agentv 2.17.2 → 2.18.1
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/README.md +13 -5
- package/dist/{chunk-MH2ZEUAO.js → chunk-JSA3EH44.js} +5 -49
- package/dist/chunk-JSA3EH44.js.map +1 -0
- package/dist/{chunk-SO4O4O2B.js → chunk-NWQTM4ZK.js} +40 -20
- package/dist/chunk-NWQTM4ZK.js.map +1 -0
- package/dist/{chunk-QAZU7YAH.js → chunk-YSLP2HLP.js} +33 -41
- package/dist/chunk-YSLP2HLP.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-AQVAKXMK.js → dist-QE7SFCGD.js} +2 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-T2FAFLD2.js → interactive-MQODOYWE.js} +3 -3
- package/package.json +1 -1
- package/dist/chunk-MH2ZEUAO.js.map +0 -1
- package/dist/chunk-QAZU7YAH.js.map +0 -1
- package/dist/chunk-SO4O4O2B.js.map +0 -1
- /package/dist/{dist-AQVAKXMK.js.map → dist-QE7SFCGD.js.map} +0 -0
- /package/dist/{interactive-T2FAFLD2.js.map → interactive-MQODOYWE.js.map} +0 -0
package/README.md
CHANGED
|
@@ -119,21 +119,28 @@ See [AGENTS.md](AGENTS.md) for development guidelines and design principles.
|
|
|
119
119
|
|
|
120
120
|
### Releasing
|
|
121
121
|
|
|
122
|
-
|
|
122
|
+
Version bump:
|
|
123
123
|
|
|
124
124
|
```bash
|
|
125
125
|
bun run release # patch bump
|
|
126
126
|
bun run release minor
|
|
127
127
|
bun run release major
|
|
128
|
-
bun run publish # publish to npm `latest`
|
|
129
128
|
```
|
|
130
129
|
|
|
131
|
-
|
|
130
|
+
Canary rollout (recommended):
|
|
131
|
+
|
|
132
|
+
```bash
|
|
133
|
+
bun run publish:next # publish current version to npm `next`
|
|
134
|
+
bun run promote:latest # promote same version to npm `latest`
|
|
135
|
+
bun run tag:next 2.18.0 # point npm `next` to an explicit version
|
|
136
|
+
bun run promote:latest 2.18.0 # point npm `latest` to an explicit version
|
|
137
|
+
```
|
|
138
|
+
|
|
139
|
+
Legacy prerelease flow (still available):
|
|
132
140
|
|
|
133
141
|
```bash
|
|
134
142
|
bun run release:next # bump/increment `-next.N`
|
|
135
143
|
bun run release:next major # start new major prerelease line
|
|
136
|
-
bun run publish:next # publish to npm `next`
|
|
137
144
|
```
|
|
138
145
|
|
|
139
146
|
## Core Concepts
|
|
@@ -347,7 +354,6 @@ targets:
|
|
|
347
354
|
|
|
348
355
|
- name: vscode_dev
|
|
349
356
|
provider: vscode
|
|
350
|
-
workspace_template: ${{ WORKSPACE_PATH }}
|
|
351
357
|
judge_target: azure-base
|
|
352
358
|
|
|
353
359
|
- name: local_agent
|
|
@@ -358,6 +364,8 @@ targets:
|
|
|
358
364
|
|
|
359
365
|
Supports: `azure`, `anthropic`, `gemini`, `codex`, `copilot`, `pi-coding-agent`, `claude`, `vscode`, `vscode-insiders`, `cli`, and `mock`.
|
|
360
366
|
|
|
367
|
+
Workspace templates are configured at eval-level under `workspace.template` (not per-target `workspace_template`).
|
|
368
|
+
|
|
361
369
|
Use `${{ VARIABLE_NAME }}` syntax to reference your `.env` file. See `.agentv/targets.yaml` after `agentv init` for detailed examples and all provider-specific fields.
