@post-print/agent-test 0.1.13 → 0.1.15
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 +72 -0
- package/dist/cli.d.ts +24 -1
- package/dist/cli.d.ts.map +1 -1
- package/dist/cli.js +58 -8
- package/dist/cli.js.map +1 -1
- package/dist/debug-bundle.d.ts +66 -0
- package/dist/debug-bundle.d.ts.map +1 -0
- package/dist/debug-bundle.js +241 -0
- package/dist/debug-bundle.js.map +1 -0
- package/dist/expect.d.ts +9 -0
- package/dist/expect.d.ts.map +1 -1
- package/dist/expect.js +125 -46
- package/dist/expect.js.map +1 -1
- package/dist/failures.d.ts +4 -0
- package/dist/failures.d.ts.map +1 -0
- package/dist/failures.js +7 -0
- package/dist/failures.js.map +1 -0
- package/dist/index.d.ts +4 -2
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +3 -1
- package/dist/index.js.map +1 -1
- package/dist/live-isolation.d.ts +7 -9
- package/dist/live-isolation.d.ts.map +1 -1
- package/dist/live-isolation.js +37 -10
- package/dist/live-isolation.js.map +1 -1
- package/dist/load-suite.d.ts.map +1 -1
- package/dist/load-suite.js +18 -7
- package/dist/load-suite.js.map +1 -1
- package/dist/mcp-config.d.ts +5 -0
- package/dist/mcp-config.d.ts.map +1 -0
- package/dist/mcp-config.js +75 -0
- package/dist/mcp-config.js.map +1 -0
- package/dist/record-trace.d.ts +8 -0
- package/dist/record-trace.d.ts.map +1 -1
- package/dist/record-trace.js +23 -4
- package/dist/record-trace.js.map +1 -1
- package/dist/run-suite.d.ts +6 -0
- package/dist/run-suite.d.ts.map +1 -1
- package/dist/run-suite.js +193 -61
- package/dist/run-suite.js.map +1 -1
- package/dist/theme.d.ts +9 -1
- package/dist/theme.d.ts.map +1 -1
- package/dist/theme.js +30 -6
- package/dist/theme.js.map +1 -1
- package/dist/types.d.ts +30 -2
- package/dist/types.d.ts.map +1 -1
- package/package.json +2 -2
package/dist/theme.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"theme.js","sourceRoot":"","sources":["../src/theme.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;
|
|
1
|
+
{"version":3,"file":"theme.js","sourceRoot":"","sources":["../src/theme.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC3C,OAAO,KAAK,MAAM,OAAO,CAAC;AAE1B,MAAM,mBAAmB,GAAG,EAAE,CAAC;AAE/B,+EAA+E;AAC/E,MAAM,UAAU,iBAAiB;IAChC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;AACjB,CAAC;AAED,MAAM,UAAU,YAAY;IAC3B,OAAO,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;AACxB,CAAC;AAED,uEAAuE;AACvE,MAAM,UAAU,YAAY,CAAC,IAAY;IACxC,IACC,OAAO,CAAC,GAAG,CAAC,wBAAwB,KAAK,GAAG;QAC5C,OAAO,CAAC,GAAG,CAAC,gBAAgB,KAAK,GAAG;QACpC,OAAO,CAAC,GAAG,CAAC,gBAAgB,KAAK,MAAM,EACtC,CAAC;QACF,OAAO,IAAI,CAAC;IACb,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAC9C,IAAI,KAAK,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QACvB,OAAO,IAAI,CAAC;IACb,CAAC;IACD,OAAO,KAAK,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC;AACzC,CAAC;AAED,qEAAqE;AACrE,MAAM,UAAU,QAAQ,CAAC,IAAY,EAAE,IAAI,GAAG,mBAAmB;IAChE,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACvD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,EAAE,CAAC;IACX,CAAC;IACD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC7B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACvC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAC5B,IAAI,GAAG,OAAO,IAAI,IAAI,EAAE,CAAC,MAAM,IAAI,IAAI,EAAE,CAAC;YACzC,OAAO,GAAG,GAAG,OAAO,IAAI,IAAI,EAAE,CAAC;QAChC,CAAC;aAAM,CAAC;YACP,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YACpB,OAAO,GAAG,IAAI,CAAC;QAChB,CAAC;IACF,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACpB,OAAO,KAAK,CAAC;AACd,CAAC;AA+BD,MAAM,UAAU,mBAAmB,CAAC,EAAU;IAC7C,IAAI,EAAE,GAAG,IAAI,EAAE,CAAC;QACf,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC;IAC9B,CAAC;IACD,OAAO,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;AACrC,CAAC;AAED,SAAS,cAAc,CAAC,KAAa;IACpC,OAAO,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,GAAG,KAAK,WAAW,CAAC;AAC1D,CAAC;AAED,MAAM,CAAC,MAAM,KAAK,GAAG;IACpB,WAAW,CAAC,IAAY,EAAE,IAAY,EAAE,KAAa;QACpD,MAAM,SAAS,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,GAAG,KAAK,YAAY,CAAC;QACpE,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;IACtH,CAAC;IAED,aAAa,CAAC,KAAa,EAAE,KAAa,EAAE,IAAY,EAAE,IAAY;QACrE,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,KAAK,IAAI,KAAK,GAAG,CAAC,IAAI,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,EAAE,CAAC;IACpG,CAAC;IAED,aAAa,CAAC,IAAY,EAAE,IAAa;QACxC,OAAO,IAAI,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/F,CAAC;IAED,SAAS,CAAC,MAA4B,EAAE,OAAe;QACtD,OAAO,KAAK,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;IAC5C,CAAC;IAED,IAAI,CAAC,CAAS;QACb,OAAO,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC;IACpC,CAAC;IAED,QAAQ,CAAC,CAAS;QACjB,OAAO,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAChC,CAAC;IAED,QAAQ,CAAC,KAAa;QACrB,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC5B,CAAC;IAED,QAAQ,CAAC,OAAe;QACvB,OAAO,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAC3B,CAAC;IAED,KAAK,CAAC,KAAa,EAAE,MAAe;QACnC,IAAI,CAAC,MAAM,EAAE,CAAC;YACb,OAAO,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;QACzB,CAAC;QACD,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,MAAM,EAAE,CAAC;IACzC,CAAC;IAED,eAAe,CAAC,MAAc;QAC7B,OAAO,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IACzE,CAAC;IAED,GAAG,CAAC,OAAe;QAClB,OAAO,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAClC,CAAC;IAED,IAAI,CAAC,OAAe;QACnB,OAAO,GAAG,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;IACxD,CAAC;IAED,MAAM,CAAC,IAAI,GAAG,MAAM;QACnB,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;IAC/D,CAAC;IAED,YAAY,CAAC,OAAe;QAC3B,OAAO,KAAK,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;IAClC,CAAC;IAED,aAAa,CAAC,IAAY;QACzB,OAAO,KAAK,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,KAAK,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;IACvE,CAAC;IAED,OAAO,CAAC,KAAa,EAAE,MAAc,EAAE,MAAc,EAAE,OAAe;QACrE,MAAM,KAAK,GAAG;YACb,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG;YAC5B,KAAK,CAAC,KAAK,CAAC,GAAG,MAAM,SAAS,CAAC;YAC/B,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,MAAM,SAAS,CAAC;YAC1E,KAAK,CAAC,GAAG,CAAC,GAAG,OAAO,UAAU,CAAC;SAC/B,CAAC;QACF,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACxB,CAAC;IAED,qBAAqB;QACpB,OAAO,KAAK,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IACvC,CAAC;IAED,kBAAkB,CAAC,IAAY;QAC9B,OAAO,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC;IACtC,CAAC;IAED,cAAc,CAAC,OAAe,EAAE,OAAe,EAAE,QAAiB,EAAE,QAAiB;QACpF,MAAM,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;QAC1D,MAAM,IAAI,GAAG,SAAS,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,OAAO,EAAE,CAAC;QAChE,IAAI,CAAC,QAAQ,EAAE,CAAC;YACf,OAAO,IAAI,CAAC;QACb,CAAC;QACD,OAAO,GAAG,IAAI,aAAa,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;IAClD,CAAC;IAED,kBAAkB,CAAC,IAAY;QAC9B,OAAO,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IAC/C,CAAC;IAED,UAAU,CAAC,KAAa;QACvB,OAAO,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,GAAG,CAAC,IAAI,cAAc,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC;IACtE,CAAC;IAED,aAAa;QACZ,OAAO,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,uBAAuB,CAAC,CAAC;IAClD,CAAC;IAED,OAAO,CAAC,KAAa;QACpB,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;IACxD,CAAC;IAED,eAAe,CAAC,OAA+B;QAC9C,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC1E,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAClF,MAAM,OAAO,GACZ,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,OAAO,CAAC,KAAK,KAAK,SAAS;YACzD,CAAC,CAAC,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,IAAI;YACxC,CAAC,CAAC,EAAE,CAAC;QACP,MAAM,QAAQ,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,mBAAmB,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;QAC9E,MAAM,eAAe,GACpB,CAAC,OAAO,CAAC,MAAM,IAAI,OAAO,CAAC,eAAe;YACzC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC;YACvC,CAAC,CAAC,SAAS,CAAC;QACd,MAAM,KAAK,GAAa;YACvB,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE;YACrB,eAAe;gBACd,CAAC,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,IAAI,IAAI,MAAM,KAAK,eAAe,KAAK,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,QAAQ,EAAE;gBACxH,CAAC,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,IAAI,IAAI,MAAM,KAAK,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,QAAQ,EAAE;SACrG,CAAC;QAEF,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,aAAa,IAAI,EAAE,EAAE,CAAC;YACnD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC;YACrD,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;YAC3F,KAAK,MAAM,OAAO,IAAI,QAAQ,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;gBACnD,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,cAAc,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;YAC/D,CAAC;QACF,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,cAAc,IAAI,EAAE,EAAE,CAAC;YACpD,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAC5F,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,QAAQ,KAAK,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;YAChF,KAAK,MAAM,OAAO,IAAI,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;gBACjD,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,cAAc,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;YACtE,CAAC;YACD,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACvC,KAAK,MAAM,OAAO,IAAI,QAAQ,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAClD,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,cAAc,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;gBACnE,CAAC;YACF,CAAC;QACF,CAAC;QAED,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,cAAc,EAAE,CAAC;YAC7C,KAAK,CAAC,IAAI,CACT,KAAK,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,eAAe,CAAC,CAAC,CAAC,EAAE,CAC1H,CAAC;QACH,CAAC;QAED,OAAO,KAAK,CAAC;IACd,CAAC;CACD,CAAC"}
|
package/dist/types.d.ts
CHANGED
|
@@ -1,4 +1,5 @@
|
|
|
1
|
-
import type { AgentHost, AgentTrace, ContextProfile, SkillContextSetting } from "@post-print/agent-harness";
|
|
1
|
+
import type { AgentHost, AgentTrace, ContextProfile, McpServerConfig, SkillContextSetting } from "@post-print/agent-harness";
|
|
2
|
+
export type { McpServerConfig } from "@post-print/agent-harness";
|
|
2
3
|
export type JudgeRubricItem = string | {
|
|
3
4
|
id?: string;
|
|
4
5
|
question: string;
|
|
@@ -10,6 +11,12 @@ export interface ScenarioRubric {
|
|
|
10
11
|
must?: string[];
|
|
11
12
|
mustNot?: string[];
|
|
12
13
|
mustRun?: string[];
|
|
14
|
+
/**
|
|
15
|
+
* Tool name substring, or `name:argFragment` where argFragment must appear in JSON args.
