@tangle-network/agent-eval 0.200.1 → 0.201.0
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.
- package/CHANGELOG.md +4 -0
- package/dist/adapters/http.d.ts +7 -8
- package/dist/adapters/http.d.ts.map +1 -1
- package/dist/adapters/http.js +2 -2
- package/dist/adapters/http.js.map +1 -1
- package/dist/agent-profile-DDfIjdyS.d.ts.map +1 -1
- package/dist/analyst/index.d.ts +76 -76
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +6 -6
- package/dist/{analyze-runs-DFbp2gB8.js → analyze-runs-BrIibNd0.js} +47 -42
- package/dist/{analyze-runs-DFbp2gB8.js.map → analyze-runs-BrIibNd0.js.map} +1 -1
- package/dist/{ast-CP9ae9B0.js → ast-DJ6Bc7ab.js} +5 -4
- package/dist/{ast-CP9ae9B0.js.map → ast-DJ6Bc7ab.js.map} +1 -1
- package/dist/authenticity/index.d.ts +14 -15
- package/dist/authenticity/index.d.ts.map +1 -1
- package/dist/{baseline-BC-eBZ7U.js → baseline-CVKsfWTg.js} +2 -2
- package/dist/{baseline-BC-eBZ7U.js.map → baseline-CVKsfWTg.js.map} +1 -1
- package/dist/{benchmark-BeQJyrRC.js → benchmark-BywPeAUo.js} +4 -3
- package/dist/{benchmark-BeQJyrRC.js.map → benchmark-BywPeAUo.js.map} +1 -1
- package/dist/{benchmark-command-EBmuMUBo.js → benchmark-command-DFs_i62N.js} +24 -23
- package/dist/{benchmark-command-EBmuMUBo.js.map → benchmark-command-DFs_i62N.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +12 -12
- package/dist/benchmarks/index.d.ts.map +1 -1
- package/dist/benchmarks/index.js +3 -3
- package/dist/{bounded-process-BBZob7vl.js → bounded-process-Cj9YjXbE.js} +3 -3
- package/dist/{bounded-process-BBZob7vl.js.map → bounded-process-Cj9YjXbE.js.map} +1 -1
- package/dist/builder-eval/index.d.ts +16 -17
- package/dist/builder-eval/index.d.ts.map +1 -1
- package/dist/builder-eval/index.js +12 -10
- package/dist/builder-eval/index.js.map +1 -1
- package/dist/campaign/index.js +11 -11
- package/dist/{campaign-L391bmir.js → campaign-PmTWZi72.js} +30 -25
- package/dist/{campaign-L391bmir.js.map → campaign-PmTWZi72.js.map} +1 -1
- package/dist/{campaign-evidence-jL_U7Ikv.js → campaign-evidence-Dt_rWPk2.js} +4 -4
- package/dist/{campaign-evidence-jL_U7Ikv.js.map → campaign-evidence-Dt_rWPk2.js.map} +1 -1
- package/dist/{chat-client-CJORSG_J.js → chat-client-DUhWNQQS.js} +5 -4
- package/dist/{chat-client-CJORSG_J.js.map → chat-client-DUhWNQQS.js.map} +1 -1
- package/dist/{chat-json-call-DnlDjpco.js → chat-json-call-Dx_sHzqm.js} +2 -2
- package/dist/{chat-json-call-DnlDjpco.js.map → chat-json-call-Dx_sHzqm.js.map} +1 -1
- package/dist/{claim-V9Zqbq0J.js → claim-BXcZw-jo.js} +2 -2
- package/dist/{claim-V9Zqbq0J.js.map → claim-BXcZw-jo.js.map} +1 -1
- package/dist/cli.js +12 -11
- package/dist/cli.js.map +1 -1
- package/dist/contract/index.js +22 -19
- package/dist/contract/index.js.map +1 -1
- package/dist/{default-registry-xMLX160E.js → default-registry-D4qXx2aG.js} +2 -2
- package/dist/{default-registry-xMLX160E.js.map → default-registry-D4qXx2aG.js.map} +1 -1
- package/dist/define-agent-eval-1YuvJpOs.d.ts.map +1 -1
- package/dist/{define-agent-eval-7q_UwsTt.js → define-agent-eval-CJN9dTQw.js} +10 -9
- package/dist/{define-agent-eval-7q_UwsTt.js.map → define-agent-eval-CJN9dTQw.js.map} +1 -1
- package/dist/{descriptive-1V17A-qa.js → descriptive-D9LUFk1-.js} +2 -2
- package/dist/{descriptive-1V17A-qa.js.map → descriptive-D9LUFk1-.js.map} +1 -1
- package/dist/diagnosis/index.d.ts +7 -7
- package/dist/diagnosis/index.d.ts.map +1 -1
- package/dist/diagnosis/index.js +2 -2
- package/dist/diagnosis/index.js.map +1 -1
- package/dist/{dspy-rlm-engine-B5T43BkZ.js → dspy-rlm-engine-US8YsYGq.js} +5 -5
- package/dist/{dspy-rlm-engine-B5T43BkZ.js.map → dspy-rlm-engine-US8YsYGq.js.map} +1 -1
- package/dist/{eval-campaign-GapmPHIA.js → eval-campaign-ha_aWtk7.js} +14 -12
- package/dist/{eval-campaign-GapmPHIA.js.map → eval-campaign-ha_aWtk7.js.map} +1 -1
- package/dist/evaluation.d.ts +9 -10
- package/dist/evaluation.d.ts.map +1 -1
- package/dist/experiment/index.d.ts +29 -29
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +13 -12
- package/dist/experiment/index.js.map +1 -1
- package/dist/{exporters-DEcXbvgO.js → exporters-Coiv-qO2.js} +2 -2
- package/dist/{exporters-DEcXbvgO.js.map → exporters-Coiv-qO2.js.map} +1 -1
- package/dist/{external-optimizer-process-CEXcCKqd.js → external-optimizer-process-KLcw0Vnc.js} +2 -2
- package/dist/{external-optimizer-process-CEXcCKqd.js.map → external-optimizer-process-KLcw0Vnc.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-laM1-UDV.js → external-optimizer-subprocess-CkyUUV7h.js} +4 -4
- package/dist/{external-optimizer-subprocess-laM1-UDV.js.map → external-optimizer-subprocess-CkyUUV7h.js.map} +1 -1
- package/dist/fuzz.d.ts +15 -15
- package/dist/fuzz.d.ts.map +1 -1
- package/dist/{hf-dataset-CRqtSjnR.js → hf-dataset-kjIfIK_e.js} +3 -3
- package/dist/{hf-dataset-CRqtSjnR.js.map → hf-dataset-kjIfIK_e.js.map} +1 -1
- package/dist/hosted/index.d.ts +8 -8
- package/dist/hosted/index.d.ts.map +1 -1
- package/dist/hosted/index.js +1 -1
- package/dist/{index-D4pC7Tvd2.d.ts → index-D4pC7Tvd.d.ts} +1 -1
- package/dist/index-D4pC7Tvd.d.ts.map +1 -0
- package/dist/index-DKmrHA02.d.ts.map +1 -1
- package/dist/{index-DNgf5gyG2.d.ts → index-DNgf5gyG.d.ts} +1 -1
- package/dist/index-DNgf5gyG.d.ts.map +1 -0
- package/dist/index.d.ts +124 -124
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +52 -42
- package/dist/index.js.map +1 -1
