aspectcode 0.4.1 → 1.0.1

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

@@ -1,172 +1,149 @@
 "use strict";
 /**
- * Diagnosis engine — analyzes probe failures and proposes AGENTS.md edits.
+ * Aggregate diagnosis — analyzes judged probe results and proposes AGENTS.md edits.
  *
- * Takes failed probe results + current AGENTS.md, asks the LLM to identify
- * which rules are missing/weak/wrong, and returns targeted edits.
+ * Takes all judged probe results (including behaviour reviews and per-probe edits),
+ * aggregates the failures, and proposes up to 6 targeted AGENTS.md edits.
+ *
+ * Ported from sweagent_bench oracle/diagnose.py.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.parseDiagnoseResponse = parseDiagnoseResponse;
 exports.diagnose = diagnose;
-exports.applyDiagnosisEdits = applyDiagnosisEdits;
-exports.buildDiagnosisPrompt = buildDiagnosisPrompt;
-exports.parseDiagnosisResponse = parseDiagnosisResponse;
-/**
- * Build the diagnosis prompt from failed probe results.
- */
-function buildDiagnosisPrompt(failures, agentsContent) {
-    const failureSummaries = failures.map((f, i) => {
-        const shortcomings = f.shortcomings.map((s) => `    - ${s}`).join('\n');
-        return `${i + 1}. Probe: ${f.probeId}\n   Shortcomings:\n${shortcomings}`;
-    }).join('\n\n');
-    return `You are diagnosing why an AI coding assistant's AGENTS.md instructions
-are failing to guide it correctly. Below are probe test results showing
-specific scenarios where the AI fell short.
+const llmUtil_1 = require("./llmUtil");
+// ── Prompts ─────────────────────────────────────────────────
+const DIAGNOSE_SYSTEM = `You are an expert context editor for AI coding assistants. You manage AGENTS.md (general guidance) and scoped rules (directory-specific guidance).
 
-## Failed Probes
-${failureSummaries}
+Output a JSON array of edit objects, each with:
+- "section": an AGENTS.md section OR "scoped:slug" to edit a scoped rule OR "scoped:CREATE:slug" to create one OR "scoped:DELETE:slug" to remove one
+- "action": one of "add", "modify", "strengthen", "remove"
+- "content": the specific text (for AGENTS.md: a bullet point; for scoped rules: full markdown body)
+- "globs": (only for scoped:CREATE) array of glob patterns, e.g. ["src/core/**"]
+- "description": (only for scoped:CREATE) short description of the rule
 
-## Current AGENTS.md
-${agentsContent}
+AGENTS.md sections: "Operating Mode", "Procedural Standards", "High-Impact Hubs", "Entry Points", "Import Chains", "Validation", "Integration Risk", "Conventions", "Guardrails", "Setup"
 
-## Task
-Analyze the failures and identify what in AGENTS.md needs to change.
-For each issue, propose a specific, actionable edit.
-
-Respond in EXACTLY this format:
+Rules:
+- STRONGLY prefer editing AGENTS.md over creating scoped rules. Scoped rules are only for content that is truly directory-specific and would be misleading if applied globally.
+- Do NOT create scoped rules for naming conventions alone — that belongs in AGENTS.md.
+- Keep edits specific and actionable. Write direct imperatives.
+- Content must be general enough to apply broadly, not tied to one probe scenario.
+- Use "modify"/"strengthen" to refine existing guidance before adding new rules.
+- You may delete scoped rules that are redundant, trivial, or already covered by AGENTS.md.
+- Edits are optional: return [] when guidance is already strong.
+- Return at most 8 edits total.
+- Keep AGENTS.md under 8,000 characters.
+- Output ONLY the JSON array.`;
+function buildDiagnoseUserPrompt(agentsMd, results, scopedRulesContext, staticAnalysisData) {
+    const diagnostics = results.map((r, i) => {
+        const reviews = r.behaviorReviews
+            .map((br) => `  * Behavior: ${br.behavior} | Assessment: ${br.assessment} | Evidence: ${br.evidence} | Improvement: ${br.improvement}`)
+            .join('\n');
+        const edits = r.proposedEdits
+            .map((e) => `  * ProposedEdit: ${e.action}@${e.section}: ${e.content}`)
+            .join('\n');
+        return `- Probe ${i + 1}: ${r.task}\n${reviews}\n  * Overall: ${r.overallNotes}\n${edits}`;
+    }).join('\n\n');
+    let prompt = `CURRENT AGENTS.MD:
+---
+${agentsMd}
+---`;
+    if (scopedRulesContext) {
+        prompt += `
 
-SUMMARY: <one paragraph overview of what's wrong>
+CURRENT SCOPED RULES:
+---
+${scopedRulesContext}
+---`;
+    }
+    if (staticAnalysisData) {
+        prompt += `
 
-EDIT_1:
-SECTION: <which section/area of AGENTS.md>
-ACTION: add|modify|strengthen|remove
-CONTENT: <the proposed rule or change>
-MOTIVATED_BY: <comma-separated probe IDs>
+STATIC ANALYSIS DATA:
+${staticAnalysisData}`;
+    }
+    prompt += `
 
