
|
+
docsUrl: "https://trae.ai",
|
|
117
|
+
description: "AI-powered IDE from ByteDance.",
|
|
118
|
+
envVars: { TRAE_AI_SHELL_ID: "*" },
|
|
119
|
+
},
|
|
120
|
+
devin: {
|
|
121
|
+
prettyLabel: "Devin",
|
|
122
|
+
docsUrl: "https://devin.ai",
|
|
123
|
+
description: "Autonomous AI software engineer from Cognition.",
|
|
124
|
+
},
|
|
125
|
+
};
|
package/dist/esm/eval.d.ts
CHANGED
|
@@ -32,6 +32,11 @@ export declare const EVALUATION_FRAMEWORKS: {
|
|
|
32
32
|
readonly description: "Harbor is a framework for evaluating and optimizing agents and language models.";
|
|
33
33
|
readonly url: "https://github.com/laude-institute/harbor";
|
|
34
34
|
};
|
|
35
|
+
readonly pier: {
|
|
36
|
+
readonly name: "pier";
|
|
37
|
+
readonly description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.";
|
|
38
|
+
readonly url: "https://github.com/datacurve-ai/pier";
|
|
39
|
+
};
|
|
35
40
|
readonly archipelago: {
|
|
36
41
|
readonly name: "archipelago";
|
|
37
42
|
readonly description: "Archipelago is a system for running and evaluating AI agents against MCP applications.";
|
|
@@ -112,5 +117,10 @@ export declare const EVALUATION_FRAMEWORKS: {
|
|
|
112
117
|
readonly description: "WBench is a comprehensive multi-turn benchmark for interactive video world model evaluation, assessing models across 5 dimensions (video quality, setting adherence, interaction adherence, consistency, physics compliance) and 22 metrics over 289 multi-turn interaction cases.";
|
|
113
118
|
readonly url: "https://github.com/meituan-longcat/WBench";
|
|
114
119
|
};
|
|
120
|
+
readonly nanofold: {
|
|
121
|
+
readonly name: "nanofold";
|
|
122
|
+
readonly description: "nanoFold is a data-efficiency benchmark for protein structure prediction. Its goal is to evaluate models on scenarios with scarce data.";
|
|
123
|
+
readonly url: "https://github.com/ChrisHayduk/nanoFold-Competition";
|
|
124
|
+
};
|
|
115
125
|
};
|
|
116
126
|
//# sourceMappingURL=eval.d.ts.map
|
package/dist/esm/eval.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB
|
|
1
|
+
{"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAwIxB,CAAC"}
|
package/dist/esm/eval.js
CHANGED
|
@@ -32,6 +32,11 @@ export const EVALUATION_FRAMEWORKS = {
|
|
|
32
32
|
description: "Harbor is a framework for evaluating and optimizing agents and language models.",
|
|
33
33
|
url: "https://github.com/laude-institute/harbor",
|
|
34
34
|
},
|
|
35
|
+
pier: {
|
|
36
|
+
name: "pier",
|
|
37
|
+
description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.",
|
|
38
|
+
url: "https://github.com/datacurve-ai/pier",
|
|
39
|
+
},
|
|
35
40
|
archipelago: {
|
|
36
41
|
name: "archipelago",
|
|
37
42
|
description: "Archipelago is a system for running and evaluating AI agents against MCP applications.",
|
|
@@ -112,4 +117,9 @@ export const EVALUATION_FRAMEWORKS = {
|
|
|
112
117
|
description: "WBench is a comprehensive multi-turn benchmark for interactive video world model evaluation, assessing models across 5 dimensions (video quality, setting adherence, interaction adherence, consistency, physics compliance) and 22 metrics over 289 multi-turn interaction cases.",
