agentv 3.9.0 → 3.9.2
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/{chunk-Q2YWV4QM.js → chunk-6ZAFWUBT.js} +29 -18
- package/dist/chunk-6ZAFWUBT.js.map +1 -0
- package/dist/{chunk-GC5P5HHZ.js → chunk-JGMJL2LV.js} +76 -42
- package/dist/chunk-JGMJL2LV.js.map +1 -0
- package/dist/{chunk-TXDPYXHY.js → chunk-OIVGGWJ3.js} +102 -43
- package/dist/chunk-OIVGGWJ3.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-PIOSPBKX.js → dist-PUPHGVKL.js} +4 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-3VTDK5NX.js → interactive-BD56NB23.js} +3 -3
- package/dist/templates/.agentv/config.yaml +4 -13
- package/dist/templates/.agentv/targets.yaml +0 -16
- package/dist/templates/{.agentv/.env.example → .env.example} +11 -9
- package/package.json +1 -1
- package/dist/chunk-GC5P5HHZ.js.map +0 -1
- package/dist/chunk-Q2YWV4QM.js.map +0 -1
- package/dist/chunk-TXDPYXHY.js.map +0 -1
- /package/dist/{dist-PIOSPBKX.js.map → dist-PUPHGVKL.js.map} +0 -0
- /package/dist/{interactive-3VTDK5NX.js.map → interactive-BD56NB23.js.map} +0 -0
package/dist/cli.js
CHANGED
|
@@ -2,9 +2,9 @@
|
|
|
2
2
|
import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
|
|
3
3
|
import {
|
|
4
4
|
runCli
|
|
5
|
-
} from "./chunk-
|
|
6
|
-
import "./chunk-
|
|
7
|
-
import "./chunk-
|
|
5
|
+
} from "./chunk-6ZAFWUBT.js";
|
|
6
|
+
import "./chunk-JGMJL2LV.js";
|
|
7
|
+
import "./chunk-OIVGGWJ3.js";
|
|
8
8
|
import "./chunk-C5GOHBQM.js";
|
|
9
9
|
import "./chunk-JK6V4KVD.js";
|
|
10
10
|
import "./chunk-HQDCIXVH.js";
|
|
@@ -69,6 +69,7 @@ import {
|
|
|
69
69
|
extractTargetsFromSuite,
|
|
70
70
|
extractTargetsFromTestCase,
|
|
71
71
|
extractTrialsConfig,
|
|
72
|
+
extractWorkersFromSuite,
|
|
72
73
|
fileExists,
|
|
73
74
|
findGitRoot,
|
|
74
75
|
freeformEvaluationSchema,
|
|
@@ -140,7 +141,7 @@ import {
|
|
|
140
141
|
transpileEvalYaml,
|
|
141
142
|
transpileEvalYamlFile,
|
|
142
143
|
trimBaselineResult
|
|
143
|
-
} from "./chunk-
|
|
144
|
+
} from "./chunk-OIVGGWJ3.js";
|
|
144
145
|
import {
|
|
145
146
|
OtlpJsonFileExporter
|
|
146
147
|
} from "./chunk-C5GOHBQM.js";
|
|
@@ -226,6 +227,7 @@ export {
|
|
|
226
227
|
extractTargetsFromSuite,
|
|
227
228
|
extractTargetsFromTestCase,
|
|
228
229
|
extractTrialsConfig,
|
|
230
|
+
extractWorkersFromSuite,
|
|
229
231
|
fileExists,
|
|
230
232
|
findGitRoot,
|
|
231
233
|
freeformEvaluationSchema,
|
|
@@ -298,4 +300,4 @@ export {
|
|
|
298
300
|
transpileEvalYamlFile,
|
|
299
301
|
trimBaselineResult
|
|
300
302
|
};
|
|
301
|
-
//# sourceMappingURL=dist-
|
|
303
|
+
//# sourceMappingURL=dist-PUPHGVKL.js.map
|
package/dist/index.js
CHANGED
|
@@ -3,9 +3,9 @@ import {
|
|
|
3
3
|
app,
|
|
4
4
|
preprocessArgv,
|
|
5
5
|
runCli
|
|
6
|
-
} from "./chunk-
|
|
7
|
-
import "./chunk-
|
|
8
|
-
import "./chunk-
|
|
6
|
+
} from "./chunk-6ZAFWUBT.js";
|
|
7
|
+
import "./chunk-JGMJL2LV.js";
|
|
8
|
+
import "./chunk-OIVGGWJ3.js";
|
|
9
9
|
import "./chunk-C5GOHBQM.js";
|
|
10
10
|
import "./chunk-JK6V4KVD.js";
|
|
11
11
|
import "./chunk-HQDCIXVH.js";
|
|
@@ -4,14 +4,14 @@ import {
|
|
|
4
4
|
fileExists,
|
|
5
5
|
findRepoRoot,
|
|
6
6
|
runEvalCommand
|
|
7
|
-
} from "./chunk-
|
|
7
|
+
} from "./chunk-JGMJL2LV.js";
|
|
8
8
|
import {
|
|
9
9
|
DEFAULT_EVAL_PATTERNS,
|
|
10
10
|
getAgentvHome,
|
|
11
11
|
listTargetNames,
|
|
12
12
|
loadConfig,
|
|
13
13
|
readTargetDefinitions
|
|
14
|
-
} from "./chunk-
|
|
14
|
+
} from "./chunk-OIVGGWJ3.js";
|
|
15
15
|
import "./chunk-C5GOHBQM.js";
|
|
16
16
|
import "./chunk-JK6V4KVD.js";
|
|
17
17
|
import "./chunk-HQDCIXVH.js";
|
|
@@ -371,4 +371,4 @@ ${ANSI_DIM}Retrying execution errors...${ANSI_RESET}
|
|
|
371
371
|
export {
|
|
372