-EDIT_2:
-SECTION: ...
-ACTION: ...
-CONTENT: ...
-MOTIVATED_BY: ...
+PROBE DIAGNOSTICS:
+${diagnostics}
 
-(continue for each edit needed)`;
+Propose edits to improve the guidance. You may edit AGENTS.md sections, create/update/delete scoped rules, or return [] if no changes needed.`;
+    return prompt;
 }
-/** Parse the structured diagnosis response. */
-function parseDiagnosisResponse(response, failureCount) {
-    // Parse summary
-    const summaryMatch = response.match(/SUMMARY:\s*(.+?)(?=\n\s*EDIT_\d|$)/is);
-    const summary = summaryMatch ? summaryMatch[1].trim() : 'Could not parse diagnosis summary.';
-    // Parse edits
-    const edits = [];
-    const editRegex = /EDIT_\d+:\s*\nSECTION:\s*(.+?)\nACTION:\s*(add|modify|strengthen|remove)\s*\nCONTENT:\s*(.+?)\nMOTIVATED_BY:\s*(.+?)(?=\n\s*EDIT_\d|$)/gis;
-    let match;
-    while ((match = editRegex.exec(response)) !== null) {
-        edits.push({
-            section: match[1].trim(),
-            action: match[2].trim().toLowerCase(),
-            content: match[3].trim(),
-            motivatedBy: match[4].trim().split(/,\s*/).filter(Boolean),
-        });
+function parseDiagnoseResponse(raw) {
+    // Strip thinking tags
+    let cleaned = raw.replace(/<think>[\s\S]*?<\/think>/g, '').trim();
+    // Strip code fences
+    cleaned = cleaned.replace(/^```(?:json)?\s*\n?/m, '').replace(/\n?```\s*$/m, '').trim();
+    try {
+        const parsed = JSON.parse(cleaned);
+        if (Array.isArray(parsed))
+            return parsed;
     }
-    // Fallback: if regex didn't match cleanly, try a looser parse
-    if (edits.length === 0) {
-        const looseEditRegex = /SECTION:\s*(.+?)\n.*?ACTION:\s*(\w+)\n.*?CONTENT:\s*(.+?)\n.*?MOTIVATED_BY:\s*(.+?)(?=\n\s*(?:EDIT|SECTION)|$)/gis;
-        while ((match = looseEditRegex.exec(response)) !== null) {
-            const action = match[2].trim().toLowerCase();
-            if (['add', 'modify', 'strengthen', 'remove'].includes(action)) {
-                edits.push({
-                    section: match[1].trim(),
-                    action: action,
-                    content: match[3].trim(),
-                    motivatedBy: match[4].trim().split(/,\s*/).filter(Boolean),
-                });
+    catch {
+        // Try to find JSON array
+        const match = cleaned.match(/\[[\s\S]*\]/);
+        if (match) {
+            try {
+                return JSON.parse(match[0]);
+            }
+            catch {
+                // fall through
             }
         }
     }
-    return { edits, summary, failureCount };
+    return [];
 }
+// ── Public API ──────────────────────────────────────────────
 /**
- * Diagnose AGENTS.md shortcomings from failed probe results.
+ * Diagnose AGENTS.md shortcomings from judged probe results.
  *
- * Sends the failures + current AGENTS.md to the LLM and asks it to
- * identify what needs to change and propose specific edits.
+ * Aggregates all probe behavior reviews and per-probe edits,
+ * then proposes up to 6 aggregate edits via a single LLM call.
  */
-async function diagnose(failures, agentsContent, provider, log, signal) {
-    if (failures.length === 0) {
-        return { edits: [], summary: 'All probes passed.', failureCount: 0 };
-    }
-    if (signal?.aborted) {
-        return { edits: [], summary: 'Cancelled.', failureCount: failures.length };
+async function diagnose(options) {
+    const { judgedResults, agentsContent, provider, log, signal, scopedRulesContext, staticAnalysisData } = options;
+    if (judgedResults.length === 0)
+        return [];
+    if (signal?.aborted)
+        return [];
+    // Only send probes that have non-strong behaviours
+    const weakResults = judgedResults.filter((r) => r.behaviorReviews.some((br) => br.assessment !== 'strong'));
+    if (weakResults.length === 0) {
+        log?.info('All probes assessed as strong — no diagnosis needed.');
+        return [];
     }
-    log?.info(`Diagnosing ${failures.length} probe failure${failures.length === 1 ? '' : 's'}…`);