|
|
16
|
+
* Matches built-in and MCP tool calls recorded on the trace.
|
|
17
|
+
*/
|
|
18
|
+
mustCallTool?: string[];
|
|
19
|
+
mustNotCallTool?: string[];
|
|
13
20
|
/** Skill folder names the agent must read via SKILL.md (e.g. grill, crystallize). */
|
|
14
21
|
mustInvokeSkill?: string[];
|
|
15
22
|
mustNotInvokeSkill?: string[];
|
|
@@ -25,6 +32,8 @@ export interface AgentScenario {
|
|
|
25
32
|
profile?: ContextProfile;
|
|
26
33
|
/** Override suite defaults for skill catalog loading. */
|
|
27
34
|
skills?: SkillContextSetting;
|
|
35
|
+
/** Inline MCP servers for live Cursor (merged over suite defaults by server name). */
|
|
36
|
+
mcpServers?: Record<string, McpServerConfig>;
|
|
28
37
|
/** Live-only: apply patch + commit in worktree so pr-mode branch diff exists. */
|
|
29
38
|
seedPatch?: string;
|
|
30
39
|
replayTrace?: string;
|
|
@@ -39,18 +48,35 @@ export interface AgentSuiteFile {
|
|
|
39
48
|
profile?: ContextProfile;
|
|
40
49
|
/** none | catalog | full — ambient-routing uses full for IDE parity. */
|
|
41
50
|
skills?: SkillContextSetting;
|
|
51
|
+
/** Inline MCP servers for live Cursor runs. */
|
|
52
|
+
mcpServers?: Record<string, McpServerConfig>;
|
|
42
53
|
};
|
|
43
54
|
scenarios: AgentScenario[];
|
|
44
55
|
}
|
|
45
56
|
export interface AssertionFailure {
|
|
46
57
|
matcher: string;
|
|
47
58
|
message: string;
|
|
59
|
+
/** Normalized failure class for console, bundles, and CI consumers. */
|
|
60
|
+
category: FailureCategory;
|
|
61
|
+
/** Optional diagnostic detail (printed in --debug). */
|
|
62
|
+
evidence?: string;
|
|
48
63
|
}
|
|
64
|
+
export type FailureCategory = "rubric_miss" | "judge_infra" | "agent_runtime" | "worktree_leak" | "recording_error";
|
|
49
65
|
export interface JudgeVerdictResult {
|
|
50
66
|
id: string;
|
|
51
67
|
question: string;
|
|
52
68
|
pass: boolean;
|
|
53
69
|
rationale: string;
|
|
70
|
+
infraError?: string;
|
|
71
|
+
rawSdkStatus?: string;
|
|
72
|
+
sdkError?: {
|
|
73
|
+
message?: string;
|
|
74
|
+
code?: string;
|
|
75
|
+
};
|
|
76
|
+
attempt?: number;
|
|
77
|
+
durationMs?: number;
|
|
78
|
+
transcriptChars?: number;
|
|
79
|
+
promptChars?: number;
|
|
54
80
|
}
|
|
55
81
|
export interface ScenarioResult {
|
|
56
82
|
suite: string;
|
|
@@ -61,8 +87,10 @@ export interface ScenarioResult {
|
|
|
61
87
|
durationMs: number;
|
|
62
88
|
/** LLM judge verdicts when judge criteria were evaluated. */
|
|
63
89
|
judgeVerdicts?: JudgeVerdictResult[];
|
|
64
|
-
/** Full agent transcript when available (for HTML reports). */
|
|
90
|
+
/** Full agent transcript when available (for HTML reports / debug bundles). */
|
|
65
91
|
trace?: AgentTrace;
|
|
92
|
+
/** Absolute path to the debug bundle directory when --debug wrote one. */
|
|
93
|
+
debugBundleDir?: string;
|
|
66
94
|
}
|
|
67
95
|
export interface SuiteRunReport {
|
|
68
96
|
suite: string;
|
