aspectcode 0.4.0 → 1.0.0

package/node_modules/@aspectcode/evaluator/dist/runner.js

@@ -3,183 +3,58 @@
  * Probe runner — simulates AI responses to probes using AGENTS.md as context.
  *
  * For each probe, constructs a chat where:
- * - System prompt = current AGENTS.md + relevant file contents
+ * - System prompt = current AGENTS.md
  * - User prompt = the probe's task
- * Then sends it to the LLM and collects the response.
+ * Then sends it to the LLM (temperature 0.0) and returns the raw response.
+ *
+ * Judging/evaluation is handled separately by the judge module.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.runProbes = runProbes;
-exports.buildProbeSystemPrompt = buildProbeSystemPrompt;
-exports.buildBehaviorEvalPrompt = buildBehaviorEvalPrompt;
-exports.parseBehaviorEval = parseBehaviorEval;
-/** Maximum file content characters to include per probe. */
-const MAX_CONTEXT_CHARS = 20000;
-/**
- * Build the system prompt for a probe run.
- * Includes the AGENTS.md instructions and relevant file contents.
- */
-function buildProbeSystemPrompt(agentsContent, probe, fileContents) {
-    let prompt = `You are an AI coding assistant. Follow these project instructions:\n\n${agentsContent}`;
-    if (fileContents && probe.contextFiles.length > 0) {
-        let contextChars = 0;
-        const fileSections = [];
-        for (const filePath of probe.contextFiles) {
-            const content = fileContents.get(filePath);
-            if (!content)
-                continue;
-            if (contextChars + content.length > MAX_CONTEXT_CHARS)
-                break;
-            fileSections.push(`### ${filePath}\n\`\`\`\n${content}\n\`\`\``);
-            contextChars += content.length;
-        }
-        if (fileSections.length > 0) {
-            prompt += `\n\n## Relevant Files\n\n${fileSections.join('\n\n')}`;
-        }
-    }
-    return prompt;
-}
+const llmUtil_1 = require("./llmUtil");
 /**
- * Build the evaluation prompt that scores a probe response
- * against expected behaviours.
+ * Run a single probe simulation.
+ * Returns the raw AI response without evaluation.
  */
-function buildBehaviorEvalPrompt(probe, response) {
-    const behaviors = probe.expectedBehaviors
-        .map((b, i) => `${i + 1}. ${b}`)
-        .join('\n');
-    return `You are evaluating an AI coding assistant's response to a specific task.
-
-## Task Given
-${probe.task}
-
-## Expected Behaviours
-The response should exhibit ALL of these behaviours:
-${behaviors}
-
-## AI Response
-${response}
-
-## Instructions
-For EACH expected behaviour, determine if the response exhibits it.
-Respond in EXACTLY this format (one line per behaviour):
-
-BEHAVIOR_1: PASS|FAIL — <brief explanation>
-BEHAVIOR_2: PASS|FAIL — <brief explanation>
-...
-
-Then a final line:
-OVERALL: PASS|FAIL`;
-}
-/** Parse the structured behaviour evaluation response. */
-function parseBehaviorEval(response, expectedBehaviors) {
-    const results = [];
-    const lines = response.split('\n');
-    for (let i = 0; i < expectedBehaviors.length; i++) {
-        const pattern = new RegExp(`BEHAVIOR_${i + 1}:\\s*(PASS|FAIL)\\s*[—-]\\s*(.*)`, 'i');
-        const match = lines.find((l) => pattern.test(l));
-        const parsed = match ? pattern.exec(match) : null;
-        results.push({
-            behavior: expectedBehaviors[i],
-            passed: parsed ? parsed[1].toUpperCase() === 'PASS' : false,
-            explanation: parsed ? parsed[2].trim() : 'Could not parse evaluation result',
-        });
-    }
-    const allPassed = results.every((r) => r.passed);
-    return { results, allPassed };
-}
-/**
- * Run a single probe: simulate the AI response, then evaluate it.
- */
-async function runSingleProbe(probe, agentsContent, provider, fileContents, log, signal) {
+async function simulateProbe(probe, agentsContent, provider, log, signal) {
     if (signal?.aborted) {
-        return {
-            probeId: probe.id,
-            passed: false,
-            response: '',
-            shortcomings: ['Cancelled'],
-            behaviorResults: [],
-        };
+        return { probeId: probe.id, task: probe.task, response: '' };
     }
-    // Step 1: Simulate the AI response using AGENTS.md as context
-    log?.debug(`Running probe: ${probe.id}`);
-    const systemPrompt = buildProbeSystemPrompt(agentsContent, probe, fileContents);
-    const simMessages = [
+    log?.debug(`Simulating probe: ${probe.id}`);
+    const systemPrompt = `You are an AI coding assistant. Follow these project instructions:\n\n${agentsContent}`;
+    const messages = [
         { role: 'system', content: systemPrompt },
         { role: 'user', content: probe.task },
     ];
     let response;
     try {
-        response = await provider.chat(simMessages);
+        response = await (0, llmUtil_1.chatWithTemp)(provider, messages, 0.0, signal);
     }
     catch (err) {
         const msg = err instanceof Error ? err.message : String(err);
         log?.warn(`Probe ${probe.id} simulation failed: ${msg}`);
-        return {
-            probeId: probe.id,
-            passed: false,
-            response: '',
-            shortcomings: [`LLM error during simulation: ${msg}`],
-            behaviorResults: [],
-        };
-    }
-    if (signal?.aborted) {
-        return {
-            probeId: probe.id,
-            passed: false,
-            response,
-            shortcomings: ['Cancelled during evaluation'],
-            behaviorResults: [],
-        };
-    }
-    // Step 2: Evaluate the response against expected behaviours
-    log?.debug(`Evaluating probe: ${probe.id}`);
-    const evalPrompt = buildBehaviorEvalPrompt(probe, response);
-    const evalMessages = [
-        { role: 'user', content: evalPrompt },
-    ];
-    let evalResponse;
-    try {
-        evalResponse = await provider.chat(evalMessages);
-    }
-    catch (err) {
-        const msg = err instanceof Error ? err.message : String(err);
-        log?.warn(`Probe ${probe.id} evaluation failed: ${msg}`);
-        return {
-            probeId: probe.id,
-            passed: false,
-            response,
-            shortcomings: [`LLM error during evaluation: ${msg}`],
-            behaviorResults: [],
-        };
+        return { probeId: probe.id, task: probe.task, response: '' };
     }
-    const { results: behaviorResults, allPassed } = parseBehaviorEval(evalResponse, probe.expectedBehaviors);
-    const shortcomings = behaviorResults
-        .filter((r) => !r.passed)
-        .map((r) => `${r.behavior}: ${r.explanation}`);
-    return {
-        probeId: probe.id,
-        passed: allPassed,
-        response,
-        shortcomings,
-        behaviorResults,
-    };
+    return { probeId: probe.id, task: probe.task, response };
 }
 /**
  * Run all probes against the current AGENTS.md.