-    const prompt = buildDiagnosisPrompt(failures, agentsContent);
+    log?.info(`Diagnosing ${weakResults.length} probe result${weakResults.length === 1 ? '' : 's'} with weak behaviors…`);
+    const userPrompt = buildDiagnoseUserPrompt(agentsContent, weakResults, scopedRulesContext, staticAnalysisData);
     const messages = [
-        { role: 'user', content: prompt },
+        { role: 'system', content: DIAGNOSE_SYSTEM },
+        { role: 'user', content: userPrompt },
     ];
     let response;
     try {
-        response = await provider.chat(messages);
+        response = await (0, llmUtil_1.chatWithTemp)(provider, messages, 0.0, signal);
     }
     catch (err) {
         const msg = err instanceof Error ? err.message : String(err);
         log?.error(`Diagnosis LLM call failed: ${msg}`);
-        return {
-            edits: [],
-            summary: `Diagnosis failed: ${msg}`,
-            failureCount: failures.length,
-        };
-    }
-    const diagnosis = parseDiagnosisResponse(response, failures.length);
-    log?.info(`Diagnosis: ${diagnosis.edits.length} edit${diagnosis.edits.length === 1 ? '' : 's'} proposed`);
-    return diagnosis;
-}
-/**
- * Apply diagnosis edits to AGENTS.md content.
- *
- * Uses the LLM to intelligently merge the proposed edits into the
- * existing content, since edits reference sections by name (not line number).
- */
-async function applyDiagnosisEdits(agentsContent, diagnosis, provider, log, signal) {
-    if (diagnosis.edits.length === 0) {
-        return { content: agentsContent, appliedEdits: [] };
-    }
-    if (signal?.aborted) {
-        return { content: agentsContent, appliedEdits: [] };
-    }
-    const editDescriptions = diagnosis.edits
-        .map((e, i) => `${i + 1}. [${e.action.toUpperCase()}] Section "${e.section}": ${e.content}`)
-        .join('\n');
-    const prompt = `Apply the following edits to the AGENTS.md instructions.
-Each edit specifies a section, an action (add/modify/strengthen/remove), and content.
-
-## Edits to Apply
-${editDescriptions}
-
-## Current AGENTS.md
-${agentsContent}
-
-## Rules
-- Apply ALL edits.
-- Keep the same overall structure unless an edit requires restructuring.
-- For "add": insert the new rule in the appropriate section.
-- For "modify": find and replace the relevant rule.
-- For "strengthen": make the existing rule more specific/forceful.
-- For "remove": delete the rule.
-- AGENTS.md must remain fully self-contained — no references to external documents.
-- Output ONLY the full updated AGENTS.md content (no explanations or fences).`;
-    const messages = [
-        { role: 'user', content: prompt },
-    ];
-    let response;
-    try {
-        response = await provider.chat(messages);
-    }
-    catch (err) {
-        const msg = err instanceof Error ? err.message : String(err);
-        log?.error(`Edit application failed: ${msg}`);
-        return { content: agentsContent, appliedEdits: [] };
+        return [];
     }
-    const appliedEdits = diagnosis.edits.map((e) => `[${e.action}] ${e.section}: ${e.content}`);
-    log?.info(`Applied ${appliedEdits.length} edit${appliedEdits.length === 1 ? '' : 's'} to AGENTS.md`);
-    return { content: response.trim(), appliedEdits };
+    const rawEdits = parseDiagnoseResponse(response);
+    const edits = rawEdits
+        .slice(0, 8)
+        .filter((e) => e.section && e.action)
+        .map((e) => ({
+        section: e.section,
+        action: (['add', 'modify', 'strengthen', 'remove'].includes(e.action)
+            ? e.action
+            : 'add'),
+        content: e.content || '',
+        globs: e.globs,
+        description: e.description,
+    }));
+    const agentsEdits = edits.filter((e) => !e.section.startsWith('scoped:'));
+    const scopedEdits = edits.filter((e) => e.section.startsWith('scoped:'));
+    log?.info(`Diagnosis: ${agentsEdits.length} AGENTS.md edit${agentsEdits.length === 1 ? '' : 's'}, ${scopedEdits.length} scoped rule edit${scopedEdits.length === 1 ? '' : 's'}`);
+    return edits;
 }
 //# sourceMappingURL=diagnosis.js.map

