@cat-factory/sandbox 0.7.36 → 0.8.0

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.
@@ -1,3 +1,4 @@
1
+ import type { SandboxFixtureKind } from '@cat-factory/contracts';
1
2
  import type { SandboxPromptVersion } from '@cat-factory/kernel';
2
3
  import type { SandboxTaskType } from './rubrics.js';
3
4
  export type SandboxAgentBucket = 'inline' | 'container';
@@ -10,6 +11,12 @@ export interface SandboxAgentKindMeta {
10
11
  bucket: SandboxAgentBucket;
11
12
  /** Which rubric the judge grades this kind's output against. */
12
13
  rubric: SandboxTaskType;
14
+ /**
15
+ * The fixture kinds this agent is exercised against (the fixture↔kind mapping the UI
16
+ * filters the library by). Source of truth here so the frontend reads it off the catalog
17
+ * instead of re-encoding the mapping in a parallel switch that can silently drift.
18
+ */
19
+ fixtureKinds: readonly SandboxFixtureKind[];
13
20
  /**
14
21
  * The version-controlled baseline prompt id (a `PROMPT_VERSIONS` key) this kind's
15
22
  * system prompt comes from. When null, the baseline text is read from
@@ -1 +1 @@
1
- {"version":3,"file":"baselines.d.ts","sourceRoot":"","sources":["../src/baselines.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAA;AAC/D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,cAAc,CAAA;AAWnD,MAAM,MAAM,kBAAkB,GAAG,QAAQ,GAAG,WAAW,CAAA;AAEvD,MAAM,WAAW,oBAAoB;IACnC,4EAA4E;IAC5E,SAAS,EAAE,MAAM,CAAA;IACjB,0DAA0D;IAC1D,KAAK,EAAE,MAAM,CAAA;IACb,gFAAgF;IAChF,MAAM,EAAE,kBAAkB,CAAA;IAC1B,gEAAgE;IAChE,MAAM,EAAE,eAAe,CAAA;IACvB;;;;OAIG;IACH,YAAY,EAAE,MAAM,GAAG,IAAI,CAAA;CAC5B;AAED,4FAA4F;AAC5F,eAAO,MAAM,mBAAmB,EAAE,SAAS,oBAAoB,EAwC9D,CAAA;AAMD,0FAA0F;AAC1F,wBAAgB,eAAe,CAAC,SAAS,EAAE,MAAM,GAAG,oBAAoB,GAAG,SAAS,CAEnF;AAED,iFAAiF;AACjF,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,oBAAoB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAM9F;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,GAAG,EAAE,MAAM,GAAG,oBAAoB,EAAE,CAmBjE"}
1
+ {"version":3,"file":"baselines.d.ts","sourceRoot":"","sources":["../src/baselines.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,wBAAwB,CAAA;AAChE,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAA;AAC/D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,cAAc,CAAA;AAWnD,MAAM,MAAM,kBAAkB,GAAG,QAAQ,GAAG,WAAW,CAAA;AAEvD,MAAM,WAAW,oBAAoB;IACnC,4EAA4E;IAC5E,SAAS,EAAE,MAAM,CAAA;IACjB,0DAA0D;IAC1D,KAAK,EAAE,MAAM,CAAA;IACb,gFAAgF;IAChF,MAAM,EAAE,kBAAkB,CAAA;IAC1B,gEAAgE;IAChE,MAAM,EAAE,eAAe,CAAA;IACvB;;;;OAIG;IACH,YAAY,EAAE,SAAS,kBAAkB,EAAE,CAAA;IAC3C;;;;OAIG;IACH,YAAY,EAAE,MAAM,GAAG,IAAI,CAAA;CAC5B;AAED,4FAA4F;AAC5F,eAAO,MAAM,mBAAmB,EAAE,SAAS,oBAAoB,EA6C9D,CAAA;AAMD,0FAA0F;AAC1F,wBAAgB,eAAe,CAAC,SAAS,EAAE,MAAM,GAAG,oBAAoB,GAAG,SAAS,CAEnF;AAED,iFAAiF;AACjF,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,oBAAoB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAM9F;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,GAAG,EAAE,MAAM,GAAG,oBAAoB,EAAE,CAmBjE"}
package/dist/baselines.js CHANGED
@@ -6,6 +6,7 @@ export const SANDBOX_AGENT_KINDS = [
6
6
  label: 'Requirements review',
7
7
  bucket: 'inline',
8
8
  rubric: 'requirement-review',
9
+ fixtureKinds: ['requirements'],
9
10
  basePromptId: 'requirement-review',
10
11
  },
11
12
  {
@@ -13,6 +14,7 @@ export const SANDBOX_AGENT_KINDS = [
13
14
  label: 'Clarity (bug-report) review',
