agentv 3.13.0 → 3.13.2

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/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-DJU4C6NS.js";
6
- import "./chunk-6H4IAXQH.js";
7
- import "./chunk-7OHZAFND.js";
5
+ } from "./chunk-X2343WOK.js";
6
+ import "./chunk-4Z5E5CYT.js";
7
+ import "./chunk-D3LNJUUB.js";
8
8
  import "./chunk-C5GOHBQM.js";
9
9
  import "./chunk-HQDCIXVH.js";
10
10
  import "./chunk-BJV6MDBE.js";
@@ -68,6 +68,7 @@ import {
68
68
  extractTargetFromSuite,
69
69
  extractTargetsFromSuite,
70
70
  extractTargetsFromTestCase,
71
+ extractThreshold,
71
72
  extractTrialsConfig,
72
73
  extractWorkersFromSuite,
73
74
  fileExists,
@@ -141,7 +142,7 @@ import {
141
142
  transpileEvalYaml,
142
143
  transpileEvalYamlFile,
143
144
  trimBaselineResult
144
- } from "./chunk-7OHZAFND.js";
145
+ } from "./chunk-D3LNJUUB.js";
145
146
  import {
146
147
  OtlpJsonFileExporter
147
148
  } from "./chunk-C5GOHBQM.js";
@@ -222,6 +223,7 @@ export {
222
223
  extractTargetFromSuite,
223
224
  extractTargetsFromSuite,
224
225
  extractTargetsFromTestCase,
226
+ extractThreshold,
225
227
  extractTrialsConfig,
226
228
  extractWorkersFromSuite,
227
229
  fileExists,
@@ -296,4 +298,4 @@ export {
296
298
  transpileEvalYamlFile,
297
299
  trimBaselineResult
298
300
  };
299
- //# sourceMappingURL=dist-SMKOBBFB.js.map
301
+ //# sourceMappingURL=dist-KPMR7RBT.js.map
package/dist/index.js CHANGED
@@ -3,9 +3,9 @@ import {
3
3
  app,
4
4
  preprocessArgv,
5
5
  runCli
6
- } from "./chunk-DJU4C6NS.js";
7
- import "./chunk-6H4IAXQH.js";
8
- import "./chunk-7OHZAFND.js";
6
+ } from "./chunk-X2343WOK.js";
7
+ import "./chunk-4Z5E5CYT.js";
8
+ import "./chunk-D3LNJUUB.js";
9
9
  import "./chunk-C5GOHBQM.js";
10
10
  import "./chunk-HQDCIXVH.js";
11
11
  import "./chunk-BJV6MDBE.js";
@@ -4,14 +4,14 @@ import {
4
4
  fileExists,
5
5
  findRepoRoot,
6
6
  runEvalCommand
7
- } from "./chunk-6H4IAXQH.js";
7
+ } from "./chunk-4Z5E5CYT.js";
8
8
  import {
9
9
  DEFAULT_EVAL_PATTERNS,
10
10
  getAgentvHome,
11
11
  listTargetNames,
12
12
  loadConfig,
13
13
  readTargetDefinitions
14
- } from "./chunk-7OHZAFND.js";
14
+ } from "./chunk-D3LNJUUB.js";
15
15
  import "./chunk-C5GOHBQM.js";
16
16
  import "./chunk-HQDCIXVH.js";
17
17
  import "./chunk-BJV6MDBE.js";
@@ -370,4 +370,4 @@ ${ANSI_DIM}Retrying execution errors...${ANSI_RESET}
370
370
  export {
371
371
  launchInteractiveWizard
372
372
  };
373
- //# sourceMappingURL=interactive-RV664PCR.js.map
373
+ //# sourceMappingURL=interactive-HVKLYGRX.js.map
@@ -0,0 +1,23 @@
1
+ # Copy this file to .env and fill in your credentials
2
+
3
+ # Azure OpenAI Configuration
4
+ AZURE_OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com/
5
+ AZURE_OPENAI_API_KEY=your-openai-api-key-here
6
+ AZURE_DEPLOYMENT_NAME=gpt-5-chat
7
+ AZURE_OPENAI_API_VERSION=2024-12-01-preview
8
+
9
+ # Google Gemini
10
+ GOOGLE_GENERATIVE_AI_API_KEY=your-gemini-api-key-here
11
+ GEMINI_MODEL_NAME=gemini-2.5-flash
12
+
13
+ # Anthropic
14
+ ANTHROPIC_API_KEY=your-anthropic-api-key-here
15
+
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
+ # CLI provider sample (used by the local_cli target)
22
+ CLI_EVALS_DIR=./docs/examples/simple/evals/local-cli
23
+ LOCAL_AGENT_TOKEN=dummytoken
@@ -1,6 +1,15 @@
1
1
  $schema: agentv-config-v2
2
2
 
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)
3
+ # Customize which files are treated as guidelines vs regular file content
4
+
5
+ # Custom guideline patterns:
6
+ guideline_patterns:
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
@@ -10,6 +10,10 @@ 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
+
13
17
  - name: codex
14
18
  provider: codex
15
19
  judge_target: azure-llm
@@ -39,6 +43,18 @@ targets:
39
43
  log_format: json # Optional: 'summary' (default) or 'json' for raw event logs
40
44
  # system_prompt: optional override (default instructs agent to include code in response)
41
45
 
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
+
42
58
  - name: azure-llm
43
59
  provider: azure
44
60
  endpoint: ${{ AZURE_OPENAI_ENDPOINT }}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "agentv",
3
- "version": "3.13.0",
3
+ "version": "3.13.2",
4
4
  "description": "CLI entry point for AgentV",
5
5
  "type": "module",
6
6
  "repository": {