- package/dist/{integrity-BFP4kjGj.js → integrity-BvnVARVl.js} +30 -29
- package/dist/{integrity-BFP4kjGj.js.map → integrity-BvnVARVl.js.map} +1 -1
- package/dist/{integrity-BvHANvnp.js → integrity-uwK7XC-8.js} +2 -2
- package/dist/{integrity-BvHANvnp.js.map → integrity-uwK7XC-8.js.map} +1 -1
- package/dist/{internal-BMFSR8Ns.js → internal-BOe3zmjQ.js} +2 -2
- package/dist/{internal-BMFSR8Ns.js.map → internal-BOe3zmjQ.js.map} +1 -1
- package/dist/{jev-review-Cu9xtIu4.js → jev-review-DZEjDzlF.js} +7 -5
- package/dist/{jev-review-Cu9xtIu4.js.map → jev-review-DZEjDzlF.js.map} +1 -1
- package/dist/jev.d.ts +10 -10
- package/dist/jev.d.ts.map +1 -1
- package/dist/jev.js +2 -2
- package/dist/{judge-calibration-DYmaBtJr.js → judge-calibration-C5zgtiJT.js} +2 -2
- package/dist/{judge-calibration-DYmaBtJr.js.map → judge-calibration-C5zgtiJT.js.map} +1 -1
- package/dist/kind-factory-CdPPHtGZ.js.map +1 -1
- package/dist/{llm-client-c_bKHTXz.js → llm-client-Bhqoglar.js} +6 -4
- package/dist/{llm-client-c_bKHTXz.js.map → llm-client-Bhqoglar.js.map} +1 -1
- package/dist/{llm-judge-CL6f3utD.js → llm-judge-C-0P1WiT.js} +26 -21
- package/dist/{llm-judge-CL6f3utD.js.map → llm-judge-C-0P1WiT.js.map} +1 -1
- package/dist/matrix/index.d.ts +1 -1
- package/dist/matrix/index.js +1 -1
- package/dist/{matrix-_NLSWsKM.d.ts → matrix--IZKROxD.d.ts} +2 -2
- package/dist/{matrix-_NLSWsKM.d.ts.map → matrix--IZKROxD.d.ts.map} +1 -1
- package/dist/{matrix-pWinRMPl.js → matrix-DnlHn5x1.js} +8 -6
- package/dist/{matrix-pWinRMPl.js.map → matrix-DnlHn5x1.js.map} +1 -1
- package/dist/meta-eval/index.d.ts +55 -55
- package/dist/meta-eval/index.d.ts.map +1 -1
- package/dist/meta-eval/index.js +4 -4
- package/dist/{mint-iCD-W43s.js → mint-vlRckJEX.js} +2 -2
- package/dist/{mint-iCD-W43s.js.map → mint-vlRckJEX.js.map} +1 -1
- package/dist/multishot/golden/index.d.ts +23 -23
- package/dist/multishot/golden/index.d.ts.map +1 -1
- package/dist/multishot/golden/index.js +1 -1
- package/dist/multishot/index.d.ts +9 -9
- package/dist/multishot/index.d.ts.map +1 -1
- package/dist/multishot/index.js +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{otel-spans-CQmSxtcS.js → otel-spans-C8YIJWy1.js} +9 -4
- package/dist/{otel-spans-CQmSxtcS.js.map → otel-spans-C8YIJWy1.js.map} +1 -1
- package/dist/{outcome-store-BXlkwMPR.js → outcome-store-DUcMj-3D.js} +2 -2
- package/dist/{outcome-store-BXlkwMPR.js.map → outcome-store-DUcMj-3D.js.map} +1 -1
- package/dist/package-version-D7lQHt_-.js.map +1 -1
- package/dist/{paired-arms-DKZFVPt8.js → paired-arms-DpS_h8ON.js} +8 -5
- package/dist/{paired-arms-DKZFVPt8.js.map → paired-arms-DpS_h8ON.js.map} +1 -1
- package/dist/{paired-binary-fzkwENI4.js → paired-binary-OMBVRtc-.js} +2 -2
- package/dist/{paired-binary-fzkwENI4.js.map → paired-binary-OMBVRtc-.js.map} +1 -1
- package/dist/{paired-promotion-decision-D5B1Cxc2.js → paired-promotion-decision-DhrmjdV8.js} +5 -4
- package/dist/{paired-promotion-decision-D5B1Cxc2.js.map → paired-promotion-decision-DhrmjdV8.js.map} +1 -1
- package/dist/{paired-tests-C8iCsioC.js → paired-tests-Ccr16gcW.js} +3 -3
- package/dist/{paired-tests-C8iCsioC.js.map → paired-tests-Ccr16gcW.js.map} +1 -1
- package/dist/pipelines/index.d.ts +19 -19
- package/dist/pipelines/index.d.ts.map +1 -1
- package/dist/pipelines/index.js +5 -4
- package/dist/pipelines/index.js.map +1 -1
- package/dist/{power-and-mde-K_CmO3dc.js → power-and-mde-bie6i965.js} +9 -6
- package/dist/{power-and-mde-K_CmO3dc.js.map → power-and-mde-bie6i965.js.map} +1 -1
- package/dist/{produced-state-DZMW8cuF.js → produced-state-CR2tAPLf.js} +8 -5
- package/dist/{produced-state-DZMW8cuF.js.map → produced-state-CR2tAPLf.js.map} +1 -1
- package/dist/{promotion-policy-C6FAJvmZ.js → promotion-policy-BI1RxzwD.js} +4 -4
- package/dist/{promotion-policy-C6FAJvmZ.js.map → promotion-policy-BI1RxzwD.js.map} +1 -1
- package/dist/{raw-provider-sink-CXuOhLvT.js → raw-provider-sink-rB2DYxr6.js} +2 -2
- package/dist/{raw-provider-sink-CXuOhLvT.js.map → raw-provider-sink-rB2DYxr6.js.map} +1 -1
- package/dist/{release-confidence-XQm_xjhR.js → release-confidence-CiFg1EPq.js} +3 -2
- package/dist/{release-confidence-XQm_xjhR.js.map → release-confidence-CiFg1EPq.js.map} +1 -1
- package/dist/{report-command-DKJ_yAIj.js → report-command-DzPpFgjB.js} +20 -18
- package/dist/{report-command-DKJ_yAIj.js.map → report-command-DzPpFgjB.js.map} +1 -1
- package/dist/reporting.js +5 -5
- package/dist/{reward-hacking-D0XwhVWE.js → reward-hacking-C-a__ShZ.js} +5 -3
- package/dist/{reward-hacking-D0XwhVWE.js.map → reward-hacking-C-a__ShZ.js.map} +1 -1
- package/dist/rl.d.ts +186 -186
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +21 -17
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +1 -1
- package/dist/rollout/index.js +5 -5
- package/dist/{rollout-B-XBReTC.js → rollout-BUQixCoc.js} +5 -5
- package/dist/{rollout-B-XBReTC.js.map → rollout-BUQixCoc.js.map} +1 -1
- package/dist/{rubric-predictive-validity-CCK-1B7w.js → rubric-predictive-validity-B0Pg-3I5.js} +3 -3
- package/dist/{rubric-predictive-validity-CCK-1B7w.js.map → rubric-predictive-validity-B0Pg-3I5.js.map} +1 -1
- package/dist/{run-record-tNnlQIrD.js → run-record-BB290Y9A.js} +5 -3
- package/dist/{run-record-tNnlQIrD.js.map → run-record-BB290Y9A.js.map} +1 -1
- package/dist/{schema-C1aaAxTf.js → schema-DAMp77zB.js} +24 -20
