newtype-profile 1.0.9 → 1.0.11

package/dist/cli/index.js

@@ -2253,7 +2253,7 @@ var require_picocolors = __commonJS((exports, module) => {
 var require_package = __commonJS((exports, module) => {
   module.exports = {
     name: "newtype-profile",
-    version: "1.0.9",
+    version: "1.0.11",
     description: "AI Agent Collaboration System for Content Creation - Based on oh-my-opencode",
     main: "dist/index.js",
     types: "dist/index.d.ts",

package/dist/hooks/chief-orchestrator/confidence-router.d.ts

@@ -0,0 +1,32 @@
+/**
+ * Confidence-based routing for fact-check results
+ *
+ * Parses confidence scores from fact-checker output and generates
+ * routing recommendations for the orchestrator.
+ */
+export interface ConfidenceResult {
+    confidence: number | null;
+    recommendation: "pass" | "polish" | "rewrite" | null;
+    directive: string | null;
+}
+/**
+ * Extract confidence score from fact-checker output
+ * Looks for pattern: **CONFIDENCE: X.XX**
+ */
+export declare function extractConfidence(output: string): number | null;
+/**
+ * Determine routing recommendation based on confidence score
+ */
+export declare function getRecommendation(confidence: number): "pass" | "polish" | "rewrite";
+/**
+ * Build routing directive for Chief based on confidence
+ */
+export declare function buildConfidenceDirective(confidence: number, sessionId: string): string;
+/**
+ * Analyze fact-check output and generate routing result
+ */
+export declare function analyzeFactCheckOutput(output: string, sessionId: string): ConfidenceResult;
+/**
+ * Check if output is from a fact-check task
+ */
+export declare function isFactCheckOutput(output: string): boolean;

package/dist/hooks/chief-orchestrator/confidence-router.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/index.js

@@ -22893,6 +22893,77 @@ function clearTrackerForSession(sessionId) {
   sessionTrackers.delete(sessionId);
 }
 
+// src/hooks/chief-orchestrator/confidence-router.ts
+function extractConfidence(output) {
+  const match = output.match(/\*\*CONFIDENCE:\s*(\d+\.?\d*)\*\*/i);
+  if (match) {
+    const value = parseFloat(match[1]);
+    if (!isNaN(value) && value >= 0 && value <= 1) {
+      return value;
+    }
+  }
+  return null;
+}
+function getRecommendation(confidence) {
+  if (confidence >= 0.8) {
+    return "pass";
+  } else if (confidence >= 0.5) {
+    return "polish";
+  } else {
+    return "rewrite";
+  }
+}
+function buildConfidenceDirective(confidence, sessionId) {
+  const recommendation = getRecommendation(confidence);
+  const confidencePercent = Math.round(confidence * 100);
+  switch (recommendation) {
+    case "pass":
+      return `[FACT-CHECK PASSED]
+Confidence: ${confidencePercent}% (HIGH)
+Action: Content verified. Ready for delivery.`;
+    case "polish":
+      return `[FACT-CHECK: NEEDS POLISH]
+Confidence: ${confidencePercent}% (MEDIUM)
+Action: Send to Editor for refinement.
+
+REQUIRED: Call chief_task with:
+  category="editing"
+  prompt="Polish the content based on fact-check feedback. Address minor uncertainties while preserving verified claims."
+  resume="${sessionId}"`;
+    case "rewrite":
+      return `[FACT-CHECK: NEEDS REWRITE]
+Confidence: ${confidencePercent}% (LOW)
+Action: Significant issues found. Send back to Writer.
+
+REQUIRED: Call chief_task with:
+  category="writing"  
+  prompt="Rewrite the content addressing the fact-check issues. Focus on: [list specific issues from fact-check report]"
+  resume="${sessionId}"
+
+NOTE: Max 2 rewrite attempts. If still failing after 2 rewrites, escalate to user.`;
+  }
+}
+function analyzeFactCheckOutput(output, sessionId) {
+  const confidence = extractConfidence(output);
+  if (confidence === null) {
+    return {
+      confidence: null,
+      recommendation: null,
+      directive: null
+    };
+  }
+  const recommendation = getRecommendation(confidence);
+  const directive = buildConfidenceDirective(confidence, sessionId);
+  return {
+    confidence,
+    recommendation,
+    directive
+  };
+}
+function isFactCheckOutput(output) {
+  return output.includes("CONFIDENCE:") || output.toLowerCase().includes("fact-check") || output.includes("\u6838\u67E5") || output.includes("verification");
+}
+
 // src/hooks/chief-orchestrator/index.ts
 var HOOK_NAME6 = "chief-orchestrator";
 var ALLOWED_PATH_PREFIX2 = ".chief/";
@@ -23465,12 +23536,28 @@ ${buildOrchestratorReminder(boulderState.plan_name, progress, subagentSessionId)
           const category = categoryMatch?.[1];
           const summarized = summarizeOutput(output.output, { category });
           const formattedSummary = formatSummarizedOutput(summarized);
+          let confidenceDirective = "";
+          if (isFactCheckOutput(output.output)) {
+            const confidenceResult = analyzeFactCheckOutput(output.output, subagentSessionId);
+            if (confidenceResult.directive) {
+              confidenceDirective = `
+
+---
+${confidenceResult.directive}
+---`;
+              log(`[${HOOK_NAME6}] Confidence routing detected`, {
+                sessionID: input.sessionID,
+                confidence: confidenceResult.confidence,
+                recommendation: confidenceResult.recommendation
+              });
+            }
+          }
           output.output = `${formattedSummary}
 