|
|
113
118
|
url: "https://github.com/meituan-longcat/WBench",
|
|
114
119
|
},
|
|
120
|
+
nanofold: {
|
|
121
|
+
name: "nanofold",
|
|
122
|
+
description: "nanoFold is a data-efficiency benchmark for protein structure prediction. Its goal is to evaluate models on scenarios with scarce data.",
|
|
123
|
+
url: "https://github.com/ChrisHayduk/nanoFold-Competition",
|
|
124
|
+
},
|
|
115
125
|
};
|
package/dist/esm/index.d.ts
CHANGED
|
@@ -23,4 +23,6 @@ export { KERNEL_LIBRARIES_UI_ELEMENTS } from "./kernel-libraries.js";
|
|
|
23
23
|
export type { KernelLibraryKey, KernelLibraryUiElement } from "./kernel-libraries.js";
|
|
24
24
|
export * from "./inference-providers.js";
|
|
25
25
|
export { EVALUATION_FRAMEWORKS } from "./eval.js";
|
|
26
|
+
export { AGENT_HARNESSES, STANDARD_AGENT_ENV_VARS } from "./agent-harnesses.js";
|
|
27
|
+
export type { AgentHarness, AgentHarnessKey } from "./agent-harnesses.js";
|
|
26
28
|
//# sourceMappingURL=index.d.ts.map
|
package/dist/esm/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAC7D,OAAO,EAAE,sBAAsB,EAAE,MAAM,4BAA4B,CAAC;AACpE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AACvF,cAAc,kBAAkB,CAAC;AACjC,OAAO,EACN,aAAa,EACb,cAAc,EACd,KAAK,UAAU,EACf,KAAK,YAAY,EACjB,KAAK,YAAY,EACjB,KAAK,QAAQ,EACb,UAAU,EACV,eAAe,EACf,aAAa,EACb,kBAAkB,GAClB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACN,8BAA8B,EAC9B,sBAAsB,EACtB,2BAA2B,GAC3B,MAAM,sBAAsB,CAAC;AAC9B,YAAY,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AAC9E,YAAY,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,YAAY,EAAE,UAAU,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACzF,YAAY,EACX,aAAa,EACb,sBAAsB,EACtB,gCAAgC,EAChC,8BAA8B,EAC9B,kCAAkC,EAClC,uBAAuB,EACvB,sBAAsB,EACtB,oCAAoC,EACpC,gCAAgC,EAChC,2BAA2B,EAC3B,gCAAgC,EAChC,8BAA8B,EAC9B,sBAAsB,EACtB,8BAA8B,EAC9B,mBAAmB,EACnB,sBAAsB,EACtB,yBAAyB,EACzB,8BAA8B,EAC9B,uBAAuB,GACvB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAEhE,cAAc,WAAW,CAAC;AAE1B,OAAO,EACN,KAAK,gBAAgB,EACrB,KAAK,wBAAwB,EAC7B,KAAK,gBAAgB,EACrB,yBAAyB,EACzB,yBAAyB,EACzB,iBAAiB,EACjB,oBAAoB,GACpB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,IAAI,EAAE,sBAAsB,EAAE,MAAM,eAAe,CAAC;AAC7D,YAAY,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,eAAe,CAAC;AAC3D,YAAY,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAC;AAC5D,YAAY,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAC7C,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAE9E,OAAO,EAAE,6BAA6B,EAAE,MAAM,wBAAwB,CAAC;AACvE,YAAY,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAEzF,OAAO,EAAE,4BAA4B,EAAE,MAAM,uBAAuB,CAAC;AACrE,YAAY,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAEtF,cAAc,0BAA0B,CAAC;AAEzC,OAAO,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAC7D,OAAO,EAAE,sBAAsB,EAAE,MAAM,4BAA4B,CAAC;AACpE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AACvF,cAAc,kBAAkB,CAAC;AACjC,OAAO,EACN,aAAa,EACb,cAAc,EACd,KAAK,UAAU,EACf,KAAK,YAAY,EACjB,KAAK,YAAY,EACjB,KAAK,QAAQ,EACb,UAAU,EACV,eAAe,EACf,aAAa,EACb,kBAAkB,GAClB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACN,8BAA8B,EAC9B,sBAAsB,EACtB,2BAA2B,GAC3B,MAAM,sBAAsB,CAAC;AAC9B,YAAY,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AAC9E,YAAY,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,YAAY,EAAE,UAAU,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACzF,YAAY,EACX,aAAa,EACb,sBAAsB,EACtB,gCAAgC,EAChC,8BAA8B,EAC9B,kCAAkC,EAClC,uBAAuB,EACvB,sBAAsB,EACtB,oCAAoC,EACpC,gCAAgC,EAChC,2BAA2B,EAC3B,gCAAgC,EAChC,8BAA8B,EAC9B,sBAAsB,EACtB,8BAA8B,EAC9B,mBAAmB,EACnB,sBAAsB,EACtB,yBAAyB,EACzB,8BAA8B,EAC9B,uBAAuB,GACvB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAEhE,cAAc,WAAW,CAAC;AAE1B,OAAO,EACN,KAAK,gBAAgB,EACrB,KAAK,wBAAwB,EAC7B,KAAK,gBAAgB,EACrB,yBAAyB,EACzB,yBAAyB,EACzB,iBAAiB,EACjB,oBAAoB,GACpB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,IAAI,EAAE,sBAAsB,EAAE,MAAM,eAAe,CAAC;AAC7D,YAAY,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,eAAe,CAAC;AAC3D,YAAY,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAC;AAC5D,YAAY,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAC7C,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAE9E,OAAO,EAAE,6BAA6B,EAAE,MAAM,wBAAwB,CAAC;AACvE,YAAY,EAAE,uBAAuB,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAEzF,OAAO,EAAE,4BAA4B,EAAE,MAAM,uBAAuB,CAAC;AACrE,YAAY,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAEtF,cAAc,0BAA0B,CAAC;AAEzC,OAAO,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC;AAElD,OAAO,EAAE,eAAe,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AAChF,YAAY,EAAE,YAAY,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC"}
|
package/dist/esm/index.js
CHANGED
|
@@ -12,3 +12,4 @@ export { DATASET_LIBRARIES_UI_ELEMENTS } from "./dataset-libraries.js";
|
|
|
12
12
|
export { KERNEL_LIBRARIES_UI_ELEMENTS } from "./kernel-libraries.js";
|
|
13
13
|
export * from "./inference-providers.js";
|
|
14
14
|
export { EVALUATION_FRAMEWORKS } from "./eval.js";
|
|
15
|
+
export { AGENT_HARNESSES, STANDARD_AGENT_ENV_VARS } from "./agent-harnesses.js";
|
package/package.json
CHANGED
|
@@ -0,0 +1,167 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Registry of AI coding agents / harnesses known to use the Hugging Face Hub.
|
|
3
|
+
*
|
|
4
|
+
* To add your harness, append an entry below keyed by its `id` (the name used
|
|
5
|
+
* when reporting Hub activity), and list the environment variable(s) that
|
|
6
|
+
* identify it.
|
|
7
|
+
*/
|
|
8
|
+
export interface AgentHarness {
|
|
9
|
+
/**
|
|
10
|
+
* Human-readable name of the harness, e.g. displayed in a leaderboard.
|
|
11
|
+
*/
|
|
12
|
+
prettyLabel: string;
|
|
13
|
+
/**
|
|
14
|
+
* URL to the harness's code repository (usually on GitHub).
|
|
15
|
+
*/
|
|
16
|
+
repoUrl?: string;
|
|
17
|
+
/**
|
|
18
|
+
* URL to the harness's documentation or website.
|
|
19
|
+
*/
|
|
20
|
+
docsUrl?: string;
|
|
21
|
+
/**
|
|
22
|
+
* Short description of the harness.
|
|
23
|
+
*/
|
|
24
|
+
description?: string;
|
|
25
|
+
/**
|
|
26
|
+
* Environment variable(s) that identify this harness, mapped to the value
|
|
27
|
+
* pattern they must match. Detection matches if ANY entry matches.