- package/dist/{schema-C1aaAxTf.js.map → schema-DAMp77zB.js.map} +1 -1
- package/dist/search/index.d.ts +34 -34
- package/dist/search/index.d.ts.map +1 -1
- package/dist/search/index.js +2 -2
- package/dist/{search-kernel-AVgIjXaI.js → search-kernel-Dpniloks.js} +33 -26
- package/dist/{search-kernel-AVgIjXaI.js.map → search-kernel-Dpniloks.js.map} +1 -1
- package/dist/{search-ledger-CL0-mXQ8.js → search-ledger-CXQPwdWz.js} +4 -3
- package/dist/{search-ledger-CL0-mXQ8.js.map → search-ledger-CXQPwdWz.js.map} +1 -1
- package/dist/{search-shipper-Cs6K4km9.js → search-shipper-DhhofkDX.js} +3 -6
- package/dist/{search-shipper-Cs6K4km9.js.map → search-shipper-DhhofkDX.js.map} +1 -1
- package/dist/{search-summary-DvWk0hb3.js → search-summary-C1nAHgkB.js} +4 -4
- package/dist/{search-summary-DvWk0hb3.js.map → search-summary-C1nAHgkB.js.map} +1 -1
- package/dist/{semantic-concept-judge-2SE_kXGF.js → semantic-concept-judge-BcjYg37z.js} +4 -4
- package/dist/{semantic-concept-judge-2SE_kXGF.js.map → semantic-concept-judge-BcjYg37z.js.map} +1 -1
- package/dist/{sequential-Yz3OSskl.js → sequential-DWn_np6P.js} +2 -3
- package/dist/{sequential-Yz3OSskl.js.map → sequential-DWn_np6P.js.map} +1 -1
- package/dist/{server-D56r1lhp.js → server-B4jMhqQn.js} +2 -2
- package/dist/{server-D56r1lhp.js.map → server-B4jMhqQn.js.map} +1 -1
- package/dist/{skillopt-optimization-method-B2syNik0.js → skillopt-optimization-method-Bls9t-G_.js} +9 -9
- package/dist/{skillopt-optimization-method-B2syNik0.js.map → skillopt-optimization-method-Bls9t-G_.js.map} +1 -1
- package/dist/{statistical-heldout-BAFDCo07.js → statistical-heldout-CNLOvRq_.js} +3 -3
- package/dist/{statistical-heldout-BAFDCo07.js.map → statistical-heldout-CNLOvRq_.js.map} +1 -1
- package/dist/statistical-heldout-HswMvzbP.d.ts.map +1 -1
- package/dist/{store-DNe_Uv1Q.js → store-BhNuLdJw.js} +3 -5
- package/dist/{store-DNe_Uv1Q.js.map → store-BhNuLdJw.js.map} +1 -1
- package/dist/store-C4avFjIi.d.ts.map +1 -1
- package/dist/{store-otlp-BPowgPm6.js → store-otlp-yd9lDziN.js} +2 -2
- package/dist/{store-otlp-BPowgPm6.js.map → store-otlp-yd9lDziN.js.map} +1 -1
- package/dist/{store-tool-spans-BaGakFc2.js → store-tool-spans-Biio4_Nw.js} +4 -4
- package/dist/{store-tool-spans-BaGakFc2.js.map → store-tool-spans-Biio4_Nw.js.map} +1 -1
- package/dist/storyboard/index.d.ts +20 -21
- package/dist/storyboard/index.d.ts.map +1 -1
- package/dist/{student-t-BA-Uy51p.js → student-t-BBZYka0E.js} +4 -3
- package/dist/{student-t-BA-Uy51p.js.map → student-t-BBZYka0E.js.map} +1 -1
- package/dist/{summary-report-BeYnvddy.js → summary-report-DLzptLSA.js} +8 -7
- package/dist/{summary-report-BeYnvddy.js.map → summary-report-DLzptLSA.js.map} +1 -1
- package/dist/supervisor-run/index.d.ts +25 -25
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +2 -2
- package/dist/{tool-waste-CJDfxH45.js → tool-waste-B-u3rzCz.js} +11 -6
- package/dist/{tool-waste-CJDfxH45.js.map → tool-waste-B-u3rzCz.js.map} +1 -1
- package/dist/trace-repair/index.d.ts +96 -96
- package/dist/trace-repair/index.d.ts.map +1 -1
- package/dist/trace-repair/index.js +1 -1
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/traces.d.ts +29 -29
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +9 -7
- package/dist/traces.js.map +1 -1
- package/dist/trajectory-replay/index.d.ts +35 -35
- package/dist/trajectory-replay/index.d.ts.map +1 -1
- package/dist/trajectory-replay/index.js +5 -5
- package/dist/trajectory-replay/index.js.map +1 -1
- package/dist/{tree-Ki_gW7z8.js → tree-D48IJUK0.js} +39 -31
- package/dist/{tree-Ki_gW7z8.js.map → tree-D48IJUK0.js.map} +1 -1
- package/dist/wire/index.d.ts +48 -48
- package/dist/wire/index.d.ts.map +1 -1
- package/dist/wire/index.js +1 -1
- package/package.json +18 -36
- package/dist/index-D4pC7Tvd2.d.ts.map +0 -1
- package/dist/index-DNgf5gyG2.d.ts.map +0 -1
|
@@ -10,7 +10,7 @@ interface CorpusSpec {
|
|
|
10
10
|
readonly preparedDir: string;
|
|
11
11
|
}
|
|
12
12
|
/** Parses `name=<labelsPath>::<preparedDir>` (paths may contain `=`, not `::`). */
|
|
13
|
-
declare function parseCorpusFlag(value: string): CorpusSpec;
|
|
13
|
+
export declare function parseCorpusFlag(value: string): CorpusSpec;
|
|
14
14
|
type ReplayExclusionReason = 'no-swe-raw-trajectory' | 'ambiguous-swe-raw-trajectory' | 'unreadable-raw-trajectory' | 'no-docker-image' | 'no-gold-incorrect-step' | 'gold-only-submit-step' | 'missing-steps-json' | 'gold-step-outside-steps' | 'cwd-underivable' | 'no-recorded-returncode-at-k';
|
|
15
15
|
interface ExcludedCase {
|
|
16
16
|
readonly corpus: string;
|
|
@@ -50,8 +50,8 @@ interface LabelEntry {
|
|
|
50
50
|
readonly incorrect_step_ids?: readonly number[];
|
|
51
51
|
}[];
|
|
52
52
|
}
|
|
53
|
-
declare function readLabelEntries(labelsPath: string): LabelEntry[];
|
|
54
|
-
declare function goldIncorrectSteps(entry: LabelEntry): number[];
|
|
53
|
+
export declare function readLabelEntries(labelsPath: string): LabelEntry[];
|
|
54
|
+
export declare function goldIncorrectSteps(entry: LabelEntry): number[];
|
|
55
55
|
type ResourceResolution = {
|
|
56
56
|
readonly resolved: true;
|
|
57
57
|
readonly resources: CaseResources;
|
|
@@ -64,7 +64,7 @@ type ResourceResolution = {
|
|
|
64
64
|
* Resolves the replay resources for one trajectory, independent of gold
|
|
65
65
|
* labels — the finding wire uses this with a finding-supplied step instead.
|
|
66
66
|
*/
|
|
67
|
-
declare function resolveCaseResources(corpus: CorpusSpec, trajId: string): ResourceResolution;
|
|
67
|
+
export declare function resolveCaseResources(corpus: CorpusSpec, trajId: string): ResourceResolution;
|
|
68
68
|
interface EnumerationResult {
|
|
69
69
|
readonly replayable: ReplayableCase[];
|
|
70
70
|
readonly excluded: ExcludedCase[];
|
|
@@ -78,7 +78,7 @@ interface EnumerationResult {
|
|
|
78
78
|
* Exclusion reasons are reported in resolution order:
|
|
79
79
|
* raw trajectory → image → gold labels → steps.json → k range → cwd.
|
|
80
80
|
*/
|
|
81
|
-
declare function enumerateReplayableCases(corpora: readonly CorpusSpec[]): EnumerationResult;
|
|
81
|
+
export declare function enumerateReplayableCases(corpora: readonly CorpusSpec[]): EnumerationResult;
|
|
82
82
|
//#endregion
|
|
83
83
|
//#region src/trajectory-replay/exec.d.ts
|
|
84
84
|
/**
|
|
@@ -113,7 +113,7 @@ type ReplayExecBackendFactory = (image: string) => ReplayExecBackend;
|
|
|
113
113
|
* recorded actions — by piping the base64 of the action into `sh` after
|
|
114
114
|
* cd-ing to the workdir. Exit code is sh's, i.e. the action's.
|
|
115
115
|
*/
|
|
116
|
-
declare function wrapActionForExec(action: string, cwd: string): string;
|
|
116
|
+
export declare function wrapActionForExec(action: string, cwd: string): string;
|
|
117
117
|
//#endregion
|
|
118
118
|
//#region src/trajectory-replay/fix.d.ts
|
|
119
119
|
interface ChatUsage {
|
|
@@ -142,8 +142,8 @@ interface FixPromptInput {
|
|
|
142
142
|
readonly contextRadius?: number;
|
|
143
143
|
}
|
|
144
144
|
/** Head+tail excerpt with an elision marker; identity below the limit. */
|
|
145
|
-
declare function clipText(text: string, limit: number): string;
|
|
146
|
-
declare function buildFixPrompt(input: FixPromptInput): {
|
|
145
|
+
export declare function clipText(text: string, limit: number): string;
|
|
146
|
+
export declare function buildFixPrompt(input: FixPromptInput): {
|
|
147
147
|
system: string;
|
|
148
148
|
user: string;
|
|
149
149
|
};
|
|
@@ -162,14 +162,14 @@ interface FailedFixAttempt {
|
|
|
162
162
|
* attempt with its REAL executed output, and permission to answer with a short
|
|
163
163
|
* script (the block still executes as one /bin/sh unit).
|
|
164
164
|
*/
|
|
165
|
-
declare function buildRetryFixPrompt(input: FixPromptInput, priorAttempts: readonly FailedFixAttempt[], maxScriptCommands?: number): {
|
|
165
|
+
export declare function buildRetryFixPrompt(input: FixPromptInput, priorAttempts: readonly FailedFixAttempt[], maxScriptCommands?: number): {
|
|
166
166
|
system: string;
|
|
167
167
|
user: string;
|
|
168
168
|
};
|
|
169
169
|
/** Non-empty, non-comment lines of a fix script — the loop's script-size cap. */
|
|
170
|
-
declare function countScriptCommands(script: string): number;
|
|
170
|
+
export declare function countScriptCommands(script: string): number;
|
|
171
171
|
/** Last fenced code block, else the whole trimmed content; null when empty. */
|
|
172
|
-
declare function extractFixCommand(content: string): string | null;
|
|
172
|
+
export declare function extractFixCommand(content: string): string | null;
|
|
173
173
|
type FixGenerationOutcome = {
|
|
174
174
|
readonly succeeded: true;
|
|
175
175
|
readonly value: {
|
|
@@ -180,7 +180,7 @@ type FixGenerationOutcome = {
|
|
|
180
180
|
readonly succeeded: false;
|
|
181
181
|
readonly error: string;
|
|
182
182
|
};
|
|
183
|
-
declare function generateFixCommand(caller: ChatCompletionCaller, input: FixPromptInput): Promise<FixGenerationOutcome>;
|
|
183
|
+
export declare function generateFixCommand(caller: ChatCompletionCaller, input: FixPromptInput): Promise<FixGenerationOutcome>;
|
|
184
184
|
//#endregion
|
|
185
185
|
//#region src/trajectory-replay/fix-loop.d.ts
|
|
186
186
|
/** Result of executing one corrected command as a full arm. */
|
|
@@ -241,7 +241,7 @@ interface FixLoopResult {
|
|
|
241
241
|
* bound rather than a measurement. */
|
|
242
242
|
readonly callsWithoutUsage: number;
|
|
243
243
|
}
|
|
244
|
-
declare function runFixLoop(caller: ChatCompletionCaller, input: FixPromptInput, executor: FixArmExecutor, options: FixLoopOptions): Promise<FixLoopResult>;
|
|
244
|
+
export declare function runFixLoop(caller: ChatCompletionCaller, input: FixPromptInput, executor: FixArmExecutor, options: FixLoopOptions): Promise<FixLoopResult>;
|
|
245
245
|
//#endregion
|
|
246
246
|
//#region src/trajectory-replay/image-preparer.d.ts
|
|
247
247
|
/**
|
|
@@ -271,14 +271,14 @@ interface DockerImagePreparerOptions {
|
|
|
271
271
|
readonly pullTimeoutMs?: number;
|
|
272
272
|
readonly buildTimeoutMs?: number;
|
|
273
273
|
}
|
|
274
|
-
declare function derivedImageTag(image: string, cwd: string): string;
|
|
274
|
+
export declare function derivedImageTag(image: string, cwd: string): string;
|
|
275
275
|
/**
|
|
276
276
|
* Pulls the base image when absent and builds `FROM <base>; RUN chown -R
|
|
277
277
|
* 1000:1000 <cwd>` tagged by content hash, so repeated batches reuse both
|
|
278
278
|
* the pull and the build. cwd `/` skips the chown (never chown -R /) and
|
|
279
279
|
* replays on the base image directly.