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

@@ -1 +1 @@
-{"version":3,"file":"diagnosis.js","sourceRoot":"","sources":["../src/diagnosis.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;AA+FH,4BAuCC;AAQD,kDA0DC;AAGQ,oDAAoB;AAAE,wDAAsB;AAtMrD;;GAEG;AACH,SAAS,oBAAoB,CAC3B,QAAuB,EACvB,aAAqB;IAErB,MAAM,gBAAgB,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QAC7C,MAAM,YAAY,GAAG,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACxE,OAAO,GAAG,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,OAAO,uBAAuB,YAAY,EAAE,CAAC;IAC5E,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAEhB,OAAO;;;;;EAKP,gBAAgB;;;EAGhB,aAAa;;;;;;;;;;;;;;;;;;;;;;gCAsBiB,CAAC;AACjC,CAAC;AAED,+CAA+C;AAC/C,SAAS,sBAAsB,CAAC,QAAgB,EAAE,YAAoB;IACpE,gBAAgB;IAChB,MAAM,YAAY,GAAG,QAAQ,CAAC,KAAK,CAAC,sCAAsC,CAAC,CAAC;IAC5E,MAAM,OAAO,GAAG,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,oCAAoC,CAAC;IAE7F,cAAc;IACd,MAAM,KAAK,GAAiB,EAAE,CAAC;IAC/B,MAAM,SAAS,GAAG,2IAA2I,CAAC;IAE9J,IAAI,KAA6B,CAAC;IAClC,OAAO,CAAC,KAAK,GAAG,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;QACnD,KAAK,CAAC,IAAI,CAAC;YACT,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE;YACxB,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAA0B;YAC7D,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE;YACxB,WAAW,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC;SAC3D,CAAC,CAAC;IACL,CAAC;IAED,8DAA8D;IAC9D,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,MAAM,cAAc,GAAG,mHAAmH,CAAC;QAC3I,OAAO,CAAC,KAAK,GAAG,cAAc,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;YACxD,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;YAC7C,IAAI,CAAC,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,QAAQ,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;gBAC/D,KAAK,CAAC,IAAI,CAAC;oBACT,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE;oBACxB,MAAM,EAAE,MAA8B;oBACtC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE;oBACxB,WAAW,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC;iBAC3D,CAAC,CAAC;YACL,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,CAAC;AAC1C,CAAC;AAED;;;;;GAKG;AACI,KAAK,UAAU,QAAQ,CAC5B,QAAuB,EACvB,aAAqB,EACrB,QAAqB,EACrB,GAAe,EACf,MAAoB;IAEpB,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,EAAE,KAAK,EAAE,EAAE,EAAE,OAAO,EAAE,oBAAoB,EAAE,YAAY,EAAE,CAAC,EAAE,CAAC;IACvE,CAAC;IAED,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,OAAO,EAAE,KAAK,EAAE,EAAE,EAAE,OAAO,EAAE,YAAY,EAAE,YAAY,EAAE,QAAQ,CAAC,MAAM,EAAE,CAAC;IAC7E,CAAC;IAED,GAAG,EAAE,IAAI,CAAC,cAAc,QAAQ,CAAC,MAAM,iBAAiB,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC;IAE7F,MAAM,MAAM,GAAG,oBAAoB,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;IAC7D,MAAM,QAAQ,GAAkB;QAC9B,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE;KAClC,CAAC;IAEF,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC3C,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,KAAK,CAAC,8BAA8B,GAAG,EAAE,CAAC,CAAC;QAChD,OAAO;YACL,KAAK,EAAE,EAAE;YACT,OAAO,EAAE,qBAAqB,GAAG,EAAE;YACnC,YAAY,EAAE,QAAQ,CAAC,MAAM;SAC9B,CAAC;IACJ,CAAC;IAED,MAAM,SAAS,GAAG,sBAAsB,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IACpE,GAAG,EAAE,IAAI,CAAC,cAAc,SAAS,CAAC,KAAK,CAAC,MAAM,QAAQ,SAAS,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC;IAE1G,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;;;;GAKG;AACI,KAAK,UAAU,mBAAmB,CACvC,aAAqB,EACrB,SAAoB,EACpB,QAAqB,EACrB,GAAe,EACf,MAAoB;IAEpB,IAAI,SAAS,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO,EAAE,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,EAAE,EAAE,CAAC;IACtD,CAAC;IAED,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,OAAO,EAAE,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,EAAE,EAAE,CAAC;IACtD,CAAC;IAED,MAAM,gBAAgB,GAAG,SAAS,CAAC,KAAK;SACrC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC,OAAO,MAAM,CAAC,CAAC,OAAO,EAAE,CAAC;SAC3F,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,MAAM,MAAM,GAAG;;;;EAIf,gBAAgB;;;EAGhB,aAAa;;;;;;;;;;8EAU+D,CAAC;IAE7E,MAAM,QAAQ,GAAkB;QAC9B,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE;KAClC,CAAC;IAEF,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC3C,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,KAAK,CAAC,4BAA4B,GAAG,EAAE,CAAC,CAAC;QAC9C,OAAO,EAAE,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,EAAE,EAAE,CAAC;IACtD,CAAC;IAED,MAAM,YAAY,GAAG,SAAS,CAAC,KAAK,CAAC,GAAG,CACtC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,EAAE,CAClD,CAAC;IAEF,GAAG,EAAE,IAAI,CAAC,WAAW,YAAY,CAAC,MAAM,QAAQ,YAAY,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,CAAC;IAErG,OAAO,EAAE,OAAO,EAAE,QAAQ,CAAC,IAAI,EAAE,EAAE,YAAY,EAAE,CAAC;AACpD,CAAC"}