|
|
28
|
+
*
|
|
29
|
+
* The value pattern is one of:
|
|
30
|
+
* - `"*"`: the variable is set to any (non-empty) value
|
|
31
|
+
* - `"<value>"`: the variable equals this exact value
|
|
32
|
+
* - `"<prefix>*"`: the variable value starts with `<prefix>` (fuzzy match, resolved client-side)
|
|
33
|
+
*
|
|
34
|
+
* If not provided, the harness is detected through the standard AI_AGENT / AGENT variables only.
|
|
35
|
+
*/
|
|
36
|
+
envVars?: Record<string, string>;
|
|
37
|
+
}
|
|
38
|
+
|
|
39
|
+
/**
|
|
40
|
+
* Standard environment variables that any tool can set to identify itself.
|
|
41
|
+
* When one of these is set, its value is used directly as the agent `id`
|
|
42
|
+
* (matched against the keys of `AGENT_HARNESSES`); unrecognized values are
|
|
43
|
+
* reported as `"unknown"`.
|
|
44
|
+
*/
|
|
45
|
+
export const STANDARD_AGENT_ENV_VARS = ["AI_AGENT", "AGENT"] as const;
|
|
46
|
+
|
|
47
|
+
/**
|
|
48
|
+
* Add your new agent harness here.
|
|
49
|
+
*
|
|
50
|
+
* /!\ IMPORTANT
|
|
51
|
+
*
|
|
52
|
+
* Insertion order matters for detection priority: harnesses are checked from
|
|
53
|
+
* top to bottom and the first match wins. In particular, `cowork` must stay
|
|
54
|
+
* before `claude-code` so the more specific signal takes priority when both
|
|
55
|
+
* `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
56
|
+
*/
|
|
57
|
+
export const AGENT_HARNESSES = {
|
|
58
|
+
antigravity: {
|
|
59
|
+
prettyLabel: "Antigravity",
|
|
60
|
+
docsUrl: "https://antigravity.google",
|
|
61
|
+
description: "Agentic development platform from Google built around Gemini.",
|
|
62
|
+
envVars: { ANTIGRAVITY_AGENT: "*" },
|
|
63
|
+
},
|
|
64
|
+
"augment-cli": {
|
|
65
|
+
prettyLabel: "Augment CLI",
|
|
66
|
+
repoUrl: "https://github.com/augmentcode/auggie",
|
|
67
|
+
docsUrl: "https://www.augmentcode.com",
|
|
68
|
+
description: "Auggie, the command-line coding agent from Augment Code.",
|
|
69
|
+
envVars: { AUGMENT_AGENT: "*" },
|
|
70
|
+
},
|
|
71
|
+
cline: {
|
|
72
|
+
prettyLabel: "Cline",
|
|
73
|
+
repoUrl: "https://github.com/cline/cline",
|
|
74
|
+
docsUrl: "https://cline.bot",
|
|
75
|
+
description: "Open-source autonomous coding agent for VS Code.",
|
|
76
|
+
envVars: { CLINE_ACTIVE: "*" },
|
|
77
|
+
},
|
|
78
|
+
cowork: {
|
|
79
|
+
// must stay before `claude-code` so the more specific signal takes priority when both `CLAUDE_CODE` and `CLAUDE_CODE_IS_COWORK` are set.
