codeharness 0.35.1 → 0.35.2
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.
|
@@ -2895,7 +2895,7 @@ function generateDockerfileTemplate(projectDir, stackOrDetections) {
|
|
|
2895
2895
|
}
|
|
2896
2896
|
|
|
2897
2897
|
// src/modules/infra/init-project.ts
|
|
2898
|
-
var HARNESS_VERSION = true ? "0.35.
|
|
2898
|
+
var HARNESS_VERSION = true ? "0.35.2" : "0.0.0-dev";
|
|
2899
2899
|
function failResult(opts, error) {
|
|
2900
2900
|
return {
|
|
2901
2901
|
status: "fail",
|
package/dist/index.js
CHANGED
|
@@ -40,7 +40,7 @@ import {
|
|
|
40
40
|
validateDockerfile,
|
|
41
41
|
warn,
|
|
42
42
|
writeState
|
|
43
|
-
} from "./chunk-
|
|
43
|
+
} from "./chunk-KJF2YA5T.js";
|
|
44
44
|
|
|
45
45
|
// src/index.ts
|
|
46
46
|
import { Command } from "commander";
|
|
@@ -3609,6 +3609,14 @@ async function executeLoopBlock(loopBlock, state, config, workItems, initialCont
|
|
|
3609
3609
|
if (loopBlock.loop.length === 0) {
|
|
3610
3610
|
return { state: currentState, errors, tasksCompleted, halted: false, lastContract: lastOutputContract };
|
|
3611
3611
|
}
|
|
3612
|
+
const lastAgentTaskInLoop = (() => {
|
|
3613
|
+
for (let i = loopBlock.loop.length - 1; i >= 0; i--) {
|
|
3614
|
+
const tn = loopBlock.loop[i];
|
|
3615
|
+
const t = config.workflow.tasks[tn];
|
|
3616
|
+
if (t && t.agent !== null) return tn;
|
|
3617
|
+
}
|
|
3618
|
+
return loopBlock.loop[loopBlock.loop.length - 1];
|
|
3619
|
+
})();
|
|
3612
3620
|
while (true) {
|
|
3613
3621
|
const nextIteration = currentState.iteration + 1;
|
|
3614
3622
|
const allCurrentIterationDone = currentState.iteration > 0 && loopBlock.loop.every((tn) => {
|
|
@@ -3751,88 +3759,91 @@ async function executeLoopBlock(loopBlock, state, config, workItems, initialCont
|
|
|
3751
3759
|
propagateVerifyFlags(taskName, dispatchResult.contract, projectDir);
|
|
3752
3760
|
accumulatedCostUsd += dispatchResult.contract?.cost_usd ?? 0;
|
|
3753
3761
|
tasksCompleted++;
|
|
3754
|
-
|
|
3755
|
-
|
|
3756
|
-
verdict =
|
|
3757
|
-
|
|
3758
|
-
|
|
3759
|
-
|
|
3760
|
-
|
|
3761
|
-
|
|
3762
|
-
|
|
3763
|
-
|
|
3764
|
-
|
|
3765
|
-
|
|
3766
|
-
|
|
3767
|
-
|
|
3768
|
-
|
|
3769
|
-
|
|
3770
|
-
|
|
3771
|
-
|
|
3772
|
-
|
|
3773
|
-
|
|
3774
|
-
|
|
3775
|
-
|
|
3776
|
-
|
|
3777
|
-
|
|
3778
|
-
|
|
3779
|
-
|
|
3780
|
-
|
|
3762
|
+
const isLastTaskInLoop = taskName === lastAgentTaskInLoop;
|
|
3763
|
+
if (isLastTaskInLoop) {
|
|
3764
|
+
let verdict = null;
|
|
3765
|
+
try {
|
|
3766
|
+
verdict = parseVerdict(dispatchResult.output);
|
|
3767
|
+
} catch (parseErr) {
|
|
3768
|
+
if (parseErr instanceof VerdictParseError && parseErr.retryable) {
|
|
3769
|
+
warn(`workflow-engine: verdict parse failed, retrying evaluator for ${taskName}`);
|
|
3770
|
+
try {
|
|
3771
|
+
const retryResult = await dispatchTaskWithResult(
|
|
3772
|
+