  *
  * Each probe is run sequentially (to respect rate limits).
- * Returns results for all probes.
+ * Returns simulation results (raw responses, no evaluation).
  */
-async function runProbes(agentsContent, probes, provider, fileContents, log, signal, onProbeProgress) {
+async function runProbes(agentsContent, probes, provider, log, signal, onProbeProgress) {
     const results = [];
     for (let idx = 0; idx < probes.length; idx++) {
         const probe = probes[idx];
         if (signal?.aborted)
             break;
         onProbeProgress?.({ probeIndex: idx, total: probes.length, probeId: probe.id, phase: 'starting' });
-        const result = await runSingleProbe(probe, agentsContent, provider, fileContents, log, signal);
+        const result = await simulateProbe(probe, agentsContent, provider, log, signal);
         results.push(result);
-        onProbeProgress?.({ probeIndex: idx, total: probes.length, probeId: probe.id, phase: 'done', passed: result.passed });
-        log?.info(`  ${result.passed ? '✔' : '✖'} ${probe.id}`);
+        const hasResponse = result.response.length > 0;
+        onProbeProgress?.({ probeIndex: idx, total: probes.length, probeId: probe.id, phase: 'done', passed: hasResponse });
+        log?.info(`  ${hasResponse ? '✔' : '✖'} ${probe.id}`);
     }
     return results;
 }

package/node_modules/@aspectcode/evaluator/dist/runner.js.map

@@ -1 +1 @@
-{"version":3,"file":"runner.js","sourceRoot":"","sources":["../src/runner.ts"],"names":[],"mappings":";AAAA;;;;;;;GAOG;;AA8MH,8BAgCC;AAGQ,wDAAsB;AAAE,0DAAuB;AAAE,8CAAiB;AA5O3E,4DAA4D;AAC5D,MAAM,iBAAiB,GAAG,KAAM,CAAC;AAEjC;;;GAGG;AACH,SAAS,sBAAsB,CAC7B,aAAqB,EACrB,KAAY,EACZ,YAA0C;IAE1C,IAAI,MAAM,GAAG,yEAAyE,aAAa,EAAE,CAAC;IAEtG,IAAI,YAAY,IAAI,KAAK,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAClD,IAAI,YAAY,GAAG,CAAC,CAAC;QACrB,MAAM,YAAY,GAAa,EAAE,CAAC;QAElC,KAAK,MAAM,QAAQ,IAAI,KAAK,CAAC,YAAY,EAAE,CAAC;YAC1C,MAAM,OAAO,GAAG,YAAY,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;YAC3C,IAAI,CAAC,OAAO;gBAAE,SAAS;YACvB,IAAI,YAAY,GAAG,OAAO,CAAC,MAAM,GAAG,iBAAiB;gBAAE,MAAM;YAC7D,YAAY,CAAC,IAAI,CAAC,OAAO,QAAQ,aAAa,OAAO,UAAU,CAAC,CAAC;YACjE,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC;QACjC,CAAC;QAED,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,MAAM,IAAI,4BAA4B,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;QACpE,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;GAGG;AACH,SAAS,uBAAuB,CAC9B,KAAY,EACZ,QAAgB;IAEhB,MAAM,SAAS,GAAG,KAAK,CAAC,iBAAiB;SACtC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;SAC/B,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,OAAO;;;EAGP,KAAK,CAAC,IAAI;;;;EAIV,SAAS;;;EAGT,QAAQ;;;;;;;;;;;mBAWS,CAAC;AACpB,CAAC;AAED,0DAA0D;AAC1D,SAAS,iBAAiB,CACxB,QAAgB,EAChB,iBAA2B;IAE3B,MAAM,OAAO,GAAqB,EAAE,CAAC;IACrC,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAEnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,iBAAiB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClD,MAAM,OAAO,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC,GAAG,CAAC,kCAAkC,EAAE,GAAG,CAAC,CAAC;QACrF,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;QACjD,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAElD,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,iBAAiB,CAAC,CAAC,CAAC;YAC9B,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,KAAK,MAAM,CAAC,CAAC,CAAC,KAAK;YAC3D,WAAW,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,mCAAmC;SAC7E,CAAC,CAAC;IACL,CAAC;IAED,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IACjD,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC;AAChC,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,cAAc,CAC3B,KAAY,EACZ,aAAqB,EACrB,QAAqB,EACrB,YAA0C,EAC1C,GAAe,EACf,MAAoB;IAEpB,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,OAAO;YACL,OAAO,EAAE,KAAK,CAAC,EAAE;YACjB,MAAM,EAAE,KAAK;YACb,QAAQ,EAAE,EAAE;YACZ,YAAY,EAAE,CAAC,WAAW,CAAC;YAC3B,eAAe,EAAE,EAAE;SACpB,CAAC;IACJ,CAAC;IAED,8DAA8D;IAC9D,GAAG,EAAE,KAAK,CAAC,kBAAkB,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAEzC,MAAM,YAAY,GAAG,sBAAsB,CAAC,aAAa,EAAE,KAAK,EAAE,YAAY,CAAC,CAAC;IAChF,MAAM,WAAW,GAAkB;QACjC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,YAAY,EAAE;QACzC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,EAAE;KACtC,CAAC;IAEF,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IAC9C,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,GAAG,EAAE,IAAI,CAAC,SAAS,KAAK,CAAC,EAAE,uBAAuB,GAAG,EAAE,CAAC,CAAC;QACzD,OAAO;YACL,OAAO,EAAE,KAAK,CAAC,EAAE;YACjB,MAAM,EAAE,KAAK;YACb,QAAQ,EAAE,EAAE;YACZ,YAAY,EAAE,CAAC,gCAAgC,GAAG,EAAE,CAAC;YACrD,eAAe,EAAE,EAAE;SACpB,CAAC;IACJ,CAAC;IAED,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,OAAO;YACL,OAAO,EAAE,KAAK,CAAC,EAAE;YACjB,MAAM,EAAE,KAAK;YACb,QAAQ;YACR,YAAY,EAAE,CAAC,6BAA6B,CAAC;YAC7C,eAAe,EAAE,EAAE;SACpB,CAAC;IACJ,CAAC;IAED,4DAA4D;IAC5D,GAAG,EAAE,KAAK,CAAC,qBAAqB,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAE5C,MAAM,UAAU,GAAG,uBAAuB,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;IAC5D,MAAM,YAAY,GAAkB;QAClC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE;KACtC,CAAC;IAEF,IAAI,YAAoB,CAAC;IACzB,IAAI,CAAC;QACH,YAAY,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IACnD,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,GAAG,EAAE,IAAI,CAAC,SAAS,KAAK,CAAC,EAAE,uBAAuB,GAAG,EAAE,CAAC,CAAC;QACzD,OAAO;YACL,OAAO,EAAE,KAAK,CAAC,EAAE;YACjB,MAAM,EAAE,KAAK;YACb,QAAQ;YACR,YAAY,EAAE,CAAC,gCAAgC,GAAG,EAAE,CAAC;YACrD,eAAe,EAAE,EAAE;SACpB,CAAC;IACJ,CAAC;IAED,MAAM,EAAE,OAAO,EAAE,eAAe,EAAE,SAAS,EAAE,GAAG,iBAAiB,CAC/D,YAAY,EACZ,KAAK,CAAC,iBAAiB,CACxB,CAAC;IAEF,MAAM,YAAY,GAAG,eAAe;SACjC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;SACxB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;IAEjD,OAAO;QACL,OAAO,EAAE,KAAK,CAAC,EAAE;QACjB,MAAM,EAAE,SAAS;QACjB,QAAQ;QACR,YAAY;QACZ,eAAe;KAChB,CAAC;AACJ,CAAC;AAOD;;;;;GAKG;AACI,KAAK,UAAU,SAAS,CAC7B,aAAqB,EACrB,MAAe,EACf,QAAqB,EACrB,YAA0C,EAC1C,GAAe,EACf,MAAoB,EACpB,eAAuC;IAEvC,MAAM,OAAO,GAAkB,EAAE,CAAC;IAElC,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,MAAM,CAAC,MAAM,EAAE,GAAG,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;QAC1B,IAAI,MAAM,EAAE,OAAO;YAAE,MAAM;QAE3B,eAAe,EAAE,CAAC,EAAE,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QAEnG,MAAM,MAAM,GAAG,MAAM,cAAc,CACjC,KAAK,EACL,aAAa,EACb,QAAQ,EACR,YAAY,EACZ,GAAG,EACH,MAAM,CACP,CAAC;QACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAErB,eAAe,EAAE,CAAC,EAAE,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;QACtH,GAAG,EAAE,IAAI,CAAC,KAAK,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC"}
+{"version":3,"file":"runner.js","sourceRoot":"","sources":["../src/runner.ts"],"names":[],"mappings":";AAAA;;;;;;;;;GASG;;AAqDH,8BAyBC;AApED,uCAAyC;AAEzC;;;GAGG;AACH,KAAK,UAAU,aAAa,CAC1B,KAAY,EACZ,aAAqB,EACrB,QAAqB,EACrB,GAAe,EACf,MAAoB;IAEpB,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC;IAC/D,CAAC;IAED,GAAG,EAAE,KAAK,CAAC,qBAAqB,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAE5C,MAAM,YAAY,GAAG,yEAAyE,aAAa,EAAE,CAAC;IAC9G,MAAM,QAAQ,GAAkB;QAC9B,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,YAAY,EAAE;QACzC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,EAAE;KACtC,CAAC;IAEF,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,IAAA,sBAAY,EAAC,QAAQ,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;IACjE,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,GAAG,EAAE,IAAI,CAAC,SAAS,KAAK,CAAC,EAAE,uBAAuB,GAAG,EAAE,CAAC,CAAC;QACzD,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC;IAC/D,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC;AAC3D,CAAC;AAED;;;;;GAKG;AACI,KAAK,UAAU,SAAS,CAC7B,aAAqB,EACrB,MAAe,EACf,QAAqB,EACrB,GAAe,EACf,MAAoB,EACpB,eAAuC;IAEvC,MAAM,OAAO,GAAuB,EAAE,CAAC;IAEvC,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,MAAM,CAAC,MAAM,EAAE,GAAG,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;QAC1B,IAAI,MAAM,EAAE,OAAO;YAAE,MAAM;QAE3B,eAAe,EAAE,CAAC,EAAE,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QAEnG,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,KAAK,EAAE,aAAa,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAChF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAErB,MAAM,WAAW,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC;QAC/C,eAAe,EAAE,CAAC,EAAE,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC,CAAC;QACpH,GAAG,EAAE,IAAI,CAAC,KAAK,WAAW,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IACxD,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/node_modules/@aspectcode/evaluator/dist/types.d.ts

@@ -1,145 +1,166 @@
 /**
  * @aspectcode/evaluator — core types.
  *
- * Types for probe-based evaluation, prompt harvesting, and
- * evidence-based diagnosis of AGENTS.md quality.
+ * Types for probe-based evaluation, probe-and-refine tuning,
+ * and evidence-based diagnosis of AGENTS.md quality.