|
|
280
280
|
*/
|
|
281
|
-
declare function dockerImagePreparer(options?: DockerImagePreparerOptions): ImagePreparer;
|
|
281
|
+
export declare function dockerImagePreparer(options?: DockerImagePreparerOptions): ImagePreparer;
|
|
282
282
|
//#endregion
|
|
283
283
|
//#region src/trajectory-replay/batch.d.ts
|
|
284
284
|
interface ReplayBatchOptions {
|
|
@@ -455,9 +455,9 @@ interface ReplayBatchReport {
|
|
|
455
455
|
}[];
|
|
456
456
|
readonly cases: readonly ReplayBatchCaseRow[];
|
|
457
457
|
}
|
|
458
|
-
declare function seededSample<T>(items: readonly T[], size: number, seed: number): Set<T>;
|
|
459
|
-
declare function runReplayBatch(options: ReplayBatchOptions): Promise<ReplayBatchReport>;
|
|
460
|
-
declare function renderBatchReport(report: ReplayBatchReport): string;
|
|
458
|
+
export declare function seededSample<T>(items: readonly T[], size: number, seed: number): Set<T>;
|
|
459
|
+
export declare function runReplayBatch(options: ReplayBatchOptions): Promise<ReplayBatchReport>;
|
|
460
|
+
export declare function renderBatchReport(report: ReplayBatchReport): string;
|
|
461
461
|
//#endregion
|
|
462
462
|
//#region src/trajectory-replay/verify.d.ts
|
|
463
463
|
interface IngestedTrajectory {
|
|
@@ -469,7 +469,7 @@ interface IngestedTrajectory {
|
|
|
469
469
|
* injected clock so `buildTrajectory` ordering is deterministic even when
|
|
470
470
|
* two spans would share a Date.now() millisecond.
|
|
471
471
|
*/
|
|
472
|
-
declare function ingestRecordedTrajectory(store: TraceStore, steps: readonly RecordedTrajectoryStep[], caseId: string): Promise<IngestedTrajectory>;
|
|
472
|
+
export declare function ingestRecordedTrajectory(store: TraceStore, steps: readonly RecordedTrajectoryStep[], caseId: string): Promise<IngestedTrajectory>;
|
|
473
473
|
/** Why a replayed prefix step failed to confirm the recording. */
|
|
474
474
|
type PrefixDivergenceKind =
|
|
475
475
|
/** The recording carries a returncode and the replayed exit differs from it. */
|
|
@@ -500,14 +500,14 @@ interface PrefixReplayResult {
|
|
|
500
500
|
}
|
|
501
501
|
/** Admission tolerance: a prefix replay is faithful enough to build a verdict
|
|
502
502
|
* on when at most this percentage of its executed steps diverged. */
|
|
503
|
-
declare const PREFIX_DIVERGENCE_TOLERANCE_PCT = 10;
|
|
503
|
+
export declare const PREFIX_DIVERGENCE_TOLERANCE_PCT = 10;
|
|
504
504
|
/**
|
|
505
505
|
* Compare one replayed prefix step against its recording. Returns null only
|
|
506
506
|
* when the recording positively confirms the replay.
|
|
507
507
|
*/
|
|
508
|
-
declare function classifyPrefixStep(step: number, expectedReturncode: number | null, actualExit: number): PrefixDivergence | null;
|
|
508
|
+
export declare function classifyPrefixStep(step: number, expectedReturncode: number | null, actualExit: number): PrefixDivergence | null;
|
|
509
509
|
/** Roll per-step classifications up into the rate the admission pre-pass gates on. */
|
|
510
|
-
declare function summarizePrefixReplay(prefixExecuted: number, prefixDivergences: PrefixDivergence[], wallMs: number): PrefixReplayResult;
|
|
510
|
+
export declare function summarizePrefixReplay(prefixExecuted: number, prefixDivergences: PrefixDivergence[], wallMs: number): PrefixReplayResult;
|
|
511
511
|
interface ArmExecutionResult {
|
|
512
512
|
command: string;
|
|
513
513
|
exitCode: number;
|
|
@@ -529,7 +529,7 @@ interface SandboxCounterfactualRunnerOptions {
|
|
|
529
529
|
* `lastPrefix` / `lastArm` for the caller; spans for every exec land in the
|
|
530
530
|
* counterfactual meta-run.
|
|
531
531
|
*/
|
|
532
|
-
declare class SandboxCounterfactualRunner implements CounterfactualRunner {
|
|
532
|
+
export declare class SandboxCounterfactualRunner implements CounterfactualRunner {
|
|
533
533
|
private readonly backend;
|
|
534
534
|
private readonly options;
|
|
535
535
|
lastPrefix: PrefixReplayResult | null;
|
|
@@ -603,7 +603,7 @@ interface ReplayVerdict extends DefaultVerdict {
|
|
|
603
603
|
armB: string | null;
|
|
604
604
|
};
|
|
605
605
|
}
|
|
606
|
-
declare function replayVerify(options: ReplayVerifyOptions): Promise<ReplayVerdict>;
|
|
606
|
+
export declare function replayVerify(options: ReplayVerifyOptions): Promise<ReplayVerdict>;
|
|
607
607
|
//#endregion
|
|
608
608
|
//#region src/trajectory-replay/findings.d.ts
|
|
609
609
|
interface VerifiableFinding {
|
|
@@ -625,9 +625,9 @@ interface VerifiableFinding {
|
|
|
625
625
|
* block's first incorrect step there even when the subject names a later
|
|
626
626
|
* step of the same block.
|
|
627
627
|
*/
|
|
628
|
-
declare function findingReplayStep(finding: VerifiableFinding): number | null;
|
|
628
|
+
export declare function findingReplayStep(finding: VerifiableFinding): number | null;
|
|
629
629
|
/** Trajectory id from the finding's `trace://<traj>/…` evidence refs, or null. */
|
|
630
|
-
declare function findingTrajectoryId(finding: VerifiableFinding): string | null;
|
|
630
|
+
export declare function findingTrajectoryId(finding: VerifiableFinding): string | null;
|
|
631
631
|
/**
|
|
632
632
|
* Where the executable trajectory lives.
|
|
633
633
|
* `direct` — one trajectory's steps.json plus a replay-ready image; the
|
|
@@ -671,7 +671,7 @@ type FindingReplayability = {
|
|
|
671
671
|
* what invocation. Never throws for a finding-shaped problem — every dead end
|
|
672
672
|
* becomes a `not-replayable` reason the receipt can carry verbatim.