task,
|
|
3773
|
+
taskName,
|
|
3774
|
+
PER_RUN_SENTINEL,
|
|
3775
|
+
definition,
|
|
3776
|
+
currentState,
|
|
3777
|
+
config,
|
|
3778
|
+
void 0,
|
|
3779
|
+
lastOutputContract ?? void 0
|
|
3780
|
+
);
|
|
3781
|
+
currentState = retryResult.updatedState;
|
|
3782
|
+
lastOutputContract = retryResult.contract;
|
|
3783
|
+
propagateVerifyFlags(taskName, retryResult.contract, projectDir);
|
|
3784
|
+
tasksCompleted++;
|
|
3785
|
+
verdict = parseVerdict(retryResult.output);
|
|
3786
|
+
} catch {
|
|
3787
|
+
verdict = buildAllUnknownVerdict(
|
|
3788
|
+
workItems,
|
|
3789
|
+
"Evaluator failed to produce valid JSON after retry"
|
|
3790
|
+
);
|
|
3791
|
+
}
|
|
3781
3792
|
}
|
|
3782
3793
|
}
|
|
3783
|
-
|
|
3784
|
-
|
|
3785
|
-
|
|
3786
|
-
|
|
3787
|
-
|
|
3788
|
-
|
|
3789
|
-
|
|
3790
|
-
|
|
3794
|
+
if (!verdict) {
|
|
3795
|
+
const tagged = parseVerdictTag(dispatchResult.output);
|
|
3796
|
+
if (tagged) {
|
|
3797
|
+
verdict = {
|
|
3798
|
+
verdict: tagged.verdict,
|
|
3799
|
+
score: { passed: tagged.verdict === "pass" ? 1 : 0, failed: tagged.verdict === "fail" ? 1 : 0, unknown: 0, total: 1 },
|
|
3800
|
+
findings: []
|
|
3801
|
+
};
|
|
3802
|
+
}
|
|
3803
|
+
}
|
|
3804
|
+
lastVerdict = verdict;
|
|
3805
|
+
if (verdict) {
|
|
3806
|
+
const score = {
|
|
3807
|
+
iteration: currentState.iteration,
|
|
3808
|
+
passed: verdict.score.passed,
|
|
3809
|
+
failed: verdict.score.failed,
|
|
3810
|
+
unknown: verdict.score.unknown,
|
|
3811
|
+
total: verdict.score.total,
|
|
3812
|
+
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
3813
|
+
};
|
|
3814
|
+
currentState = {
|
|
3815
|
+
...currentState,
|
|
3816
|
+
evaluator_scores: [...currentState.evaluator_scores, score]
|
|
3817
|
+
};
|
|
3818
|
+
} else {
|
|
3819
|
+
const totalItems = workItems.length;
|
|
3820
|
+
const score = {
|
|
3821
|
+
iteration: currentState.iteration,
|
|
3822
|
+
passed: 0,
|
|
3823
|
+
failed: 0,
|
|
3824
|
+
unknown: totalItems,
|
|
3825
|
+
total: totalItems,
|
|
3826
|
+
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
3827
|
+
};
|
|
3828
|
+
currentState = {
|
|
3829
|
+
...currentState,
|
|
3830
|
+
evaluator_scores: [...currentState.evaluator_scores, score]
|
|
3791
3831
|
};
|
|
3792
3832
|
}
|
|
3793
|
-
|
|
3794
|
-
|
|
3795
|
-
|
|
3796
|
-
|
|
3797
|
-
|
|
3798
|
-
|
|
3799
|
-
|
|
3800
|
-
|
|
3801
|
-
|
|
3802
|
-
|
|
3803
|
-
|
|
3804
|
-
|
|
3805
|
-
...currentState,
|
|
3806
|
-
evaluator_scores: [...currentState.evaluator_scores, score]
|
|
3807
|
-
};
|
|
3808
|
-
} else {
|
|
3809
|
-
const totalItems = workItems.length;
|
|
3810
|
-
const score = {
|
|
3811
|
-
iteration: currentState.iteration,
|
|
3812
|
-
passed: 0,
|
|
3813
|
-
failed: 0,
|
|
3814
|
-
unknown: totalItems,
|
|
3815
|
-
total: totalItems,
|
|
3816
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
3817
|
-
};
|
|
3818
|
-
currentState = {
|
|
3819
|
-
...currentState,
|
|
3820
|
-
evaluator_scores: [...currentState.evaluator_scores, score]
|
|
3821
|
-
};
|
|
3833
|
+
const cbDecision = evaluateProgress(currentState.evaluator_scores);
|
|
3834
|
+
if (cbDecision.halt) {
|
|
3835
|
+
currentState = {
|
|
3836
|
+
...currentState,
|
|
3837
|
+
circuit_breaker: {
|
|
3838
|
+
triggered: true,
|
|
3839
|
+
reason: cbDecision.reason,
|
|
3840
|
+
score_history: cbDecision.scoreHistory
|
|
3841
|
+
}
|
|
3842
|
+
};
|
|
3843
|
+
writeWorkflowState(currentState, projectDir);
|
|
3844
|
+
}
|
|
3822
3845
|
}
|
|
3823
3846
|
writeWorkflowState(currentState, projectDir);
|
|
3824
|
-
const cbDecision = evaluateProgress(currentState.evaluator_scores);
|
|
3825
|
-
if (cbDecision.halt) {
|
|
3826
|
-
currentState = {
|
|
3827
|
-
...currentState,
|
|
3828
|
-
circuit_breaker: {
|
|
3829
|
-
triggered: true,
|
|
3830
|
-
reason: cbDecision.reason,
|
|
3831
|
-
score_history: cbDecision.scoreHistory
|
|
3832
|
-
}
|
|
3833
|
-
};
|
|
3834
|
-
writeWorkflowState(currentState, projectDir);
|
|
3835
|
-
}
|
|
3836
3847
|
} catch (err) {
|
|
3837
3848
|
const engineError = handleDispatchError(err, taskName, PER_RUN_SENTINEL);
|
|
3838
3849
|
errors.push(engineError);
|
|
@@ -11188,7 +11199,7 @@ function registerTeardownCommand(program) {
|
|
|
11188
11199
|
} else if (otlpMode === "remote-routed") {
|
|
11189
11200
|
if (!options.keepDocker) {
|
|
11190
11201
|
try {
|
|
11191
|
-
const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-
|
|
11202
|
+
const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-EIWOFRFK.js");
|
|
11192
11203
|
stopCollectorOnly2();
|
|
11193
11204
|
result.docker.stopped = true;
|
|
11194
11205
|
if (!isJson) {
|
|
@@ -11220,7 +11231,7 @@ function registerTeardownCommand(program) {
|
|
|
11220
11231
|
info("Shared stack: kept running (other projects may use it)");
|
|
11221
11232
|
}
|
|
11222
11233
|
} else if (isLegacyStack) {
|
|
11223
|
-
const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-
|
|
11234
|
+
const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-EIWOFRFK.js");
|
|
11224
11235
|
let stackRunning = false;
|
|
11225
11236
|
try {
|
|
11226
11237
|
stackRunning = isStackRunning2(composeFile);
|
|
@@ -14207,7 +14218,7 @@ function registerDriversCommand(program) {
|
|
|
14207
14218
|
}
|
|
14208
14219
|
|
|
14209
14220
|
// src/index.ts
|
|
14210
|
-
var VERSION = true ? "0.35.
|
|
14221
|
+
var VERSION = true ? "0.35.2" : "0.0.0-dev";
|
|
14211
14222
|
function createProgram() {
|
|
14212
14223
|
const program = new Command();
|
|
14213
14224
|
program.name("codeharness").description("Makes autonomous coding agents produce software that actually works").version(VERSION).option("--json", "Output in machine-readable JSON format");
|