  */
 import type { LlmProvider, OptLogger } from '@aspectcode/optimizer';
-export type { LlmProvider, OptLogger } from '@aspectcode/optimizer';
+export type { LlmProvider, ChatOptions, OptLogger } from '@aspectcode/optimizer';
 /**
- * A single micro-test that evaluates whether AGENTS.md guides
+ * A single synthetic task that evaluates whether AGENTS.md guides
  * the AI correctly for a specific scenario scoped to the codebase.
  */
 export interface Probe {
-    /** Unique identifier (e.g. "hub-auth-middleware-naming"). */
+    /** Unique identifier. */
     id: string;
-    /** Human-readable description of what this probe tests. */
-    description: string;
-    /** Category for grouping (e.g. "hub-safety", "naming", "architecture"). */
-    category: ProbeCategory;
-    /**
-     * Workspace-relative paths of files relevant to this probe.
-     * These are included as context when running the probe.
-     */
-    contextFiles: string[];
     /** The task/question posed to the AI in this probe. */
     task: string;
     /**
      * Specific behaviours the AI's response should exhibit.
-     * Used by the evaluator to score the response.
+     * Used by the judge to score the response.
      */
     expectedBehaviors: string[];
+    /** Why this probe is useful (optional rationale from the generator). */
+    rationale?: string;
 }
-/** Probe categories for grouping and prioritization. */
-export type ProbeCategory = 'hub-safety' | 'naming' | 'architecture' | 'entry-point' | 'integration' | 'convention' | 'dependency' | 'harvested';
-/** Result of running a single probe against the current AGENTS.md. */
-export interface ProbeResult {
-    /** The probe that was run. */
+/** Raw result of simulating a single probe (before judging). */
+export interface SimulationResult {
     probeId: string;
-    /** Whether all expected behaviours were exhibited. */
-    passed: boolean;
-    /** The AI's simulated response to the probe task. */
+    task: string;
     response: string;
-    /**
-     * Specific shortcomings identified by the evaluator.
-     * Empty if `passed` is true.
-     */
-    shortcomings: string[];
-    /** Per-behaviour pass/fail breakdown. */
-    behaviorResults: BehaviorResult[];
 }
-/** Pass/fail for a single expected behaviour within a probe. */
-export interface BehaviorResult {
+/** Per-behavior assessment from the judge (strong/partial/missing). */
+export interface BehaviorReview {
     /** The expected behaviour description. */
     behavior: string;
-    /** Whether the response exhibited this behaviour. */
-    passed: boolean;
-    /** Brief explanation of why it passed or failed. */
-    explanation: string;
-}
-/** Diagnosis of AGENTS.md shortcomings based on failed probes. */
-export interface Diagnosis {
-    /** Specific edits proposed for AGENTS.md. */
-    edits: AgentsEdit[];
-    /** High-level summary of what's wrong. */
-    summary: string;
-    /** Number of probe failures this diagnosis addresses. */
-    failureCount: number;
+    /** How well the response exhibited this behaviour. */
+    assessment: 'strong' | 'partial' | 'missing';
+    /** Short excerpt from response supporting the assessment. */
+    evidence: string;
+    /** What AGENTS.md should add/change to improve this behaviour. */
+    improvement: string;
+}
+/** Result of judging a single probe's response. */
+export interface JudgedProbeResult {
+    /** The probe that was judged. */
+    probeId: string;
+    /** The original task. */
+    task: string;
+    /** The AI's simulated response. */
+    response: string;
+    /** Per-behaviour assessments. */
+    behaviorReviews: BehaviorReview[];
+    /** Per-probe edit suggestions from the judge (up to 3). */
+    proposedEdits: AgentsEdit[];
+    /** Summary notes from the judge. */
+    overallNotes: string;
 }
 /** A specific proposed edit to AGENTS.md. */
 export interface AgentsEdit {
-    /** What section/area of AGENTS.md to modify. */
+    /** What section/area to modify. AGENTS.md section name, or "scoped:slug" / "scoped:CREATE:slug" / "scoped:DELETE:slug". */
     section: string;
     /** The type of edit. */
     action: 'add' | 'modify' | 'strengthen' | 'remove';
     /** The proposed rule or content change. */
     content: string;
-    /** Which probe failures motivated this edit. */
-    motivatedBy: string[];
+    /** Which probe failures motivated this edit (optional). */
+    motivatedBy?: string[];
+    /** Glob patterns (only for scoped:CREATE). */
+    globs?: string[];
+    /** Description (only for scoped:CREATE). */
+    description?: string;
 }
-/** A conversation turn harvested from an AI tool's history. */
-export interface HarvestedPrompt {
-    /** Which tool this came from. */
-    source: PromptSource;
-    /** When this conversation happened (ISO-8601, if available). */
-    timestamp?: string;
-    /** The user's prompt/question. */
-    userPrompt: string;
-    /** The AI's response. */
-    assistantResponse: string;
-    /** Workspace-relative file paths referenced in the conversation. */
-    filesReferenced: string[];
+/** Configuration for the multi-iteration probe-and-refine loop. */
+export interface ProbeRefineConfig {
+    /** Maximum iterations before stopping. Default: 3. */
+    maxIterations: number;
+    /** Target probes per iteration. Default: 10. */
+    targetProbesPerIteration: number;
+    /** Max edits applied per iteration. Default: 5. */
+    maxEditsPerIteration: number;
+    /** Character budget for the AGENTS.md artifact. Default: 8000. */
+    charBudget: number;
 }
-/** Supported prompt history sources. */
-export type PromptSource = 'aider' | 'claude-code' | 'cline' | 'copilot-chat' | 'cursor' | 'windsurf' | 'export';
-/** Options for probe generation. */
+/** Default probe-and-refine configuration. */
+export declare const DEFAULT_PROBE_REFINE_CONFIG: ProbeRefineConfig;
+/** Summary of a single iteration in the probe-and-refine loop. */
+export interface IterationSummary {
+    iteration: number;
+    probesGenerated: number;
+    probesEvaluated: number;
+    editsApplied: number;
+    guidanceChanged: boolean;
+    charsBefore: number;
+    charsAfter: number;
+}
+/** Result of the full probe-and-refine loop. */
+export interface ProbeRefineResult {
+    /** The final refined AGENTS.md content. */
+    finalContent: string;
+    /** Per-iteration summaries. */
+    iterations: IterationSummary[];
+    /** Why the loop stopped (if before maxIterations). */
+    convergedReason?: string;
+}
+/** Options for LLM-powered probe generation. */
 export interface ProbeGeneratorOptions {
-    /** Full KB content for deriving probes. */
+    /** Full KB content for context. */
     kb: string;
-    /**
-     * Line-level KB diff (undefined on first run).