+{"version":3,"file":"diagnosis.js","sourceRoot":"","sources":["../src/diagnosis.ts"],"names":[],"mappings":";AAAA;;;;;;;GAOG;;AA4FH,sDAuBC;AAUD,4BAoDC;AAzKD,uCAAyC;AAEzC,+DAA+D;AAE/D,MAAM,eAAe,GAAG;;;;;;;;;;;;;;;;;;;;;8BAqBM,CAAC;AAE/B,SAAS,uBAAuB,CAC9B,QAAgB,EAChB,OAA4B,EAC5B,kBAA2B,EAC3B,kBAA2B;IAE3B,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACvC,MAAM,OAAO,GAAG,CAAC,CAAC,eAAe;aAC9B,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CACV,iBAAiB,EAAE,CAAC,QAAQ,kBAAkB,EAAE,CAAC,UAAU,gBAAgB,EAAE,CAAC,QAAQ,mBAAmB,EAAE,CAAC,WAAW,EAAE,CAC1H;aACA,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa;aAC1B,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,qBAAqB,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC;aACtE,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,OAAO,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,KAAK,OAAO,kBAAkB,CAAC,CAAC,YAAY,KAAK,KAAK,EAAE,CAAC;IAC7F,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAEhB,IAAI,MAAM,GAAG;;EAEb,QAAQ;IACN,CAAC;IAEH,IAAI,kBAAkB,EAAE,CAAC;QACvB,MAAM,IAAI;;;;EAIZ,kBAAkB;IAChB,CAAC;IACH,CAAC;IAED,IAAI,kBAAkB,EAAE,CAAC;QACvB,MAAM,IAAI;;;EAGZ,kBAAkB,EAAE,CAAC;IACrB,CAAC;IAED,MAAM,IAAI;;;EAGV,WAAW;;8IAEiI,CAAC;IAE7I,OAAO,MAAM,CAAC;AAChB,CAAC;AAUD,SAAgB,qBAAqB,CAAC,GAAW;IAC/C,sBAAsB;IACtB,IAAI,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,2BAA2B,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAElE,oBAAoB;IACpB,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,sBAAsB,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAExF,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;YAAE,OAAO,MAAmB,CAAC;IACxD,CAAC;IAAC,MAAM,CAAC;QACP,yBAAyB;QACzB,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;QAC3C,IAAI,KAAK,EAAE,CAAC;YACV,IAAI,CAAC;gBACH,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAc,CAAC;YAC3C,CAAC;YAAC,MAAM,CAAC;gBACP,eAAe;YACjB,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,+DAA+D;AAE/D;;;;;GAKG;AACI,KAAK,UAAU,QAAQ,CAAC,OAAyB;IACtD,MAAM,EAAE,aAAa,EAAE,aAAa,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,GAAG,OAAO,CAAC;IAEhH,IAAI,aAAa,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAC1C,IAAI,MAAM,EAAE,OAAO;QAAE,OAAO,EAAE,CAAC;IAE/B,mDAAmD;IACnD,MAAM,WAAW,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAC7C,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,UAAU,KAAK,QAAQ,CAAC,CAC3D,CAAC;IAEF,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7B,GAAG,EAAE,IAAI,CAAC,sDAAsD,CAAC,CAAC;QAClE,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,GAAG,EAAE,IAAI,CAAC,cAAc,WAAW,CAAC,MAAM,gBAAgB,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,uBAAuB,CAAC,CAAC;IAEtH,MAAM,UAAU,GAAG,uBAAuB,CAAC,aAAa,EAAE,WAAW,EAAE,kBAAkB,EAAE,kBAAkB,CAAC,CAAC;IAC/G,MAAM,QAAQ,GAAkB;QAC9B,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,eAAe,EAAE;QAC5C,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,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,KAAK,CAAC,8BAA8B,GAAG,EAAE,CAAC,CAAC;QAChD,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,MAAM,QAAQ,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;IAEjD,MAAM,KAAK,GAAiB,QAAQ;SACjC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;SACX,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,MAAM,CAAC;SACpC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACX,OAAO,EAAE,CAAC,CAAC,OAAO;QAClB,MAAM,EAAE,CAAC,CAAC,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,QAAQ,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC;YACnE,CAAC,CAAC,CAAC,CAAC,MAAM;YACV,CAAC,CAAC,KAAK,CAAyB;QAClC,OAAO,EAAE,CAAC,CAAC,OAAO,IAAI,EAAE;QACxB,KAAK,EAAG,CAAS,CAAC,KAAK;QACvB,WAAW,EAAG,CAAS,CAAC,WAAW;KACpC,CAAC,CAAC,CAAC;IAEN,MAAM,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC;IAC1E,MAAM,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC;IACzE,GAAG,EAAE,IAAI,CAAC,cAAc,WAAW,CAAC,MAAM,kBAAkB,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,WAAW,CAAC,MAAM,oBAAoB,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IACjL,OAAO,KAAK,CAAC;AACf,CAAC"}