 ${progressTable}
 
 ${fileChanges ? `
-${fileChanges}` : ""}
+${fileChanges}` : ""}${confidenceDirective}
 <system-reminder>
 ${buildStandaloneVerificationReminder(subagentSessionId)}
 </system-reminder>`;
@@ -44501,7 +44588,22 @@ Approach:
 - Official documents over media reports
 - Academic/peer-reviewed over informal
 - Note confidence levels for each claim
-</Category_Context>`;
+</Category_Context>
+
+<Output_Format>
+CRITICAL: You MUST end your response with a confidence score in this EXACT format:
+
+---
+**CONFIDENCE: X.XX**
+
+Where X.XX is a number between 0.00 and 1.00:
+- 0.90-1.00: All claims verified with authoritative sources
+- 0.70-0.89: Most claims verified, minor uncertainties
+- 0.50-0.69: Some claims unverified or conflicting sources
+- 0.00-0.49: Significant issues, major claims unverified or false
+
+This score determines whether the content passes review or needs revision.
+</Output_Format>`;
 var ARCHIVE_CATEGORY_PROMPT_APPEND = `<Category_Context>
 You are working on ARCHIVE/KNOWLEDGE-BASE tasks.
 
@@ -44582,11 +44684,11 @@ Approach:
 </Category_Context>`;
 var DEFAULT_CATEGORIES = {
   research: {
-    model: "google/antigravity-gemini-3-pro-high",
+    model: "google/antigravity-gemini-3-flash",
     temperature: 0.5
   },
   "fact-check": {
-    model: "google/antigravity-gemini-3-pro-high",
+    model: "google/antigravity-gemini-3-flash",
     temperature: 0.2
   },
   archive: {
@@ -48911,35 +49013,49 @@ function getModelLimit(state2, providerID, modelID) {
 var DEFAULT_MODEL = "anthropic/claude-opus-4-5";
 function createChiefAgent(model = DEFAULT_MODEL) {
   return {
-    description: "Chief Editor - orchestrates content creation team. Two modes: discussion partner for brainstorming, execution coordinator for deliverables. Final quality gate.",
+    description: "Chief - thought partner for exploration, coordinator for execution. Opinionated, direct, challenges flawed thinking.",
     mode: "primary",
     model,
     temperature: 0.3,
     prompt: `<Role>
-You are "Chief" \u2014 the Chief Editor with two modes of operation.
+You are "Chief" \u2014 a thought partner who happens to have a team.
 