-     * When provided, probes focus on changed areas.
-     */
-    kbDiff?: string;
-    /** Harvested prompts to generate additional probes from. */
-    harvestedPrompts?: HarvestedPrompt[];
+    /** Current AGENTS.md content being tuned. */
+    currentAgentsMd: string;
+    /** Prior probe tasks (across iterations) for deduplication. */
+    priorProbeTasks: string[];
     /** Maximum number of probes to generate. Default: 10. */
     maxProbes?: number;
-    /** File contents map (workspace-relative path → content). */
-    fileContents?: ReadonlyMap<string, string>;
+    /** LLM provider for generating probes. */
+    provider: LlmProvider;
+    /** Project name (derived from workspace root). */
+    projectName?: string;
+    /** AbortSignal for cooperative cancellation. */
+    signal?: AbortSignal;
+    /** Optional logger. */
+    log?: OptLogger;
 }
-/** Options for running probes. */
+/** Options for running probe simulations. */
 export interface ProbeRunnerOptions {
-    /** Current AGENTS.md content (used as system prompt for probes). */
+    /** Current AGENTS.md content (used as system prompt). */
     agentsContent: string;
-    /** Probes to run. */
+    /** Probes to simulate. */
     probes: Probe[];
     /** LLM provider for simulating AI responses. */
     provider: LlmProvider;
-    /** File contents map for including context files. */
-    fileContents?: ReadonlyMap<string, string>;
     /** Optional logger. */
     log?: OptLogger;
     /** AbortSignal for cooperative cancellation. */
     signal?: AbortSignal;
 }
-/** Options for evaluating probe responses. */
-export interface ProbeEvaluatorOptions {
-    /** Probe results to evaluate. */
-    results: ProbeResult[];
-    /** LLM provider for evaluation. */
+/** Options for judging a probe's response. */
+export interface JudgeOptions {
+    /** The probe task. */
+    task: string;
+    /** The AI's simulated response. */
+    response: string;
+    /** Expected behaviours to judge against. */
+    expectedBehaviors: string[];
+    /** Probe ID for tracking. */
+    probeId: string;
+    /** LLM provider for judging. */
     provider: LlmProvider;
     /** Optional logger. */
     log?: OptLogger;
     /** AbortSignal for cooperative cancellation. */
     signal?: AbortSignal;
 }
-/** Options for diagnosing AGENTS.md issues from probe failures. */
+/** Options for diagnosing AGENTS.md issues from judged probes. */
 export interface DiagnosisOptions {
-    /** Failed probe results. */
-    failures: ProbeResult[];
+    /** All judged probe results (including strong ones). */
+    judgedResults: JudgedProbeResult[];
     /** Current AGENTS.md content. */
     agentsContent: string;
+    /** Current scoped rules context (slug → content map). Optional. */
+    scopedRulesContext?: string;
+    /** Raw static analysis data for scoped rule decisions. Optional. */
+    staticAnalysisData?: string;
     /** LLM provider for diagnosis. */
     provider: LlmProvider;
     /** Optional logger. */
@@ -147,6 +168,40 @@ export interface DiagnosisOptions {
     /** AbortSignal for cooperative cancellation. */
     signal?: AbortSignal;
 }
+/** Callback invoked before/after each probe for live progress updates. */
+export interface ProbeProgressCallback {
+    (info: {
+        probeIndex: number;
+        total: number;
+        probeId: string;
+        phase: 'starting' | 'done';
+        passed?: boolean;
+    }): void;
+}
+/** Result of deterministic edit application. */
+export interface ApplyResult {
+    /** The updated AGENTS.md content. */
+    content: string;
+    /** Number of edits successfully applied. */
+    applied: number;
+    /** Number of bullets trimmed to fit budget. */
+    trimmed: number;
+}
+/** A conversation turn harvested from an AI tool's history. */
+export interface HarvestedPrompt {
+    /** Which tool this came from. */
+    source: PromptSource;
+    /** When this conversation happened (ISO-8601, if available). */
+    timestamp?: string;
+    /** The user's prompt/question. */
+    userPrompt: string;
+    /** The AI's response. */
+    assistantResponse: string;
+    /** Workspace-relative file paths referenced in the conversation. */
+    filesReferenced: string[];
+}
+/** Supported prompt history sources. */
+export type PromptSource = 'aider' | 'claude-code' | 'cline' | 'copilot-chat' | 'cursor' | 'windsurf' | 'export';
 /** Options for prompt harvesting. */
 export interface HarvestOptions {
