@uinaf/skillcheck 1.7.2 → 1.7.3
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/scenario.js +2 -0
- package/docs/usage.md +8 -0
- package/package.json +4 -4
package/dist/scenario.js
CHANGED
|
@@ -286,6 +286,8 @@ function buildConfig(s, trials, opts, paths) {
|
|
|
286
286
|
...opts.judgeEffort ? { effort: opts.judgeEffort } : {},
|
|
287
287
|
apiKeyRequired: false,
|
|
288
288
|
max_turns: 3,
|
|
289
|
+
env: { CLAUDE_CODE_DISABLE_AUTO_MEMORY: "1" },
|
|
290
|
+
title: "skillcheck judge",
|
|
289
291
|
output_format: {
|
|
290
292
|
type: "json_schema",
|
|
291
293
|
schema: {
|
package/docs/usage.md
CHANGED
|
@@ -174,6 +174,14 @@ skillcheck run <scenario-dir> --agent claude-opus-5-5 --agent-effort medium \
|
|
|
174
174
|
--judge claude-opus-5-5 --judge-effort high --trials 3
|
|
175
175
|
```
|
|
176
176
|
|
|
177
|
+
The SDK judge runs Claude Code with auto memory off and a fixed session title.
|
|
178
|
+
Each call gets a fresh temp working directory, and the memory section would
|
|
179
|
+
put that path into the system prompt; without it, every call sends the same
|
|
180
|
+
system prompt, which the prompt cache can share. A gateway that moves the
|
|
181
|
+
caller's system prompt behind the first user message prevents that sharing.
|
|
182
|
+
The fixed title saves the second request, on the judge model, that Claude Code
|
|
183
|
+
otherwise sends per call to name the session.
|
|
184
|
+
|
|
177
185
|
## Sweep
|
|
178
186
|
|
|
179
187
|
```sh
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@uinaf/skillcheck",
|
|
3
|
-
"version": "1.7.
|
|
3
|
+
"version": "1.7.3",
|
|
4
4
|
"description": "Lint and eval harness for agent skills",
|
|
5
5
|
"homepage": "https://github.com/uinaf/skillcheck#readme",
|
|
6
6
|
"bugs": {
|
|
@@ -34,7 +34,7 @@
|
|
|
34
34
|
},
|
|
35
35
|
"devDependencies": {
|
|
36
36
|
"@anthropic-ai/claude-agent-sdk": "^0.3.281",
|
|
37
|
-
"@openai/codex-sdk": "^0.
|
|
37
|
+
"@openai/codex-sdk": "^0.157.0",
|
|
38
38
|
"@types/node": "^26.6.2",
|
|
39
39
|
"promptfoo": "^0.123.1",
|
|
40
40
|
"vite": "catalog:",
|
|
@@ -42,7 +42,7 @@
|
|
|
42
42
|
},
|
|
43
43
|
"peerDependencies": {
|
|
44
44
|
"@anthropic-ai/claude-agent-sdk": "^0.3.281",
|
|
45
|
-
"@openai/codex-sdk": "^0.156.1",
|
|
45
|
+
"@openai/codex-sdk": "^0.156.1 || ^0.157.0",
|
|
46
46
|
"promptfoo": "^0.123.1"
|
|
47
47
|
},
|
|
48
48
|
"peerDependenciesMeta": {
|
|
@@ -66,5 +66,5 @@
|
|
|
66
66
|
"engines": {
|
|
67
67
|
"node": ">=24"
|
|
68
68
|
},
|
|
69
|
-
"packageManager": "pnpm@12.
|
|
69
|
+
"packageManager": "pnpm@12.7.0"
|
|
70
70
|
}
|