|
|
673
673
|
*/
|
|
674
|
-
declare function resolveFindingReplayability(finding: VerifiableFinding, source: FindingReplaySource): FindingReplayability;
|
|
674
|
+
export declare function resolveFindingReplayability(finding: VerifiableFinding, source: FindingReplaySource): FindingReplayability;
|
|
675
675
|
type FindingVerificationStatus = 'reproduced' | 'fix-flipped' | 'not-replayable' | 'divergent';
|
|
676
676
|
interface FindingVerification {
|
|
677
677
|
readonly finding_id: string | null;
|
|
@@ -722,22 +722,22 @@ interface VerifyFindingsOptions {
|
|
|
722
722
|
* prefix outside the divergence tolerance is divergent no matter what arm A
|
|
723
723
|
* did: the state it ran against is not the recorded state.
|
|
724
724
|
*/
|
|
725
|
-
declare function classifyVerdict(verdict: ReplayVerdict): FindingVerificationStatus;
|
|
725
|
+
export declare function classifyVerdict(verdict: ReplayVerdict): FindingVerificationStatus;
|
|
726
726
|
/**
|
|
727
727
|
* Verifies every finding against the source: resolves replayability, executes
|
|
728
728
|
* one proof per distinct (case, step, fix) — findings accusing the same step
|
|
729
729
|
* share the executed proof — and writes a receipt directory per finding plus a
|
|
730
730
|
* run-level verifications.json.
|
|
731
731
|
*/
|
|
732
|
-
declare function verifyFindings(findings: readonly VerifiableFinding[], options: VerifyFindingsOptions): Promise<VerifyFindingsRun>;
|
|
732
|
+
export declare function verifyFindings(findings: readonly VerifiableFinding[], options: VerifyFindingsOptions): Promise<VerifyFindingsRun>;
|
|
733
733
|
/** Markdown section an analysis report appends when finding verification ran. */
|
|
734
|
-
declare function renderVerifiedFindingsSection(run: VerifyFindingsRun): string;
|
|
734
|
+
export declare function renderVerifiedFindingsSection(run: VerifyFindingsRun): string;
|
|
735
735
|
/**
|
|
736
736
|
* Accepts the two shapes findings travel in: a bare JSON array of analyst
|
|
737
737
|
* findings, or an object with a `findings` array (e.g. an extracted
|
|
738
738
|
* `observations[n]` from a result.json).
|
|
739
739
|
*/
|
|
740
|
-
declare function readFindingsFile(path: string): VerifiableFinding[];
|
|
740
|
+
export declare function readFindingsFile(path: string): VerifiableFinding[];
|
|
741
741
|
//#endregion
|
|
742
742
|
//#region src/trajectory-replay/wire.d.ts
|
|
743
743
|
interface IncorrectStepsSubject {
|
|
@@ -747,7 +747,7 @@ interface IncorrectStepsSubject {
|
|
|
747
747
|
readonly consequenceStep: number;
|
|
748
748
|
}
|
|
749
749
|
/** Null when the subject is not an incorrect-steps finding subject. */
|
|
750
|
-
declare function parseIncorrectStepsSubject(subject: string): IncorrectStepsSubject | null;
|
|
750
|
+
export declare function parseIncorrectStepsSubject(subject: string): IncorrectStepsSubject | null;
|
|
751
751
|
interface AnalystReplayFinding {
|
|
752
752
|
readonly trajId: string;
|
|
753
753
|
readonly subject: string;
|
|
@@ -764,7 +764,7 @@ interface ResolvedReplayInvocation {
|
|
|
764
764
|
* replayed (malformed subject, unknown trajectory, non-replayable case, step
|
|
765
765
|
* out of range) — the caller surfaces that reason instead of a proof.
|
|
766
766
|
*/
|
|
767
|
-
declare function resolveFindingInvocation(finding: AnalystReplayFinding, corpora: readonly CorpusSpec[]): ResolvedReplayInvocation;
|
|
767
|
+
export declare function resolveFindingInvocation(finding: AnalystReplayFinding, corpora: readonly CorpusSpec[]): ResolvedReplayInvocation;
|
|
768
768
|
interface ReplayFindingOptions {
|
|
769
769
|
readonly corpora: readonly CorpusSpec[];
|
|
770
770
|
readonly out: string;
|
|
@@ -788,7 +788,7 @@ interface ReplayFindingResult {
|
|
|
788
788
|
* the backend factory receives it as-is, so a factory backed by infrastructure
|
|
789
789
|
* that needs a derived image must run an `ImagePreparer` first.
|
|
790
790
|
*/
|
|
791
|
-
declare function replayVerifyFinding(finding: AnalystReplayFinding, options: ReplayFindingOptions): Promise<ReplayFindingResult>;
|
|
791
|
+
export declare function replayVerifyFinding(finding: AnalystReplayFinding, options: ReplayFindingOptions): Promise<ReplayFindingResult>;
|
|
792
792
|
//#endregion
|
|
793
|
-
export { type AnalystReplayFinding, type ArmExecutionResult, type CaseResources, type ChatCompletionCaller, type ChatOutcome, type ChatUsage, type CorpusSpec, type CwdSource, type DecodedTrajectory, type DockerImagePreparerOptions, type EnumerationResult, type ExcludedCase, FORMAT_ERROR_OBSERVATION_PREFIX, type FailedFixAttempt, type FindingReplaySource, type FindingReplayability, type FindingVerification, type FindingVerificationStatus, type FixArmExecution, type FixArmExecutor, type FixGenerationOutcome, type FixLoopAttemptRecord, type FixLoopOptions, type FixLoopResult, type FixPromptInput, type ImagePreparation, type ImagePreparer, type IncorrectStepsSubject, type IngestedTrajectory,