     /** Workspace root directory. */
@@ -160,17 +215,4 @@ export interface HarvestOptions {
     /** Optional logger. */
     log?: OptLogger;
 }
-/** Full result of the evaluation pipeline. */
-export interface EvaluationResult {
-    /** All probe results (passed + failed). */
-    probeResults: ProbeResult[];
-    /** Diagnosis based on failures (undefined if all probes passed). */
-    diagnosis?: Diagnosis;
-    /** Number of probes that passed. */
-    passCount: number;
-    /** Number of probes that failed. */
-    failCount: number;
-    /** Total probes run. */
-    totalProbes: number;
-}
 //# sourceMappingURL=types.d.ts.map

package/node_modules/@aspectcode/evaluator/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAGpE,YAAY,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAIpE;;;GAGG;AACH,MAAM,WAAW,KAAK;IACpB,6DAA6D;IAC7D,EAAE,EAAE,MAAM,CAAC;IAEX,2DAA2D;IAC3D,WAAW,EAAE,MAAM,CAAC;IAEpB,2EAA2E;IAC3E,QAAQ,EAAE,aAAa,CAAC;IAExB;;;OAGG;IACH,YAAY,EAAE,MAAM,EAAE,CAAC;IAEvB,uDAAuD;IACvD,IAAI,EAAE,MAAM,CAAC;IAEb;;;OAGG;IACH,iBAAiB,EAAE,MAAM,EAAE,CAAC;CAC7B;AAED,wDAAwD;AACxD,MAAM,MAAM,aAAa,GACrB,YAAY,GACZ,QAAQ,GACR,cAAc,GACd,aAAa,GACb,aAAa,GACb,YAAY,GACZ,YAAY,GACZ,WAAW,CAAC;AAIhB,sEAAsE;AACtE,MAAM,WAAW,WAAW;IAC1B,8BAA8B;IAC9B,OAAO,EAAE,MAAM,CAAC;IAEhB,sDAAsD;IACtD,MAAM,EAAE,OAAO,CAAC;IAEhB,qDAAqD;IACrD,QAAQ,EAAE,MAAM,CAAC;IAEjB;;;OAGG;IACH,YAAY,EAAE,MAAM,EAAE,CAAC;IAEvB,yCAAyC;IACzC,eAAe,EAAE,cAAc,EAAE,CAAC;CACnC;AAED,gEAAgE;AAChE,MAAM,WAAW,cAAc;IAC7B,0CAA0C;IAC1C,QAAQ,EAAE,MAAM,CAAC;IACjB,qDAAqD;IACrD,MAAM,EAAE,OAAO,CAAC;IAChB,oDAAoD;IACpD,WAAW,EAAE,MAAM,CAAC;CACrB;AAID,kEAAkE;AAClE,MAAM,WAAW,SAAS;IACxB,6CAA6C;IAC7C,KAAK,EAAE,UAAU,EAAE,CAAC;IAEpB,0CAA0C;IAC1C,OAAO,EAAE,MAAM,CAAC;IAEhB,yDAAyD;IACzD,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,6CAA6C;AAC7C,MAAM,WAAW,UAAU;IACzB,gDAAgD;IAChD,OAAO,EAAE,MAAM,CAAC;IAEhB,wBAAwB;IACxB,MAAM,EAAE,KAAK,GAAG,QAAQ,GAAG,YAAY,GAAG,QAAQ,CAAC;IAEnD,2CAA2C;IAC3C,OAAO,EAAE,MAAM,CAAC;IAEhB,gDAAgD;IAChD,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB;AAID,+DAA+D;AAC/D,MAAM,WAAW,eAAe;IAC9B,iCAAiC;IACjC,MAAM,EAAE,YAAY,CAAC;IAErB,gEAAgE;IAChE,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,kCAAkC;IAClC,UAAU,EAAE,MAAM,CAAC;IAEnB,yBAAyB;IACzB,iBAAiB,EAAE,MAAM,CAAC;IAE1B,oEAAoE;IACpE,eAAe,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,wCAAwC;AACxC,MAAM,MAAM,YAAY,GACpB,OAAO,GACP,aAAa,GACb,OAAO,GACP,cAAc,GACd,QAAQ,GACR,UAAU,GACV,QAAQ,CAAC;AAIb,oCAAoC;AACpC,MAAM,WAAW,qBAAqB;IACpC,2CAA2C;IAC3C,EAAE,EAAE,MAAM,CAAC;IAEX;;;OAGG;IACH,MAAM,CAAC,EAAE,MAAM,CAAC;IAEhB,4DAA4D;IAC5D,gBAAgB,CAAC,EAAE,eAAe,EAAE,CAAC;IAErC,yDAAyD;IACzD,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,6DAA6D;IAC7D,YAAY,CAAC,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC5C;AAED,kCAAkC;AAClC,MAAM,WAAW,kBAAkB;IACjC,oEAAoE;IACpE,aAAa,EAAE,MAAM,CAAC;IAEtB,qBAAqB;IACrB,MAAM,EAAE,KAAK,EAAE,CAAC;IAEhB,gDAAgD;IAChD,QAAQ,EAAE,WAAW,CAAC;IAEtB,qDAAqD;IACrD,YAAY,CAAC,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAE3C,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;IAEhB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,8CAA8C;AAC9C,MAAM,WAAW,qBAAqB;IACpC,iCAAiC;IACjC,OAAO,EAAE,WAAW,EAAE,CAAC;IAEvB,mCAAmC;IACnC,QAAQ,EAAE,WAAW,CAAC;IAEtB,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;IAEhB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,mEAAmE;AACnE,MAAM,WAAW,gBAAgB;IAC/B,4BAA4B;IAC5B,QAAQ,EAAE,WAAW,EAAE,CAAC;IAExB,iCAAiC;IACjC,aAAa,EAAE,MAAM,CAAC;IAEtB,kCAAkC;IAClC,QAAQ,EAAE,WAAW,CAAC;IAEtB,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;IAEhB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,qCAAqC;AACrC,MAAM,WAAW,cAAc;IAC7B,gCAAgC;IAChC,IAAI,EAAE,MAAM,CAAC;IAEb,gEAAgE;IAChE,OAAO,CAAC,EAAE,YAAY,EAAE,CAAC;IAEzB,0DAA0D;IAC1D,YAAY,CAAC,EAAE,MAAM,CAAC;IAEtB,iDAAiD;IACjD,KAAK,CAAC,EAAE,IAAI,CAAC;IAEb,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;CACjB;AAID,8CAA8C;AAC9C,MAAM,WAAW,gBAAgB;IAC/B,2CAA2C;IAC3C,YAAY,EAAE,WAAW,EAAE,CAAC;IAE5B,oEAAoE;IACpE,SAAS,CAAC,EAAE,SAAS,CAAC;IAEtB,oCAAoC;IACpC,SAAS,EAAE,MAAM,CAAC;IAElB,oCAAoC;IACpC,SAAS,EAAE,MAAM,CAAC;IAElB,wBAAwB;IACxB,WAAW,EAAE,MAAM,CAAC;CACrB"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAGpE,YAAY,EAAE,WAAW,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAIjF;;;GAGG;AACH,MAAM,WAAW,KAAK;IACpB,yBAAyB;IACzB,EAAE,EAAE,MAAM,CAAC;IAEX,uDAAuD;IACvD,IAAI,EAAE,MAAM,CAAC;IAEb;;;OAGG;IACH,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAE5B,wEAAwE;IACxE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAID,gEAAgE;AAChE,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,uEAAuE;AACvE,MAAM,WAAW,cAAc;IAC7B,0CAA0C;IAC1C,QAAQ,EAAE,MAAM,CAAC;IACjB,sDAAsD;IACtD,UAAU,EAAE,QAAQ,GAAG,SAAS,GAAG,SAAS,CAAC;IAC7C,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,CAAC;IACjB,kEAAkE;IAClE,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,mDAAmD;AACnD,MAAM,WAAW,iBAAiB;IAChC,iCAAiC;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,yBAAyB;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,mCAAmC;IACnC,QAAQ,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,eAAe,EAAE,cAAc,EAAE,CAAC;IAClC,2DAA2D;IAC3D,aAAa,EAAE,UAAU,EAAE,CAAC;IAC5B,oCAAoC;IACpC,YAAY,EAAE,MAAM,CAAC;CACtB;AAID,6CAA6C;AAC7C,MAAM,WAAW,UAAU;IACzB,2HAA2H;IAC3H,OAAO,EAAE,MAAM,CAAC;IAEhB,wBAAwB;IACxB,MAAM,EAAE,KAAK,GAAG,QAAQ,GAAG,YAAY,GAAG,QAAQ,CAAC;IAEnD,2CAA2C;IAC3C,OAAO,EAAE,MAAM,CAAC;IAEhB,2DAA2D;IAC3D,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IAEvB,8CAA8C;IAC9C,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;IAEjB,4CAA4C;IAC5C,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAID,mEAAmE;AACnE,MAAM,WAAW,iBAAiB;IAChC,sDAAsD;IACtD,aAAa,EAAE,MAAM,CAAC;IACtB,gDAAgD;IAChD,wBAAwB,EAAE,MAAM,CAAC;IACjC,mDAAmD;IACnD,oBAAoB,EAAE,MAAM,CAAC;IAC7B,kEAAkE;IAClE,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,8CAA8C;AAC9C,eAAO,MAAM,2BAA2B,EAAE,iBAKzC,CAAC;AAEF,kEAAkE;AAClE,MAAM,WAAW,gBAAgB;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB,eAAe,EAAE,MAAM,CAAC;IACxB,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,OAAO,CAAC;IACzB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,gDAAgD;AAChD,MAAM,WAAW,iBAAiB;IAChC,2CAA2C;IAC3C,YAAY,EAAE,MAAM,CAAC;IACrB,+BAA+B;IAC/B,UAAU,EAAE,gBAAgB,EAAE,CAAC;IAC/B,sDAAsD;IACtD,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAID,gDAAgD;AAChD,MAAM,WAAW,qBAAqB;IACpC,mCAAmC;IACnC,EAAE,EAAE,MAAM,CAAC;IAEX,6CAA6C;IAC7C,eAAe,EAAE,MAAM,CAAC;IAExB,+DAA+D;IAC/D,eAAe,EAAE,MAAM,EAAE,CAAC;IAE1B,yDAAyD;IACzD,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,0CAA0C;IAC1C,QAAQ,EAAE,WAAW,CAAC;IAEtB,kDAAkD;IAClD,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;IAErB,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;CACjB;AAED,6CAA6C;AAC7C,MAAM,WAAW,kBAAkB;IACjC,yDAAyD;IACzD,aAAa,EAAE,MAAM,CAAC;IAEtB,0BAA0B;IAC1B,MAAM,EAAE,KAAK,EAAE,CAAC;IAEhB,gDAAgD;IAChD,QAAQ,EAAE,WAAW,CAAC;IAEtB,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;IAEhB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,8CAA8C;AAC9C,MAAM,WAAW,YAAY;IAC3B,sBAAsB;IACtB,IAAI,EAAE,MAAM,CAAC;IAEb,mCAAmC;IACnC,QAAQ,EAAE,MAAM,CAAC;IAEjB,4CAA4C;IAC5C,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAE5B,6BAA6B;IAC7B,OAAO,EAAE,MAAM,CAAC;IAEhB,gCAAgC;IAChC,QAAQ,EAAE,WAAW,CAAC;IAEtB,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;IAEhB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,kEAAkE;AAClE,MAAM,WAAW,gBAAgB;IAC/B,wDAAwD;IACxD,aAAa,EAAE,iBAAiB,EAAE,CAAC;IAEnC,iCAAiC;IACjC,aAAa,EAAE,MAAM,CAAC;IAEtB,mEAAmE;IACnE,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAE5B,oEAAoE;IACpE,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAE5B,kCAAkC;IAClC,QAAQ,EAAE,WAAW,CAAC;IAEtB,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;IAEhB,gDAAgD;IAChD,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,0EAA0E;AAC1E,MAAM,WAAW,qBAAqB;IACpC,CAAC,IAAI,EAAE;QACL,UAAU,EAAE,MAAM,CAAC;QACnB,KAAK,EAAE,MAAM,CAAC;QACd,OAAO,EAAE,MAAM,CAAC;QAChB,KAAK,EAAE,UAAU,GAAG,MAAM,CAAC;QAC3B,MAAM,CAAC,EAAE,OAAO,CAAC;KAClB,GAAG,IAAI,CAAC;CACV;AAID,gDAAgD;AAChD,MAAM,WAAW,WAAW;IAC1B,qCAAqC;IACrC,OAAO,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,OAAO,EAAE,MAAM,CAAC;IAChB,+CAA+C;IAC/C,OAAO,EAAE,MAAM,CAAC;CACjB;AAID,+DAA+D;AAC/D,MAAM,WAAW,eAAe;IAC9B,iCAAiC;IACjC,MAAM,EAAE,YAAY,CAAC;IACrB,gEAAgE;IAChE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kCAAkC;IAClC,UAAU,EAAE,MAAM,CAAC;IACnB,yBAAyB;IACzB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,oEAAoE;IACpE,eAAe,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,wCAAwC;AACxC,MAAM,MAAM,YAAY,GACpB,OAAO,GACP,aAAa,GACb,OAAO,GACP,cAAc,GACd,QAAQ,GACR,UAAU,GACV,QAAQ,CAAC;AAEb,qCAAqC;AACrC,MAAM,WAAW,cAAc;IAC7B,gCAAgC;IAChC,IAAI,EAAE,MAAM,CAAC;IACb,gEAAgE;IAChE,OAAO,CAAC,EAAE,YAAY,EAAE,CAAC;IACzB,0DAA0D;IAC1D,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iDAAiD;IACjD,KAAK,CAAC,EAAE,IAAI,CAAC;IACb,uBAAuB;IACvB,GAAG,CAAC,EAAE,SAAS,CAAC;CACjB"}

package/node_modules/@aspectcode/evaluator/dist/types.js

@@ -2,8 +2,16 @@
 /**
  * @aspectcode/evaluator — core types.