package/dist/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,SAAS,EACT,UAAU,EACV,cAAc,EACd,mBAAmB,EACnB,MAAM,2BAA2B,CAAC;AAEnC,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG;IAAE,EAAE,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzE,MAAM,WAAW,cAAc;IAC9B,IAAI,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACjC,kGAAkG;IAClG,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,qFAAqF;IACrF,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,WAAW,CAAC,EAAE,OAAO,GAAG,UAAU,GAAG,UAAU,GAAG,MAAM,CAAC;IACzD,4DAA4D;IAC5D,KAAK,CAAC,EAAE,eAAe,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,aAAa;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,OAAO,CAAC,EAAE,cAAc,CAAC;IACzB,yDAAyD;IACzD,MAAM,CAAC,EAAE,mBAAmB,CAAC;IAC7B,iFAAiF;IACjF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,cAAc,CAAC;IACvB,IAAI,CAAC,EAAE,OAAO,CAAC;CACf;AAED,MAAM,WAAW,cAAc;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE;QACV,IAAI,CAAC,EAAE,SAAS,CAAC;QACjB,OAAO,CAAC,EAAE,cAAc,CAAC;QACzB,wEAAwE;QACxE,MAAM,CAAC,EAAE,mBAAmB,CAAC;
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,SAAS,EACT,UAAU,EACV,cAAc,EACd,eAAe,EACf,mBAAmB,EACnB,MAAM,2BAA2B,CAAC;AAEnC,YAAY,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAEjE,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG;IAAE,EAAE,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzE,MAAM,WAAW,cAAc;IAC9B,IAAI,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACjC,kGAAkG;IAClG,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,qFAAqF;IACrF,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,WAAW,CAAC,EAAE,OAAO,GAAG,UAAU,GAAG,UAAU,GAAG,MAAM,CAAC;IACzD,4DAA4D;IAC5D,KAAK,CAAC,EAAE,eAAe,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,aAAa;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,OAAO,CAAC,EAAE,cAAc,CAAC;IACzB,yDAAyD;IACzD,MAAM,CAAC,EAAE,mBAAmB,CAAC;IAC7B,sFAAsF;IACtF,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC7C,iFAAiF;IACjF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,cAAc,CAAC;IACvB,IAAI,CAAC,EAAE,OAAO,CAAC;CACf;AAED,MAAM,WAAW,cAAc;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE;QACV,IAAI,CAAC,EAAE,SAAS,CAAC;QACjB,OAAO,CAAC,EAAE,cAAc,CAAC;QACzB,wEAAwE;QACxE,MAAM,CAAC,EAAE,mBAAmB,CAAC;QAC7B,+CAA+C;QAC/C,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;KAC7C,CAAC;IACF,SAAS,EAAE,aAAa,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,gBAAgB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,uEAAuE;IACvE,QAAQ,EAAE,eAAe,CAAC;IAC1B,uDAAuD;IACvD,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,eAAe,GACxB,aAAa,GACb,aAAa,GACb,eAAe,GACf,eAAe,GACf,iBAAiB,CAAC;AAErB,MAAM,WAAW,kBAAkB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC/C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAC7B,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,6DAA6D;IAC7D,aAAa,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACrC,+EAA+E;IAC/E,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,0EAA0E;IAC1E,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,cAAc;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,SAAS,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,cAAc,EAAE,CAAC;CAC1B"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@post-print/agent-test",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.15",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"bin": {
|
|
6
6
|
"agent-test": "./dist/cli.js"
|
|
@@ -25,7 +25,7 @@
|
|
|
25
25
|
"clean": "rm -rf dist tsconfig.tsbuildinfo"
|
|
26
26
|
},
|
|
27
27
|
"dependencies": {
|
|
28
|
-
"@post-print/agent-harness": "0.1.
|
|
28
|
+
"@post-print/agent-harness": "0.1.15",
|
|
29
29
|
"chalk": "5"
|
|
30
30
|
},
|
|
31
31
|
"devDependencies": {
|