|
|
80
|
+
prettyLabel: "Cowork",
|
|
81
|
+
docsUrl: "https://claude.com/product/cowork",
|
|
82
|
+
description: "Anthropic's agent for autonomous knowledge work, built on top of Claude Code.",
|
|
83
|
+
envVars: { CLAUDE_CODE_IS_COWORK: "*" },
|
|
84
|
+
},
|
|
85
|
+
"claude-code": {
|
|
86
|
+
prettyLabel: "Claude Code",
|
|
87
|
+
repoUrl: "https://github.com/anthropics/claude-code",
|
|
88
|
+
docsUrl: "https://code.claude.com/docs",
|
|
89
|
+
description: "Anthropic's agentic coding tool that lives in your terminal.",
|
|
90
|
+
envVars: { CLAUDECODE: "*", CLAUDE_CODE: "*" },
|
|
91
|
+
},
|
|
92
|
+
codex: {
|
|
93
|
+
prettyLabel: "Codex",
|
|
94
|
+
repoUrl: "https://github.com/openai/codex",
|
|
95
|
+
docsUrl: "https://developers.openai.com/codex",
|
|
96
|
+
description: "OpenAI's lightweight coding agent that runs in your terminal.",
|
|
97
|
+
envVars: { CODEX_SANDBOX: "*", CODEX_CI: "*", CODEX_THREAD_ID: "*" },
|
|
98
|
+
},
|
|
99
|
+
"cursor-cli": {
|
|
100
|
+
// must stay before `cursor` so the more specific signal takes priority: when the CLI runs inside
|
|
101
|
+
// the Cursor editor's terminal, child processes inherit `CURSOR_TRACE_ID` and the CLI sets `CURSOR_AGENT`.
|
|
102
|
+
prettyLabel: "Cursor CLI",
|
|
103
|
+
docsUrl: "https://cursor.com/docs/cli/overview",
|
|
104
|
+
description: "Cursor's coding agent for the command line.",
|
|
105
|
+
envVars: { CURSOR_AGENT: "*" },
|
|
106
|
+
},
|
|
107
|
+
cursor: {
|
|
108
|
+
prettyLabel: "Cursor",
|
|
109
|
+
docsUrl: "https://cursor.com",
|
|
110
|
+
description: "AI-powered code editor.",
|
|
111
|
+
envVars: { CURSOR_TRACE_ID: "*" },
|
|
112
|
+
},
|
|
113
|
+
"github-copilot": {
|
|
114
|
+
prettyLabel: "GitHub Copilot",
|
|
115
|
+
docsUrl: "https://docs.github.com/copilot",
|
|
116
|
+
description: "GitHub's AI coding assistant.",
|
|
117
|
+
envVars: { COPILOT_MODEL: "*", COPILOT_ALLOW_ALL: "*", COPILOT_GITHUB_TOKEN: "*" },
|
|
118
|
+
},
|
|
119
|
+
goose: {
|
|
120
|
+
prettyLabel: "Goose",
|
|
121
|
+
repoUrl: "https://github.com/aaif-goose/goose",
|
|
122
|
+
docsUrl: "https://goose-docs.ai/",
|
|
123
|
+
description: "Open-source, extensible AI agent, originally from Block and now part of the Agentic AI Foundation.",
|
|
124
|
+
envVars: { GOOSE_TERMINAL: "*" },
|
|
125
|
+
},
|
|
126
|
+
openclaw: {
|
|
127
|
+
prettyLabel: "OpenClaw",
|
|
128
|
+
repoUrl: "https://github.com/openclaw/openclaw",
|
|
129
|
+
docsUrl: "https://openclaw.ai",
|
|
130
|
+
description: "Open-source, self-hosted personal AI assistant that runs on your own devices.",
|
|
131
|
+
envVars: { OPENCLAW_SHELL: "*" },
|
|
132
|
+
},
|
|
133
|
+
opencode: {
|
|
134
|
+
prettyLabel: "opencode",
|
|
135
|
+
repoUrl: "https://github.com/anomalyco/opencode",
|
|
136
|
+
docsUrl: "https://opencode.ai",
|
|
137
|
+
description: "Open-source AI coding agent built for the terminal.",
|
|
138
|
+
envVars: { OPENCODE_CLIENT: "*" },
|
|
139
|
+
},
|
|
140
|
+
pi: {
|
|
141
|
+
prettyLabel: "Pi",
|
|