-**Mode 1 - Discussion Partner**: When users want to explore, brainstorm, or think through ideas, you engage directly with strong logical reasoning. You are their intellectual sparring partner.
+**Mode 1 - Thought Partner**: When users want to explore ideas, you think WITH them. You have your own opinions. You challenge flawed logic directly. You're not a facilitator \u2014 you're a sparring partner.
 
-**Mode 2 - Execution Coordinator**: When users have clear deliverables, you decompose tasks, orchestrate your team, and ensure quality.
+**Mode 2 - Execution Coordinator**: When users have clear deliverables, you decompose, delegate, and deliver.
 
-You seamlessly switch between modes based on user intent.
+You switch modes based on user intent.
 </Role>
 
+<Philosophy>
+## How You Think
+- **Go Fundamental**: Dig into principles, not surface-level tools. Why does this work? What's the underlying mechanism?
+- **First-Party Sources First**: Official docs, original papers, author's own words. Be skeptical of second-hand interpretations, especially media reports.
+- **Systematic Thinking**: Everything can become a workflow. Seek reusable patterns, not one-off solutions.
+- **Anti-Fragile by Default**: Multiple backups, no single point of failure, keep options open.
+
+## Your Stance
+- You are NOT a neutral information courier. You have opinions.
+- When you spot a logical flaw, you say it directly: "This has a problem\u2014"
+- You might be wrong. User can push back. That's how good thinking happens.
+- You don't lecture. Your values show through your judgments, not through preaching.
+</Philosophy>
+
 <Core_Capabilities>
-## As Discussion Partner
-1. **Logical Reasoning**: Analyze problems, identify assumptions, spot gaps
-2. **Structured Thinking**: Break fuzzy ideas into clear components
-3. **Socratic Dialogue**: Ask probing questions that deepen understanding
-4. **Devil's Advocate**: Challenge ideas constructively to stress-test them
-5. **Synthesis**: Connect dots, find patterns, propose frameworks
-6. **Silent Delegation**: While discussing, identify research needs and dispatch agents in background \u2014 user doesn't need to know
+## As Thought Partner
+1. **Cut to the Core**: What's the REAL question here? Strip away noise.
+2. **Find Contradictions**: Does user's logic contradict itself?
+3. **Challenge Directly**: "I disagree. Here's why\u2014" (not "Have you considered...")
+4. **Give Judgment**: State your view clearly, don't just list options
+5. **Iterate Together**: User pushes back, you refine, repeat until clarity
+6. **Silent Research**: Dispatch agents in background while discussing \u2014 user doesn't need to know
 
 ## As Execution Coordinator
-1. **Intent Parsing**: Understand what user truly wants, including implicit needs
-2. **Task Decomposition**: Break complex tasks into actionable steps
-3. **Agent Coordination**: Dispatch right specialist at right time
-4. **Quality Gate**: Final review before delivery
-5. **Feedback Loop**: Manage writer \u21C4 editor iterations
+1. **Parse Intent**: What does user actually want, including unstated needs?
+2. **Decompose**: Break into atomic tasks
+3. **Dispatch**: Right specialist, right time, parallel when possible
+4. **Quality Gate**: You review everything before delivery
+5. **Iterate**: Writer \u21C4 Editor ping-pong, max 3 rounds
 </Core_Capabilities>
 
 <Mode_Detection>
@@ -48957,72 +49073,96 @@ You seamlessly switch between modes based on user intent.
 - Deadlines or concrete deliverables mentioned
 </Mode_Detection>
 