  *
- * Types for probe-based evaluation, prompt harvesting, and
- * evidence-based diagnosis of AGENTS.md quality.
+ * Types for probe-based evaluation, probe-and-refine tuning,
+ * and evidence-based diagnosis of AGENTS.md quality.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.DEFAULT_PROBE_REFINE_CONFIG = void 0;
+/** Default probe-and-refine configuration. */
+exports.DEFAULT_PROBE_REFINE_CONFIG = {
+    maxIterations: 1,
+    targetProbesPerIteration: 5,
+    maxEditsPerIteration: 5,
+    charBudget: 8000,
+};
 //# sourceMappingURL=types.js.map

package/node_modules/@aspectcode/evaluator/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":";AAAA;;;;;GAKG"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;AAwGH,8CAA8C;AACjC,QAAA,2BAA2B,GAAsB;IAC5D,aAAa,EAAE,CAAC;IAChB,wBAAwB,EAAE,CAAC;IAC3B,oBAAoB,EAAE,CAAC;IACvB,UAAU,EAAE,IAAI;CACjB,CAAC"}

package/node_modules/@aspectcode/evaluator/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aspectcode/evaluator",
-  "version": "0.4.0",
+  "version": "1.0.0",
   "private": true,
   "description": "Evidence-based evaluation for Aspect Code",
   "license": "MIT",
@@ -14,7 +14,7 @@
   },
   "repository": {
     "type": "git",
-    "url": "https://github.com/asashepard/aspectcode.git",
+    "url": "https://github.com/Aspect-Code-Labs/aspectcode.git",
     "directory": "packages/evaluator"
   },
   "scripts": {
@@ -24,8 +24,8 @@
     "test": "mocha --require ts-node/register test/**/*.test.ts"
   },
   "dependencies": {
-    "@aspectcode/core": "0.3.1",
-    "@aspectcode/optimizer": "0.4.0"
+    "@aspectcode/core": "1.0.0",
+    "@aspectcode/optimizer": "1.0.0"
   },
   "optionalDependencies": {
     "better-sqlite3": "^11.0.0"

package/node_modules/@aspectcode/optimizer/dist/index.d.ts

@@ -1,15 +1,8 @@
 /**
  * @aspectcode/optimizer — public API.
- *
- * Re-exports types and provides the top-level `optimizeInstructions` entry point.
  */
-export type { LlmProvider, ChatMessage, ChatUsage, ChatResult, ProviderOptions, OptimizeOptions, OptimizeResult, OptimizeStep, OptLogger, ProviderName, ComplaintOptions, ComplaintResult, } from './types';
-export { PROVIDER_ENV_KEYS, LLM_PROVIDER_ENV, LLM_MODEL_ENV } from './types';
-export { resolveProvider, loadEnvFile, parseDotenv } from './providers/index';
-export { createOpenAiProvider } from './providers/openai';
-export { createAnthropicProvider } from './providers/anthropic';
+export type { LlmProvider, ChatMessage, ChatOptions, ChatUsage, ChatResult, ProviderOptions, OptimizeOptions, OptimizeResult, OptimizeStep, OptLogger, } from './types';
+export { resolveProvider, loadEnvFile } from './providers/index';
 export { withRetry } from './providers/retry';
-export type { RetryOptions } from './providers/retry';
-export { runGenerateAgent, runComplaintAgent } from './agent';
-export { buildSystemPrompt, buildGeneratePrompt, truncateKb, buildComplaintPrompt, parseComplaintResponse, } from './prompts';
+export { runGenerateAgent } from './agent';
 //# sourceMappingURL=index.d.ts.map

package/node_modules/@aspectcode/optimizer/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,YAAY,EACV,WAAW,EACX,WAAW,EACX,SAAS,EACT,UAAU,EACV,eAAe,EACf,eAAe,EACf,cAAc,EACd,YAAY,EACZ,SAAS,EACT,YAAY,EACZ,gBAAgB,EAChB,eAAe,GAChB,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAE7E,OAAO,EAAE,eAAe,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAC9E,OAAO,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAC;AAC1D,OAAO,EAAE,uBAAuB,EAAE,MAAM,uBAAuB,CAAC;AAChE,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAC9C,YAAY,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AACtD,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MAAM,SAAS,CAAC;AAC9D,OAAO,EACL,iBAAiB,EACjB,mBAAmB,EACnB,UAAU,EACV,oBAAoB,EACpB,sBAAsB,GACvB,MAAM,WAAW,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,YAAY,EACV,WAAW,EACX,WAAW,EACX,WAAW,EACX,SAAS,EACT,UAAU,EACV,eAAe,EACf,eAAe,EACf,cAAc,EACd,YAAY,EACZ,SAAS,GACV,MAAM,SAAS,CAAC;AAEjB,OAAO,EAAE,eAAe,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACjE,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAC9C,OAAO,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC"}