14
15
  bucket: 'inline',
15
16
  rubric: 'requirement-review',
17
+ fixtureKinds: ['clarity'],
16
18
  basePromptId: 'clarity-review',
17
19
  },
18
20
  {
@@ -20,6 +22,7 @@ export const SANDBOX_AGENT_KINDS = [
20
22
  label: 'Code reviewer',
21
23
  bucket: 'inline',
22
24
  rubric: 'code-review',
25
+ fixtureKinds: ['code-review'],
23
26
  basePromptId: 'review',
24
27
  },
25
28
  {
@@ -31,6 +34,7 @@ export const SANDBOX_AGENT_KINDS = [
31
34
  label: 'Architecture-proposal review',
32
35
  bucket: 'inline',
33
36
  rubric: 'requirement-review',
37
+ fixtureKinds: ['architecture'],
34
38
  basePromptId: null,
35
39
  },
36
40
  {
@@ -38,6 +42,7 @@ export const SANDBOX_AGENT_KINDS = [
38
42
  label: 'Coder (implementation)',
39
43
  bucket: 'container',
40
44
  rubric: 'implementation',
45
+ fixtureKinds: ['repo-feature', 'repo-bug'],
41
46
  basePromptId: 'build',
42
47
  },
43
48
  ];
@@ -1 +1 @@
1
- {"version":3,"file":"baselines.js","sourceRoot":"","sources":["../src/baselines.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAA;AAgC1F,4FAA4F;AAC5F,MAAM,CAAC,MAAM,mBAAmB,GAAoC;IAClE;QACE,SAAS,EAAE,qBAAqB;QAChC,KAAK,EAAE,qBAAqB;QAC5B,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,oBAAoB;KACnC;IACD;QACE,SAAS,EAAE,gBAAgB;QAC3B,KAAK,EAAE,6BAA6B;QACpC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,gBAAgB;KAC/B;IACD;QACE,SAAS,EAAE,UAAU;QACrB,KAAK,EAAE,eAAe;QACtB,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,aAAa;QACrB,YAAY,EAAE,QAAQ;KACvB;IACD;QACE,kFAAkF;QAClF,8EAA8E;QAC9E,oFAAoF;QACpF,2DAA2D;QAC3D,SAAS,EAAE,qBAAqB;QAChC,KAAK,EAAE,8BAA8B;QACrC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,IAAI;KACnB;IACD;QACE,SAAS,EAAE,OAAO;QAClB,KAAK,EAAE,wBAAwB;QAC/B,MAAM,EAAE,WAAW;QACnB,MAAM,EAAE,gBAAgB;QACxB,YAAY,EAAE,OAAO;KACtB;CACF,CAAA;AAED,MAAM,OAAO,GAAG,IAAI,GAAG,CACrB,mBAAmB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,CACjD,CAAA;AAED,0FAA0F;AAC1F,MAAM,UAAU,eAAe,CAAC,SAAiB;IAC/C,OAAO,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAA;AAC/B,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,kBAAkB,CAAC,IAA0B;IAC3D,IAAI,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,YAAY,IAAI,eAAe,EAAE,CAAC;QAC9D,MAAM,SAAS,GAAG,eAAe,CAAC,IAAI,CAAC,YAA4C,CAAC,CAAA;QACpF,OAAO,EAAE,IAAI,EAAE,SAAS,CAAC,IAAI,EAAE,KAAK,EAAE,kBAAkB,CAAC,SAAS,CAAC,EAAE,EAAE,SAAS,CAAC,OAAO,CAAC,EAAE,CAAA;IAC7F,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,eAAe,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,kBAAkB,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC,EAAE,CAAA;AAChG,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAAC,GAAW;IACvC,OAAO,mBAAmB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QACtC,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,kBAAkB,CAAC,IAAI,CAAC,CAAA;QAChD,OAAO;YACL,EAAE,EAAE,YAAY,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,SAAS,EAAE;YACrD,SAAS,EAAE,YAAY,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,SAAS,EAAE;YAC5D,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,IAAI,EAAE,KAAK;YACX,MAAM,EAAE,UAAU;YAClB,UAAU,EAAE,IAAI;YAChB,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,OAAO,EAAE,CAAC;YACV,QAAQ,EAAE,IAAI;YACd,MAAM,EAAE,EAAE;YACV,SAAS,EAAE,GAAG;YACd,SAAS,EAAE,IAAI;YACf,UAAU,EAAE,IAAI;SACjB,CAAA;IACH,CAAC,CAAC,CAAA;AACJ,CAAC"}
1