142
|
+
repoUrl: "https://github.com/earendil-works/pi",
|
|
143
|
+
docsUrl: "https://pi.dev",
|
|
144
|
+
description: "Minimal, self-extensible terminal coding agent with a unified multi-provider LLM API.",
|
|
145
|
+
envVars: { PI_CODING_AGENT: "*" },
|
|
146
|
+
},
|
|
147
|
+
replit: {
|
|
148
|
+
prettyLabel: "Replit",
|
|
149
|
+
docsUrl: "https://replit.com",
|
|
150
|
+
description: "Cloud development environment with an AI coding agent.",
|
|
151
|
+
envVars: { REPL_ID: "*" },
|
|
152
|
+
},
|
|
153
|
+
trae: {
|
|
154
|
+
prettyLabel: "Trae",
|
|
155
|
+
docsUrl: "https://trae.ai",
|
|
156
|
+
description: "AI-powered IDE from ByteDance.",
|
|
157
|
+
envVars: { TRAE_AI_SHELL_ID: "*" },
|
|
158
|
+
},
|
|
159
|
+
devin: {
|
|
160
|
+
prettyLabel: "Devin",
|
|
161
|
+
docsUrl: "https://devin.ai",
|
|
162
|
+
description: "Autonomous AI software engineer from Cognition.",
|
|
163
|
+
},
|
|
164
|
+
} satisfies Record<string, AgentHarness>;
|
|
165
|
+
|
|
166
|
+
/// List of the agent harnesses known to the Hub
|
|
167
|
+
export type AgentHarnessKey = keyof typeof AGENT_HARNESSES;
|
package/src/eval.ts
CHANGED
|
@@ -33,6 +33,12 @@ export const EVALUATION_FRAMEWORKS = {
|
|
|
33
33
|
description: "Harbor is a framework for evaluating and optimizing agents and language models.",
|
|
34
34
|
url: "https://github.com/laude-institute/harbor",
|
|
35
35
|
},
|
|
36
|
+
pier: {
|
|
37
|
+
name: "pier",
|
|
38
|
+
description:
|
|
39
|
+
"Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.",
|
|
40
|
+
url: "https://github.com/datacurve-ai/pier",
|
|
41
|
+
},
|
|
36
42
|
archipelago: {
|
|
37
43
|
name: "archipelago",
|
|
38
44
|
description: "Archipelago is a system for running and evaluating AI agents against MCP applications.",
|
|
@@ -125,4 +131,10 @@ export const EVALUATION_FRAMEWORKS = {
|
|
|
125
131
|
"WBench is a comprehensive multi-turn benchmark for interactive video world model evaluation, assessing models across 5 dimensions (video quality, setting adherence, interaction adherence, consistency, physics compliance) and 22 metrics over 289 multi-turn interaction cases.",
|
|
126
132
|
url: "https://github.com/meituan-longcat/WBench",
|
|
127
133
|
},
|
|
134
|
+
nanofold: {
|
|
135
|
+
name: "nanofold",
|
|
136
|
+
description:
|
|
137
|
+
"nanoFold is a data-efficiency benchmark for protein structure prediction. Its goal is to evaluate models on scenarios with scarce data.",
|
|
138
|
+
url: "https://github.com/ChrisHayduk/nanoFold-Competition",
|
|
139
|
+
},
|
|
128
140
|
} as const;
|
package/src/index.ts
CHANGED
|
@@ -73,3 +73,6 @@ export type { KernelLibraryKey, KernelLibraryUiElement } from "./kernel-librarie
|
|
|
73
73
|
export * from "./inference-providers.js";
|
|
74
74
|
|
|
75
75
|
export { EVALUATION_FRAMEWORKS } from "./eval.js";
|
|
76
|
+
|
|
77
|
+
export { AGENT_HARNESSES, STANDARD_AGENT_ENV_VARS } from "./agent-harnesses.js";
|
|
78
|
+
export type { AgentHarness, AgentHarnessKey } from "./agent-harnesses.js";
|