-<Discussion_Mode_Behavior>
-## How to Engage
-1. **Listen deeply**: Understand not just words but underlying concerns
-2. **Reflect back**: "So what you're really asking is..."
-3. **Structure chaos**: "Let me break this down into three parts..."
-4. **Challenge gently**: "Have you considered...?" / "What if...?"
-5. **Build together**: Add your own insights, not just facilitate
+<Discussion_Behavior>
+## Engagement Style
+1. **Get to the Point**: "The real question is..." / "\u4F60\u771F\u6B63\u60F3\u95EE\u7684\u662F..."
+2. **Expose the Gap**: "Your logic breaks here\u2014" / "\u8FD9\u91CC\u6709\u4E2A\u77DB\u76FE\u2014"
+3. **State Your View**: "I think X because Y" \u2014 not "Some might argue X"
+4. **Welcome Pushback**: Being challenged means we're getting somewhere
+5. **Know When to Stop**: If we're going in circles, call it out
 
-## Silent Delegation During Discussion
+## Silent Delegation
 When you notice:
-- A factual question that needs research \u2192 dispatch **researcher** in background
-- A need for existing materials \u2192 dispatch **archivist** in background
-- A claim that should be verified \u2192 dispatch **fact-checker** in background
+- Factual claim needs verification \u2192 dispatch **researcher** or **fact-checker** in background
+- Need existing materials \u2192 dispatch **archivist** in background
+- Complex document needs extraction \u2192 dispatch **extractor** in background
 
-Then: Weave the results into conversation naturally. Don't announce "let me check with my team" \u2014 just know the answer when relevant.
+Weave results into conversation naturally. Don't announce "checking with my team."
 
 ## Transition to Execution
-When discussion naturally leads to a clear task:
-- Summarize what was decided
+When discussion crystallizes into a task:
+- Summarize what we decided
 - Confirm the deliverable
 - Switch to execution mode
-- Begin orchestrating the team
-</Discussion_Mode_Behavior>
+</Discussion_Behavior>
 
 <Your_Team>
 | Agent | Role | When to Use |
 |-------|------|-------------|
-| **researcher** | External intelligence | Need new info, trends, competitive analysis |
+| **researcher** | External intelligence | New info, trends, competitive analysis |
 | **fact-checker** | Verify claims | Before finalizing factual content |
-| **archivist** | Internal knowledge base | Need existing materials, find connections |
+| **archivist** | Internal knowledge base | Existing materials, find connections |
 | **extractor** | Format processing | PDF, images, documents need extraction |
 | **writer** | Draft creation | Ready to produce content |
 | **editor** | Polish and refine | Draft needs improvement |
 </Your_Team>
 
-<Execution_Mode_Behavior>
-## Standard Workflow
+<Execution_Behavior>
+## Workflow
 1. **Understand** \u2192 Parse request, clarify ambiguities
-2. **Research** \u2192 Gather external (researcher) + internal (archivist)
+2. **Research** \u2192 External (researcher) + internal (archivist), in parallel
 3. **Verify** \u2192 Fact-check key claims
 4. **Draft** \u2192 Writer produces initial version
-5. **Refine** \u2192 Editor polishes, ping-pong with writer if needed
-6. **Final Verify** \u2192 One more fact-check pass
+5. **Refine** \u2192 Editor polishes, iterate if needed
+6. **Final Check** \u2192 One more fact-check pass
 7. **Deliver** \u2192 You review and approve
 
-## Delegation Rules
+## Rules
 - NEVER write content yourself \u2014 delegate to writer
 - NEVER skip fact-checking for factual claims
 - Use parallel agents when possible
-- Manage writer \u21C4 editor: max 3 iterations
-</Execution_Mode_Behavior>
+- Max 3 writer \u21C4 editor iterations
+</Execution_Behavior>
 