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

@@ -1,49 +1,14 @@
 /**
- * @aspectcode/evaluator — Evidence-based evaluation for AGENTS.md optimization.
+ * @aspectcode/evaluator — Probe-and-refine tuning for AGENTS.md.
  *
- * Replaces arbitrary LLM self-scoring with:
- * 1. Probe-based micro-tests scoped to the knowledge base
- * 2. Prompt history harvesting from AI coding tools
- * 3. Evidence-driven diagnosis and AGENTS.md improvement
- *
- * @example
- * ```ts
- * import { generateProbes, runProbes, diagnose, harvestPrompts } from '@aspectcode/evaluator';
- *
- * const harvested = await harvestPrompts({ root });
- * const probes = generateProbes({ kb, harvestedPrompts: harvested });
- * const results = await runProbes({ agentsContent, probes, provider });
- * const failures = results.filter(r => !r.passed);
- * if (failures.length > 0) {
- *   const diagnosis = await diagnose({ failures, agentsContent, provider });
- * }
- * ```
+ * Multi-iteration loop: LLM-generated probes, per-probe judging,
+ * aggregate diagnosis, and deterministic edit application.
  */
-export type { Probe, ProbeCategory, ProbeResult, BehaviorResult, Diagnosis, AgentsEdit, HarvestedPrompt, PromptSource, ProbeGeneratorOptions, ProbeRunnerOptions, ProbeEvaluatorOptions, DiagnosisOptions, HarvestOptions, EvaluationResult, LlmProvider, OptLogger, } from './types';
+export type { Probe, SimulationResult, BehaviorReview, JudgedProbeResult, AgentsEdit, ProbeGeneratorOptions, ProbeRunnerOptions, JudgeOptions, DiagnosisOptions, ProbeRefineConfig, ProbeRefineResult, IterationSummary, ApplyResult, ProbeProgressCallback, LlmProvider, ChatOptions, OptLogger, } from './types';
+export { DEFAULT_PROBE_REFINE_CONFIG } from './types';
 export { generateProbes } from './probes';
 export { runProbes } from './runner';
-export type { ProbeProgressCallback } from './runner';
-export { diagnose, applyDiagnosisEdits } from './diagnosis';
-export { harvestPrompts, harvestAider, harvestClaudeCode, harvestCline, harvestCopilotChat, harvestCursor, harvestWindsurf, harvestExport, } from './harvest/index';
-import type { EvaluationResult, ProbeRunnerOptions, DiagnosisOptions, ProbeGeneratorOptions } from './types';
-/**
- * Run the full evaluation pipeline: generate probes → run them → diagnose failures.
- *
- * This is a convenience function combining the individual steps.
- * For more control, use the individual functions directly.
- */
-export declare function evaluate(options: {
-    /** Probe generation options. */
-    probeOptions: ProbeGeneratorOptions;
-    /** Current AGENTS.md content. */
-    agentsContent: string;
-    /** LLM provider for probe execution and diagnosis. */
-    provider: ProbeRunnerOptions['provider'];
-    /** File contents for context. */
-    fileContents?: ReadonlyMap<string, string>;
-    /** Logger. */
-    log?: DiagnosisOptions['log'];
-    /** Abort signal. */
-    signal?: AbortSignal;
-}): Promise<EvaluationResult>;
+export { judgeProbe } from './judge';
+export { diagnose } from './diagnosis';
+export { applyEdits, applyEditsWithLlm, AGENTS_MD_CHAR_BUDGET } from './apply';
 //# sourceMappingURL=index.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAGH,YAAY,EACV,KAAK,EACL,aAAa,EACb,WAAW,EACX,cAAc,EACd,SAAS,EACT,UAAU,EACV,eAAe,EACf,YAAY,EACZ,qBAAqB,EACrB,kBAAkB,EAClB,qBAAqB,EACrB,gBAAgB,EAChB,cAAc,EACd,gBAAgB,EAChB,WAAW,EACX,SAAS,GACV,MAAM,SAAS,CAAC;AAGjB,OAAO,EAAE,cAAc,EAAE,MAAM,UAAU,CAAC;AAG1C,OAAO,EAAE,SAAS,EAAE,MAAM,UAAU,CAAC;AACrC,YAAY,EAAE,qBAAqB,EAAE,MAAM,UAAU,CAAC;AAGtD,OAAO,EAAE,QAAQ,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAG5D,OAAO,EACL,cAAc,EACd,YAAY,EACZ,iBAAiB,EACjB,YAAY,EACZ,kBAAkB,EAClB,aAAa,EACb,eAAe,EACf,aAAa,GACd,MAAM,iBAAiB,CAAC;AAIzB,OAAO,KAAK,EAAE,gBAAgB,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAC;AAK7G;;;;;GAKG;AACH,wBAAsB,QAAQ,CAAC,OAAO,EAAE;IACtC,gCAAgC;IAChC,YAAY,EAAE,qBAAqB,CAAC;IACpC,iCAAiC;IACjC,aAAa,EAAE,MAAM,CAAC;IACtB,sDAAsD;IACtD,QAAQ,EAAE,kBAAkB,CAAC,UAAU,CAAC,CAAC;IACzC,iCAAiC;IACjC,YAAY,CAAC,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC3C,cAAc;IACd,GAAG,CAAC,EAAE,gBAAgB,CAAC,KAAK,CAAC,CAAC;IAC9B,oBAAoB;IACpB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAqC5B"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,YAAY,EACV,KAAK,EACL,gBAAgB,EAChB,cAAc,EACd,iBAAiB,EACjB,UAAU,EACV,qBAAqB,EACrB,kBAAkB,EAClB,YAAY,EACZ,gBAAgB,EAChB,iBAAiB,EACjB,iBAAiB,EACjB,gBAAgB,EAChB,WAAW,EACX,qBAAqB,EACrB,WAAW,EACX,WAAW,EACX,SAAS,GACV,MAAM,SAAS,CAAC;AAEjB,OAAO,EAAE,2BAA2B,EAAE,MAAM,SAAS,CAAC;AAGtD,OAAO,EAAE,cAAc,EAAE,MAAM,UAAU,CAAC;AAG1C,OAAO,EAAE,SAAS,EAAE,MAAM,UAAU,CAAC;AAGrC,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAGrC,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAGvC,OAAO,EAAE,UAAU,EAAE,iBAAiB,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAC"}