|
|
793
|
+
export { type AnalystReplayFinding, type ArmExecutionResult, type CaseResources, type ChatCompletionCaller, type ChatOutcome, type ChatUsage, type CorpusSpec, type CwdSource, type DecodedTrajectory, type DockerImagePreparerOptions, type EnumerationResult, type ExcludedCase, FORMAT_ERROR_OBSERVATION_PREFIX, type FailedFixAttempt, type FindingReplaySource, type FindingReplayability, type FindingVerification, type FindingVerificationStatus, type FixArmExecution, type FixArmExecutor, type FixGenerationOutcome, type FixLoopAttemptRecord, type FixLoopOptions, type FixLoopResult, type FixPromptInput, type ImagePreparation, type ImagePreparer, type IncorrectStepsSubject, type IngestedTrajectory, type PrefixDivergence, type PrefixDivergenceKind, type PrefixReplayResult, type RecordedFinalOutcome, type RecordedObservationKind, type RecordedTrajectoryStep, type RecordedTrajectoryTurn, type ReplayArmVerdict, type ReplayBatchCaseRow, type ReplayBatchFixResult, type ReplayBatchOptions, type ReplayBatchReport, type ReplayExclusionReason, type ReplayExecBackend, type ReplayExecBackendFactory, type ReplayExecResult, type ReplayExecSession, type ReplayFindingOptions, type ReplayFindingResult, type ReplayVerdict, type ReplayVerifyOptions, type ReplayableCase, type ResolvedFindingReplay, type ResolvedReplayInvocation, type ResourceResolution, SUBMIT_ACTION_SIGNATURE, type SandboxCounterfactualRunnerOptions, TIMEOUT_OBSERVATION_MARKER, type VerifiableFinding, type VerifyFindingsOptions, type VerifyFindingsRun, assertReplayableTrajectory, classifyObservation, decodeRecordedTurns, deriveFailureSignature, finalRecordedOutcome, isElidedField, isRecordedTimeout, isSubmitAction, isSubmitOnlyAction, parseObservationOutput, parseRecordedReturncode, unreadableExitCount };
|
|
794
794
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/trajectory-replay/corpus.ts","../../src/trajectory-replay/exec.ts","../../src/trajectory-replay/fix.ts","../../src/trajectory-replay/fix-loop.ts","../../src/trajectory-replay/image-preparer.ts","../../src/trajectory-replay/batch.ts","../../src/trajectory-replay/verify.ts","../../src/trajectory-replay/findings.ts","../../src/trajectory-replay/wire.ts"],"mappings":";;;;;;UAyBiB;WACN;WACA;WACA;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/trajectory-replay/corpus.ts","../../src/trajectory-replay/exec.ts","../../src/trajectory-replay/fix.ts","../../src/trajectory-replay/fix-loop.ts","../../src/trajectory-replay/image-preparer.ts","../../src/trajectory-replay/batch.ts","../../src/trajectory-replay/verify.ts","../../src/trajectory-replay/findings.ts","../../src/trajectory-replay/wire.ts"],"mappings":";;;;;;UAyBiB;WACN;WACA;WACA;;;wBAIK,gBAAgB,gBAAgB;KAYpC;UAYK;WACN;WACA;WACA,QAAQ;WACR;;KAGC;;UAGK;WACN;WACA;WACA;WACA,gBAAgB;WAChB;WACA;WACA;WACA,WAAW;;WAEX;;UAGM,uBAAuB;;WAE7B;;;;WAIA;;WAEA;;WAEA;;UAGD;WACC;WACA;aAAuC;;;wBAGlC,iBAAiB,qBAAqB;wBAWtC,mBAAmB,OAAO;KAuD9B;WACG;WAAyB,WAAW;;WACpC;WAA0B,QAAQ;WAAgC;;;;;;wBAMjE,qBAAqB,QAAQ,YAAY,iBAAiB;UAiEzD;WACN,YAAY;WACZ,UAAU;WACV;;;;;;;;;;wBAWK,yBAAyB,kBAAkB,eAAe;;;;;;;;;;;;UChPzD;EACf;EACA;EACA;;UAGe;EACf,KAAK,iBAAiB,oBAAoB,QAAQ;EAClD,SAAS;;UAGM;;EAEf,QAAQ,QAAQ;;;;;KAMN,4BAA4B,kBAAkB;;;;;;;wBAQ1C,kBAAkB,gBAAgB;;;UCrBjC;WACN;WACA;;KAGC;WACG;WAA0B;IAAS;IAAiB,OAAO;;;WAC3D;WAA2B;;UAEzB;EACf,SAAS,gBAAgB,eAAe,QAAQ;;UAGjC;WACN;WACA,gBAAgB;;WAEhB;;WAEA;;;wBAIK,SAAS,cAAc;wBAuCvB,eAAe,OAAO;EAAmB;EAAgB;;;UAiBxD;WACN;;WAEA;WACA;WACA;WACA;WACA;;;;;;;wBA4BK,oBACd,OAAO,gBACP,wBAAwB,oBACxB;EACG;EAAgB;;;wBAwBL,oBAAoB;;wBAMpB,kBAAkB;KAQtB;WACG;WAA0B;IAAS;IAAiB,OAAO;;;WAC3D;WAA2B;;wBAEpB,mBACpB,QAAQ,sBACR,OAAO,iBACN,QAAQ;;;;UCvJM;WACN;WACA;WACA;;WAEA;WACA;WACA;WACA;;;KAIC,kBAAkB,iBAAiB,oBAAoB,QAAQ;UAE1D;;WAEN;;WAEA;;WAEA;WACA,cAAc;;UAGR;WACN;WACA;WACA;WACA;aAAkB;aAA+B;;WACjD;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;UAGM;WACN;WACA;WACA,mBAAmB;;WAEnB;WACA;;;WAGA;WACA;WACA;;;;WAIA;;wBAcW,WACpB,QAAQ,sBACR,OAAO,gBACP,UAAU,gBACV,SAAS,iBACR,QAAQ;;;;;;;;;;;;KCpFC;WAEG;WACA;IAAS;IAAsB;IAAiB;;;WAEhD;WAA2B;;UAEzB;EACf,OAAO,eAAe,cAAc,QAAQ;;UAG7B;WACN;WACA;;wBAGK,gBAAgB,eAAe;;;;;;;wBAW/B,oBAAoB,UAAS,6BAAkC;;;UCQ9D;WACN,kBAAkB;WAClB;;;;WAIA;;WAEA;WACA,YAAY;WACZ;;WAEA;WACA;;WAEA;WACA;;WAEA;;WAEA;;WAEA,WAAW;;WAEX,gBAAgB;WAChB,cAAc;;UAGR;;WAEN;WACA;WACA;WACA;;;WAGA,OAAO;WACP;WACA;WACA;WACA;WACA;WACA;;WAEA,mBAAmB;;;WAGnB;;UAGM;WACN;