372
|
launchInteractiveWizard
|
|
373
373
|
};
|
|
374
|
-
//# sourceMappingURL=interactive-
|
|
374
|
+
//# sourceMappingURL=interactive-BD56NB23.js.map
|
|
@@ -1,15 +1,6 @@
|
|
|
1
1
|
$schema: agentv-config-v2
|
|
2
2
|
|
|
3
|
-
#
|
|
4
|
-
|
|
5
|
-
#
|
|
6
|
-
|
|
7
|
-
- "**/*.instructions.md"
|
|
8
|
-
- "**/*.prompt.md"
|
|
9
|
-
- "**/SKILL.md"
|
|
10
|
-
|
|
11
|
-
# Notes:
|
|
12
|
-
# - Patterns use standard glob syntax (via micromatch library)
|
|
13
|
-
# - Paths are normalized to forward slashes for cross-platform compatibility
|
|
14
|
-
# - Only files matching these patterns are loaded as guidelines
|
|
15
|
-
# - All other files referenced in tests are treated as regular file content
|
|
3
|
+
# Execution defaults (overridden by CLI flags)
|
|
4
|
+
# execution:
|
|
5
|
+
# pool_workspaces: true # Reuse materialized workspaces across eval runs
|
|
6
|
+
# pool_slots: 10 # Max pool slots on disk (1-50, default: 10)
|
|
@@ -10,10 +10,6 @@ targets:
|
|
|
10
10
|
model: ${{ AZURE_DEPLOYMENT_NAME }}
|
|
11
11
|
# version: ${{ AZURE_OPENAI_API_VERSION }} # Optional: uncomment to override default (2024-12-01-preview)
|
|
12
12
|
|
|
13
|
-
- name: vscode
|
|
14
|
-
provider: vscode
|
|
15
|
-
judge_target: azure-llm
|
|
16
|
-
|
|
17
13
|
- name: codex
|
|
18
14
|
provider: codex
|
|
19
15
|
judge_target: azure-llm
|
|
@@ -43,18 +39,6 @@ targets:
|
|
|
43
39
|
log_format: json # Optional: 'summary' (default) or 'json' for raw event logs
|
|
44
40
|
# system_prompt: optional override (default instructs agent to include code in response)
|
|
45
41
|
|
|
46
|
-
- name: vscode_projectx
|
|
47
|
-
provider: vscode
|
|
48
|
-
workspace_template: ${{ PROJECTX_WORKSPACE_PATH }}
|
|
49
|
-
provider_batching: false
|
|
50
|
-
judge_target: azure-llm
|
|
51
|
-
|
|
52
|
-
- name: vscode_insiders_projectx
|
|
53
|
-
provider: vscode-insiders
|
|
54
|
-
workspace_template: ${{ PROJECTX_WORKSPACE_PATH }}
|
|
55
|
-
provider_batching: false
|
|
56
|
-
judge_target: azure-llm
|
|
57
|
-
|
|
58
42
|
- name: azure-llm
|
|
59
43
|
provider: azure
|
|
60
44
|
endpoint: ${{ AZURE_OPENAI_ENDPOINT }}
|
|
@@ -1,23 +1,25 @@
|
|
|
1
1
|
# Copy this file to .env and fill in your credentials
|
|
2
2
|
|
|
3
|
+
# Eval run mode (used by agentv-bench skill)
|
|
4
|
+
AGENT_EVAL_MODE=agent # agent | cli
|
|
5
|
+
|
|
3
6
|
# Azure OpenAI Configuration
|
|
4
7
|
AZURE_OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com/
|
|
5
8
|
AZURE_OPENAI_API_KEY=your-openai-api-key-here
|
|
6
|
-
AZURE_DEPLOYMENT_NAME=gpt-5-
|
|
9
|
+
AZURE_DEPLOYMENT_NAME=gpt-5-mini
|
|
7
10
|
AZURE_OPENAI_API_VERSION=2024-12-01-preview
|
|
8
11
|
|
|
12
|
+
# OpenAI
|
|
13
|
+
OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com/
|
|
14
|
+
OPENAI_API_KEY=your-openai-api-key-here
|
|
15
|
+
OPENAI_MODEL=gpt-5-mini
|
|
16
|
+
|
|
9
17
|
# Google Gemini
|
|
10
18
|
GOOGLE_GENERATIVE_AI_API_KEY=your-gemini-api-key-here
|
|
11
|
-
GEMINI_MODEL_NAME=gemini-
|
|
19
|
+
GEMINI_MODEL_NAME=gemini-3-flash-preview
|
|
12
20
|
|
|
13
21
|
# Anthropic
|
|
14
22
|
ANTHROPIC_API_KEY=your-anthropic-api-key-here
|
|
15
23
|
|
|
16
|
-
# VS Code Workspace Paths for Execution Targets
|
|
17
|
-
# Note: Using forward slashes is recommended for paths in .env files
|
|
18
|
-
# to avoid issues with escape characters.
|
|
19
|
-
PROJECTX_WORKSPACE_PATH=C:/Users/your-username/OneDrive - Company Pty Ltd/sample.code-workspace
|
|
20
|
-
|
|
21
24
|
# CLI provider sample (used by the local_cli target)
|
|
22
|
-
CLI_EVALS_DIR=./docs/examples/simple/evals/local-cli
|
|
23
|
-
LOCAL_AGENT_TOKEN=dummytoken
|
|
25
|
+
CLI_EVALS_DIR=./docs/examples/simple/evals/local-cli
|