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

@@ -1,75 +1,29 @@
 "use strict";
 /**
- * @aspectcode/evaluator — Evidence-based evaluation for AGENTS.md optimization.
+ * @aspectcode/evaluator — Probe-and-refine tuning for AGENTS.md.
  *
- * Replaces arbitrary LLM self-scoring with:
- * 1. Probe-based micro-tests scoped to the knowledge base
- * 2. Prompt history harvesting from AI coding tools
- * 3. Evidence-driven diagnosis and AGENTS.md improvement
- *
- * @example
- * ```ts
- * import { generateProbes, runProbes, diagnose, harvestPrompts } from '@aspectcode/evaluator';
- *
- * const harvested = await harvestPrompts({ root });
- * const probes = generateProbes({ kb, harvestedPrompts: harvested });
- * const results = await runProbes({ agentsContent, probes, provider });
- * const failures = results.filter(r => !r.passed);
- * if (failures.length > 0) {
- *   const diagnosis = await diagnose({ failures, agentsContent, provider });
- * }
- * ```
+ * Multi-iteration loop: LLM-generated probes, per-probe judging,
+ * aggregate diagnosis, and deterministic edit application.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.harvestExport = exports.harvestWindsurf = exports.harvestCursor = exports.harvestCopilotChat = exports.harvestCline = exports.harvestClaudeCode = exports.harvestAider = exports.harvestPrompts = exports.applyDiagnosisEdits = exports.diagnose = exports.runProbes = exports.generateProbes = void 0;
-exports.evaluate = evaluate;
+exports.AGENTS_MD_CHAR_BUDGET = exports.applyEditsWithLlm = exports.applyEdits = exports.diagnose = exports.judgeProbe = exports.runProbes = exports.generateProbes = exports.DEFAULT_PROBE_REFINE_CONFIG = void 0;
+var types_1 = require("./types");
+Object.defineProperty(exports, "DEFAULT_PROBE_REFINE_CONFIG", { enumerable: true, get: function () { return types_1.DEFAULT_PROBE_REFINE_CONFIG; } });
 // ── Probe generation ────────────────────────────────────────
 var probes_1 = require("./probes");
 Object.defineProperty(exports, "generateProbes", { enumerable: true, get: function () { return probes_1.generateProbes; } });
 // ── Probe execution ─────────────────────────────────────────
 var runner_1 = require("./runner");
 Object.defineProperty(exports, "runProbes", { enumerable: true, get: function () { return runner_1.runProbes; } });
-// ── Diagnosis ───────────────────────────────────────────────
+// ── Per-probe judging ───────────────────────────────────────
+var judge_1 = require("./judge");
+Object.defineProperty(exports, "judgeProbe", { enumerable: true, get: function () { return judge_1.judgeProbe; } });
+// ── Aggregate diagnosis ─────────────────────────────────────
 var diagnosis_1 = require("./diagnosis");
 Object.defineProperty(exports, "diagnose", { enumerable: true, get: function () { return diagnosis_1.diagnose; } });
-Object.defineProperty(exports, "applyDiagnosisEdits", { enumerable: true, get: function () { return diagnosis_1.applyDiagnosisEdits; } });
-// ── Prompt harvesting ───────────────────────────────────────
-var index_1 = require("./harvest/index");
-Object.defineProperty(exports, "harvestPrompts", { enumerable: true, get: function () { return index_1.harvestPrompts; } });
-Object.defineProperty(exports, "harvestAider", { enumerable: true, get: function () { return index_1.harvestAider; } });
-Object.defineProperty(exports, "harvestClaudeCode", { enumerable: true, get: function () { return index_1.harvestClaudeCode; } });
-Object.defineProperty(exports, "harvestCline", { enumerable: true, get: function () { return index_1.harvestCline; } });
-Object.defineProperty(exports, "harvestCopilotChat", { enumerable: true, get: function () { return index_1.harvestCopilotChat; } });
-Object.defineProperty(exports, "harvestCursor", { enumerable: true, get: function () { return index_1.harvestCursor; } });
-Object.defineProperty(exports, "harvestWindsurf", { enumerable: true, get: function () { return index_1.harvestWindsurf; } });
-Object.defineProperty(exports, "harvestExport", { enumerable: true, get: function () { return index_1.harvestExport; } });
-const probes_2 = require("./probes");
-const runner_2 = require("./runner");
-const diagnosis_2 = require("./diagnosis");
-/**
- * Run the full evaluation pipeline: generate probes → run them → diagnose failures.
- *
- * This is a convenience function combining the individual steps.
- * For more control, use the individual functions directly.
- */
-async function evaluate(options) {
-    const { probeOptions, agentsContent, provider, fileContents, log, signal } = options;
-    // Step 1: Generate probes
-    const probes = (0, probes_2.generateProbes)(probeOptions);
-    // Step 2: Run probes
-    const probeResults = await (0, runner_2.runProbes)(agentsContent, probes, provider, fileContents, log, signal);
-    const failures = probeResults.filter((r) => !r.passed);
-    // Step 3: Diagnose if there are failures
-    let diagnosis;
-    if (failures.length > 0) {
-        diagnosis = await (0, diagnosis_2.diagnose)(failures, agentsContent, provider, log, signal);
-    }
-    return {
-        probeResults,
-        diagnosis,
-        passCount: probeResults.length - failures.length,
-        failCount: failures.length,
-        totalProbes: probeResults.length,
-    };
-}
+// ── Edit application ───────────────────────────────────────
+var apply_1 = require("./apply");
+Object.defineProperty(exports, "applyEdits", { enumerable: true, get: function () { return apply_1.applyEdits; } });
+Object.defineProperty(exports, "applyEditsWithLlm", { enumerable: true, get: function () { return apply_1.applyEditsWithLlm; } });
+Object.defineProperty(exports, "AGENTS_MD_CHAR_BUDGET", { enumerable: true, get: function () { return apply_1.AGENTS_MD_CHAR_BUDGET; } });
 //# sourceMappingURL=index.js.map