+ {"version":3,"file":"baselines.js","sourceRoot":"","sources":["../src/baselines.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAA;AAuC1F,4FAA4F;AAC5F,MAAM,CAAC,MAAM,mBAAmB,GAAoC;IAClE;QACE,SAAS,EAAE,qBAAqB;QAChC,KAAK,EAAE,qBAAqB;QAC5B,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,CAAC,cAAc,CAAC;QAC9B,YAAY,EAAE,oBAAoB;KACnC;IACD;QACE,SAAS,EAAE,gBAAgB;QAC3B,KAAK,EAAE,6BAA6B;QACpC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,CAAC,SAAS,CAAC;QACzB,YAAY,EAAE,gBAAgB;KAC/B;IACD;QACE,SAAS,EAAE,UAAU;QACrB,KAAK,EAAE,eAAe;QACtB,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,aAAa;QACrB,YAAY,EAAE,CAAC,aAAa,CAAC;QAC7B,YAAY,EAAE,QAAQ;KACvB;IACD;QACE,kFAAkF;QAClF,8EAA8E;QAC9E,oFAAoF;QACpF,2DAA2D;QAC3D,SAAS,EAAE,qBAAqB;QAChC,KAAK,EAAE,8BAA8B;QACrC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,CAAC,cAAc,CAAC;QAC9B,YAAY,EAAE,IAAI;KACnB;IACD;QACE,SAAS,EAAE,OAAO;QAClB,KAAK,EAAE,wBAAwB;QAC/B,MAAM,EAAE,WAAW;QACnB,MAAM,EAAE,gBAAgB;QACxB,YAAY,EAAE,CAAC,cAAc,EAAE,UAAU,CAAC;QAC1C,YAAY,EAAE,OAAO;KACtB;CACF,CAAA;AAED,MAAM,OAAO,GAAG,IAAI,GAAG,CACrB,mBAAmB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,CACjD,CAAA;AAED,0FAA0F;AAC1F,MAAM,UAAU,eAAe,CAAC,SAAiB;IAC/C,OAAO,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAA;AAC/B,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,kBAAkB,CAAC,IAA0B;IAC3D,IAAI,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,YAAY,IAAI,eAAe,EAAE,CAAC;QAC9D,MAAM,SAAS,GAAG,eAAe,CAAC,IAAI,CAAC,YAA4C,CAAC,CAAA;QACpF,OAAO,EAAE,IAAI,EAAE,SAAS,CAAC,IAAI,EAAE,KAAK,EAAE,kBAAkB,CAAC,SAAS,CAAC,EAAE,EAAE,SAAS,CAAC,OAAO,CAAC,EAAE,CAAA;IAC7F,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,eAAe,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,kBAAkB,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC,EAAE,CAAA;AAChG,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAAC,GAAW;IACvC,OAAO,mBAAmB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QACtC,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,kBAAkB,CAAC,IAAI,CAAC,CAAA;QAChD,OAAO;YACL,EAAE,EAAE,YAAY,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,SAAS,EAAE;YACrD,SAAS,EAAE,YAAY,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,SAAS,EAAE;YAC5D,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,IAAI,EAAE,KAAK;YACX,MAAM,EAAE,UAAU;YAClB,UAAU,EAAE,IAAI;YAChB,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,OAAO,EAAE,CAAC;YACV,QAAQ,EAAE,IAAI;YACd,MAAM,EAAE,EAAE;YACV,SAAS,EAAE,GAAG;YACd,SAAS,EAAE,IAAI;YACf,UAAU,EAAE,IAAI;SACjB,CAAA;IACH,CAAC,CAAC,CAAA;AACJ,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@cat-factory/sandbox",
3
- "version": "0.7.36",
3
+ "version": "0.8.0",
4
4
  "description": "Parallel prompt/model testing surface: versioned prompt candidates, experiment matrices, judge + objective grading. Isolated from the core product so it can be extracted.",
5
5
  "repository": {
6
6
  "type": "git",
@@ -24,10 +24,10 @@
24
24
  "access": "public"
25
25
  },
26
26
  "dependencies": {
27
- "@cat-factory/agents": "0.14.6",
28
- "@cat-factory/contracts": "0.24.0",
29
- "@cat-factory/kernel": "0.27.0",
30
- "@cat-factory/sandbox-fixtures": "0.7.21"
27
+ "@cat-factory/agents": "0.14.7",
28
+ "@cat-factory/contracts": "0.25.0",
29
+ "@cat-factory/kernel": "0.28.0",
30
+ "@cat-factory/sandbox-fixtures": "0.7.22"
31
31
  },
32
32
  "devDependencies": {
33
33
  "typescript": "7.0.1-rc",