|
|
362
370
|
|
|
363
371
|
## Evaluation Features
|
|
@@ -11,7 +11,7 @@ import {
|
|
|
11
11
|
validateEvalFile,
|
|
12
12
|
validateFileReferences,
|
|
13
13
|
validateTargetsFile
|
|
14
|
-
} from "./chunk-
|
|
14
|
+
} from "./chunk-YSLP2HLP.js";
|
|
15
15
|
import {
|
|
16
16
|
assembleLlmJudgePrompt,
|
|
17
17
|
buildPromptInputs,
|
|
@@ -27,7 +27,7 @@ import {
|
|
|
27
27
|
toCamelCaseDeep,
|
|
28
28
|
toSnakeCaseDeep as toSnakeCaseDeep2,
|
|
29
29
|
trimBaselineResult
|
|
30
|
-
} from "./chunk-
|
|
30
|
+
} from "./chunk-NWQTM4ZK.js";
|
|
31
31
|
import {
|
|
32
32
|
__commonJS,
|
|
33
33
|
__esm,
|
|
@@ -3957,52 +3957,16 @@ var evalRunCommand = command({
|
|
|
3957
3957
|
long: "verbose",
|
|
3958
3958
|
description: "Enable verbose logging"
|
|
3959
3959
|
}),
|
|
3960
|
-
keepWorkspaces: flag({
|
|
3961
|
-
long: "keep-workspaces",
|
|
3962
|
-
description: "Always keep temporary workspaces after evaluation (default: keep on failure only)"
|
|
3963
|
-
}),
|
|
3964
|
-
cleanupWorkspaces: flag({
|
|
3965
|
-
long: "cleanup-workspaces",
|
|
3966
|
-
description: "Always cleanup temporary workspaces, even on failure"
|
|
3967
|
-
}),
|
|
3968
|
-
poolWorkspaces: flag({
|
|
3969
|
-
long: "pool-workspaces",
|
|
3970
|
-
description: "Enable workspace pooling (default for shared workspaces with repos)"
|
|
3971
|
-
}),
|
|
3972
|
-
noPool: flag({
|
|
3973
|
-
long: "no-pool",
|
|
3974
|
-
description: "Disable workspace pooling (clone fresh each run)"
|
|
3975
|
-
}),
|
|
3976
|
-
workspace: option({
|
|
3977
|
-
type: optional(string),
|
|
3978
|
-
long: "workspace",
|
|
3979
|
-
description: "Use an existing directory as the workspace directly (skips clone/copy/pool)"
|
|
3980
|
-
}),
|
|
3981
3960
|
workspaceMode: option({
|
|
3982
3961
|
type: optional(string),
|
|
3983
3962
|
long: "workspace-mode",
|
|
3984
|
-
description: "Workspace mode: 'pooled', '
|
|
3963
|
+
description: "Workspace mode: 'pooled', 'temp', or 'static'"
|
|
3985
3964
|
}),
|
|
3986
3965
|
workspacePath: option({
|
|
3987
3966
|
type: optional(string),
|
|
3988
3967
|
long: "workspace-path",
|
|
3989
3968
|
description: "Static workspace directory path (used when workspace mode is static)"
|
|
3990
3969
|
}),
|
|
3991
|
-
workspaceClean: option({
|
|
3992
|
-
type: optional(string),
|
|
3993
|
-
long: "workspace-clean",
|
|
3994
|
-
description: "Pooled reset clean mode: 'standard' or 'full'"
|
|
3995
|
-
}),
|
|
3996
|
-
retainOnSuccess: option({
|
|
3997
|
-
type: optional(string),
|
|
3998
|
-
long: "retain-on-success",
|
|
3999
|
-
description: "Workspace retention on success: 'keep' or 'cleanup'"
|
|
4000
|
-
}),
|
|
4001
|
-
retainOnFailure: option({
|
|
4002
|
-
type: optional(string),
|
|
4003
|
-
long: "retain-on-failure",
|
|
4004
|
-
description: "Workspace retention on failure: 'keep' or 'cleanup'"
|
|
4005
|
-
}),
|
|
4006
3970
|
otelFile: option({
|
|
4007
3971
|
type: optional(string),
|
|
4008
3972
|
long: "otel-file",
|
|
@@ -4042,7 +4006,7 @@ var evalRunCommand = command({
|
|
|
4042
4006
|
},
|
|
4043
4007
|
handler: async (args) => {
|
|
4044
4008
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
4045
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
4009
|
+
const { launchInteractiveWizard } = await import("./interactive-MQODOYWE.js");
|
|
4046
4010
|
await launchInteractiveWizard();
|
|
4047
4011
|
return;
|
|
4048
4012
|
}
|
|
@@ -4064,16 +4028,8 @@ var evalRunCommand = command({
|
|
|
4064
4028
|
cache: args.cache,
|
|
4065
4029
|
noCache: args.noCache,
|
|
4066
4030
|
verbose: args.verbose,
|
|
4067
|
-
keepWorkspaces: args.keepWorkspaces,
|
|
4068
|
-
cleanupWorkspaces: args.cleanupWorkspaces,
|
|
4069
|
-
poolWorkspaces: args.poolWorkspaces,
|
|
4070
|
-
noPool: args.noPool,
|
|
4071
|
-
workspace: args.workspace,
|
|
4072
4031
|
workspaceMode: args.workspaceMode,
|
|
4073
4032
|
workspacePath: args.workspacePath,
|
|
4074
|
-
workspaceClean: args.workspaceClean,
|
|
4075
|
-
retainOnSuccess: args.retainOnSuccess,
|
|
4076
|
-
retainOnFailure: args.retainOnFailure,
|
|
4077
4033
|
trace: false,
|
|
4078
4034
|
otelFile: args.otelFile,
|
|
4079
4035
|
traceFile: args.traceFile,
|
|
@@ -5929,4 +5885,4 @@ export {
|
|
|
5929
5885
|
preprocessArgv,
|
|
5930
5886
|
runCli
|
|
5931
5887
|
};
|
|
5932
|
-
//# sourceMappingURL=chunk-
|
|
5888
|
+
//# sourceMappingURL=chunk-JSA3EH44.js.map
|