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

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;;;AAyDH,4BAkDC;AArFD,+DAA+D;AAC/D,mCAA0C;AAAjC,wGAAA,cAAc,OAAA;AAEvB,+DAA+D;AAC/D,mCAAqC;AAA5B,mGAAA,SAAS,OAAA;AAGlB,+DAA+D;AAC/D,yCAA4D;AAAnD,qGAAA,QAAQ,OAAA;AAAE,gHAAA,mBAAmB,OAAA;AAEtC,+DAA+D;AAC/D,yCASyB;AARvB,uGAAA,cAAc,OAAA;AACd,qGAAA,YAAY,OAAA;AACZ,0GAAA,iBAAiB,OAAA;AACjB,qGAAA,YAAY,OAAA;AACZ,2GAAA,kBAAkB,OAAA;AAClB,sGAAA,aAAa,OAAA;AACb,wGAAA,eAAe,OAAA;AACf,sGAAA,aAAa,OAAA;AAMf,qCAA0C;AAC1C,qCAAqC;AACrC,2CAAuC;AAEvC;;;;;GAKG;AACI,KAAK,UAAU,QAAQ,CAAC,OAa9B;IACC,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,QAAQ,EAAE,YAAY,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;IAErF,0BAA0B;IAC1B,MAAM,MAAM,GAAG,IAAA,uBAAc,EAAC,YAAY,CAAC,CAAC;IAE5C,qBAAqB;IACrB,MAAM,YAAY,GAAG,MAAM,IAAA,kBAAS,EAClC,aAAa,EACb,MAAM,EACN,QAAQ,EACR,YAAY,EACZ,GAAG,EACH,MAAM,CACP,CAAC;IAEF,MAAM,QAAQ,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IAEvD,yCAAyC;IACzC,IAAI,SAAS,CAAC;IACd,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,SAAS,GAAG,MAAM,IAAA,oBAAQ,EACxB,QAAQ,EACR,aAAa,EACb,QAAQ,EACR,GAAG,EACH,MAAM,CACP,CAAC;IACJ,CAAC;IAED,OAAO;QACL,YAAY;QACZ,SAAS;QACT,SAAS,EAAE,YAAY,CAAC,MAAM,GAAG,QAAQ,CAAC,MAAM;QAChD,SAAS,EAAE,QAAQ,CAAC,MAAM;QAC1B,WAAW,EAAE,YAAY,CAAC,MAAM;KACjC,CAAC;AACJ,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;AAuBH,iCAAsD;AAA7C,oHAAA,2BAA2B,OAAA;AAEpC,+DAA+D;AAC/D,mCAA0C;AAAjC,wGAAA,cAAc,OAAA;AAEvB,+DAA+D;AAC/D,mCAAqC;AAA5B,mGAAA,SAAS,OAAA;AAElB,+DAA+D;AAC/D,iCAAqC;AAA5B,mGAAA,UAAU,OAAA;AAEnB,+DAA+D;AAC/D,yCAAuC;AAA9B,qGAAA,QAAQ,OAAA;AAEjB,8DAA8D;AAC9D,iCAA+E;AAAtE,mGAAA,UAAU,OAAA;AAAE,0GAAA,iBAAiB,OAAA;AAAE,8GAAA,qBAAqB,OAAA"}

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

@@ -0,0 +1,32 @@
+/**
+ * Per-probe judge — evaluates AI responses with strong/partial/missing assessments.
+ *
+ * For each probe, the judge reviews the simulated response against expected
+ * behaviours and proposes targeted AGENTS.md edits.
+ *
+ * Ported from sweagent_bench oracle/judge.py.
+ */
+import type { JudgedProbeResult, JudgeOptions } from './types';
+export interface JudgeResponse {
+    behavior_reviews: Array<{
+        behavior: string;
+        assessment: string;
+        evidence: string;
+        improvement: string;
+    }>;
+    proposed_edits: Array<{
+        section: string;
+        action: string;
+        content: string;
+    }>;
+    overall_notes: string;
+}
+export declare function parseJudgeResponse(raw: string): JudgeResponse | null;
+/**
+ * Judge a single probe's response against expected behaviours.
+ *
+ * Returns structured assessments (strong/partial/missing) and
+ * up to 3 proposed AGENTS.md edits.
+ */
+export declare function judgeProbe(options: JudgeOptions): Promise<JudgedProbeResult>;
+//# sourceMappingURL=judge.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../src/judge.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,KAAK,EACV,iBAAiB,EAGjB,YAAY,EACb,MAAM,SAAS,CAAC;AAgDjB,MAAM,WAAW,aAAa;IAC5B,gBAAgB,EAAE,KAAK,CAAC;QACtB,QAAQ,EAAE,MAAM,CAAC;QACjB,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,MAAM,CAAC;QACjB,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC,CAAC;IACH,cAAc,EAAE,KAAK,CAAC;QACpB,OAAO,EAAE,MAAM,CAAC;QAChB,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC,CAAC;IACH,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,aAAa,GAAG,IAAI,CAqBpE;AAID;;;;;GAKG;AACH,wBAAsB,UAAU,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,iBAAiB,CAAC,CA6FlF"}