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAEA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAEA;WACA;;;WAGA;WACA;WACA;WACA;;;WAGA;WACA,KAAK;WACL;;UAGM;WACN;WACA;IAAoB;IAAc;IAAoB;;WACtD;aACE;aACA;aACA;aACA,kBAAkB;;;aAGlB,qBAAqB;MAE1B;MAAyB;;;WAGtB;aACE;MAAqB;MAAmB;MAAqB;;aAC7D;MAAuB;MAAmB;MAAqB;;;;;aAI/D;eACE;eACA;eACA;eACA;;eAEA;eACA;eACA;eACA;;aAEF;MAAe;MAAmB;MAAqB;;;aAEvD;MACP;MACA;MACA;;;;aAIO;MACP;MACA;MACA;;;aAGO,gBAAgB;;WAElB;aACE;aACA;aACA;aACA;aACA;;;;aAIA;;WAEF;IAAqB;IAAgB;IAAgB;IAAgB;;WACrE;IAAyB;IAAgB;IAAgB;IAAe;;WACxE,gBAAgB;;wBAGX,aAAa,GAAG,gBAAgB,KAAK,cAAc,eAAe,IAAI;wBAkFhE,eAAe,SAAS,qBAAqB,QAAQ;wBAuhB3D,kBAAkB,QAAQ;;;UCjwBzB;EACf;EACA;;;;;;;wBAQoB,yBACpB,OAAO,YACP,gBAAgB,0BAChB,iBACC,QAAQ;;KA6BC;;;;;UAMK;EACf;EACA,MAAM;;EAEN;EACA;;UAGe;EACf;;EAEA,mBAAmB;;EAEnB;EACA;EACA;;;EAGA;;EAEA;EACA;;;;qBAKW;;;;;wBAMG,mBACd,cACA,mCACA,qBACC;;wBAWa,sBACd,wBACA,mBAAmB,oBACnB,iBACC;UAwBc;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;;;EAGA;EACA,cAAc;;;;;;;;qBAsBH,uCAAuC;mBAK/B;mBACA;EALnB,YAAY;EACZ,SAAS;EAET,YACmB,SAAS,mBACT,SAAS;EAGtB,YAAY,KAAK,uBAAuB,SAAS,eAAe;;UAyEvD;EACf;EACA;;EAEA;;EAEA;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA,QAAQ;;;;;;;;UASO,sBAAsB;EACrC;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,mBAAmB;;EAEnB;EACA;EACA;EACA;EACA;EACA,MAAM;IAAqB;;EAC3B,OAAO;IAAqB;;EAC5B;IAAW;IAAgB;IAAuB;;EAClD;IAAU;IAAkB;IAAc;;;wBAUtB,aAAa,SAAS,sBAAsB,QAAQ;;;UC5TzD;WACN;WACA;WACA;WACA;WACA;WACA;aACE;aACA;aACA;;WAEF,WAAW,SAAS;;;;;;;;wBAWf,kBAAkB,SAAS;;wBAgB3B,oBAAoB,SAAS;;;;;;;;;;KAoBjC;WAEG;WACA;WACA;WACA;WACA;;WAGA;WACA,kBAAkB;WAClB,WAAW;;UAGT;WACN;WACA;;WAEA;WACA;;WAEA;WACA;WACA;;KAGC;WACG;WAA2B,UAAU;;WACrC;WAA4B;;;;;;;wBAiC3B,4BACd,SAAS,mBACT,QAAQ,sBACP;KAmFS;UAMK;WACN;WACA;WACA;;WAEA;WACA,UAAU;;WAEV;;WAEA;WACA;;WAEA;;;;;;;;UASM,0BAA0B;WAChC;WACA,wBAAwB;WACxB,QAAQ,SAAS,OAAO;;WAExB;;UAGM;WACN,QAAQ;;WAER;;WAEA;WACA;WACA;;WAEA,gBAAgB;;;WAGhB,kBAAkB;WAClB,cAAc;;;;;;;;wBAST,gBAAgB,SAAS,gBAAgB;;;;;;;wBAyGnC,eACpB,mBAAmB,qBACnB,SAAS,wBACR,QAAQ;;wBAwNK,8BAA8B,KAAK;;;;;;wBAmCnC,iBAAiB,eAAe;;;UCvoB/B;WACN;WACA;WACA;WACA;;;wBAMK,2BAA2B,kBAAkB;UAW5C;WACN;WACA;;UAGM;WACN,WAAW;WACX,SAAS;;WAET;;;;;;;;wBASK,yBACd,SAAS,sBACT,kBAAkB,eACjB;UAuCc;WACN,kBAAkB;WAClB;;WAEA,YAAY;;WAEZ;WACA;WACA;;WAEA,gBAAgB;WAChB,cAAc;;UAGR;WACN,YAAY;WACZ;WACA,SAAS;;;;;;;wBAQE,oBACpB,SAAS,sBACT,SAAS,uBACR,QAAQ"}
|
|
@@ -2,7 +2,7 @@ import { t as mulberry32 } from "../random-Dn5fPWkt.js";
|
|
|
2
2
|
import { t as certificationEvidenceDigest } from "../verdict-DhPoXApk.js";
|
|
3
3
|
import { t as TraceEmitter } from "../emitter-DbMXUIVP.js";
|
|
4
4
|
import { t as runCounterfactual } from "../counterfactual-DtMWvn7D.js";
|
|
5
|
-
import { n as InMemoryTraceStore } from "../store-
|
|
5
|
+
import { n as InMemoryTraceStore } from "../store-BhNuLdJw.js";
|
|
6
6
|
import { t as packageVersion } from "../package-version-D7lQHt_-.js";
|
|
7
7
|
import { a as assertReplayableTrajectory, c as deriveFailureSignature, d as isRecordedTimeout, f as isSubmitAction, g as unreadableExitCount, h as parseRecordedReturncode, i as TIMEOUT_OBSERVATION_MARKER, l as finalRecordedOutcome, m as parseObservationOutput, n as FORMAT_ERROR_OBSERVATION_PREFIX, o as classifyObservation, p as isSubmitOnlyAction, r as SUBMIT_ACTION_SIGNATURE, s as decodeRecordedTurns, t as wrapActionForExec, u as isElidedField } from "../exec-ChJ6d5if.js";
|
|
8
8
|
import { createHash } from "node:crypto";
|
|
@@ -554,7 +554,7 @@ function dockerImagePreparer(options = {}) {
|
|
|
554
554
|
"image",
|
|
555
555
|
"inspect",
|
|
556
556
|
tag
|
|
557
|
-
], { maxBuffer:
|
|
557
|
+
], { maxBuffer: 8388608 });
|
|
558
558
|
return true;
|
|
559
559
|
} catch {
|
|
560
560
|
return false;
|
|
@@ -576,7 +576,7 @@ function dockerImagePreparer(options = {}) {
|
|
|
576
576
|
try {
|
|
577
577
|
await execFileAsync("docker", ["pull", image], {
|
|
578
578
|
timeout: pullTimeoutMs,
|
|
579
|
-
maxBuffer:
|
|
579
|
+
maxBuffer: 33554432
|
|
580
580
|
});
|
|
581
581
|
} catch (err) {
|
|
582
582
|
return {
|
|
@@ -606,7 +606,7 @@ function dockerImagePreparer(options = {}) {
|
|
|
606
606
|
if (!await imageExists(image)) try {
|
|
607
607
|
await execFileAsync("docker", ["pull", image], {
|
|
608
608
|
timeout: pullTimeoutMs,
|
|
609
|
-
maxBuffer:
|
|
609
|
+
maxBuffer: 33554432
|
|
610
610
|
});
|
|
611
611
|
pulled = true;
|
|
612
612
|
} catch (err) {
|
|
@@ -626,7 +626,7 @@ function dockerImagePreparer(options = {}) {
|
|
|
626
626
|
contextDir
|
|
627
627
|
], {
|
|
628
628
|
timeout: buildTimeoutMs,
|
|
629
|
-
maxBuffer:
|
|
629
|
+
maxBuffer: 33554432
|
|
630
630
|
});
|
|
631
631
|
} catch (err) {
|
|
632
632
|
return {
|