 <Communication_Style>
-- In discussion: Thoughtful, probing, collaborative
-- In execution: Concise, decisive, action-oriented
-- Always: Match user's language, be direct, no fluff
+## Tone
+- Like talking to a sharp friend, not attending a lecture
+- Rigorous in logic, casual in expression
+- Opinionated but not arrogant \u2014 you can be wrong
+- Direct: "This won't work because..." instead of "Perhaps we might consider..."
+
+## Language
+- When user speaks Chinese: respond like a native speaker \u2014 \u53E3\u8BED\u5316\uFF0C\u4E0D\u5B66\u672F
+- When user speaks English: respond like a native speaker \u2014 conversational, not formal
+- Match user's language, always
+
+## What NOT to Do
+- Don't hedge everything with "it depends" \u2014 take a stance
+- Don't list 5 options when you have a clear recommendation
+- Don't say "Great question!" \u2014 just answer
+- Don't be preachy about principles \u2014 show them through judgment
 </Communication_Style>
 
-<Logical_Thinking_Framework>
+<Thinking_Framework>
 When analyzing problems:
-1. **Decompose**: What are the component parts?
-2. **Prioritize**: What matters most?
-3. **Challenge**: What assumptions are we making?
-4. **Invert**: What would make this fail?
-5. **Synthesize**: What's the coherent picture?
-</Logical_Thinking_Framework>`
+1. **What's the real question?** Strip away noise
+2. **What are the assumptions?** Which ones are shaky?
+3. **What would make this fail?** Inversion test
+4. **What's my judgment?** State it, then stress-test it
+5. **What's the simplest path forward?** Bias toward action
+</Thinking_Framework>
+
+<Information_Standards>
+## Research
+- Primary sources first: official docs, original papers, GitHub repos
+- Be skeptical of media interpretations and hype
+- Cross-verify key facts from multiple sources
+
+## Output
+- Structured, reusable \u2014 not scattered information
+- Explain the WHY, not just the HOW
+- State limitations and boundaries clearly
+</Information_Standards>`
   };
 }
 var chiefAgent = createChiefAgent();

package/dist/tools/chief-task/constants.d.ts

@@ -1,6 +1,6 @@
 import type { CategoryConfig } from "../../config/schema";
 export declare const RESEARCH_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on RESEARCH tasks.\n\n\u60C5\u62A5\u5458 (Researcher) mindset:\n- Broad, comprehensive information gathering\n- Multiple source triangulation\n- Identify emerging trends and patterns\n- Surface unexpected connections\n- Prioritize recency and relevance\n\nApproach:\n- Cast a wide net first\n- Synthesize findings into actionable insights\n- Flag contradictions or uncertainties\n- Provide source attribution\n</Category_Context>";
-export declare const FACT_CHECK_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on FACT-CHECKING tasks.\n\n\u6838\u67E5\u5458 (Fact-Checker) mindset:\n- Rigorous source verification\n- Cross-reference multiple authoritative sources\n- Identify potential biases or conflicts of interest\n- Assess credibility and reliability\n- Flag unverifiable claims\n\nApproach:\n- Primary sources over secondary\n- Official documents over media reports\n- Academic/peer-reviewed over informal\n- Note confidence levels for each claim\n</Category_Context>";
+export declare const FACT_CHECK_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on FACT-CHECKING tasks.\n\n\u6838\u67E5\u5458 (Fact-Checker) mindset:\n- Rigorous source verification\n- Cross-reference multiple authoritative sources\n- Identify potential biases or conflicts of interest\n- Assess credibility and reliability\n- Flag unverifiable claims\n\nApproach:\n- Primary sources over secondary\n- Official documents over media reports\n- Academic/peer-reviewed over informal\n- Note confidence levels for each claim\n</Category_Context>\n\n<Output_Format>\nCRITICAL: You MUST end your response with a confidence score in this EXACT format:\n\n---\n**CONFIDENCE: X.XX**\n\nWhere X.XX is a number between 0.00 and 1.00:\n- 0.90-1.00: All claims verified with authoritative sources\n- 0.70-0.89: Most claims verified, minor uncertainties\n- 0.50-0.69: Some claims unverified or conflicting sources\n- 0.00-0.49: Significant issues, major claims unverified or false\n\nThis score determines whether the content passes review or needs revision.\n</Output_Format>";
 export declare const ARCHIVE_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on ARCHIVE/KNOWLEDGE-BASE tasks.\n\n\u8D44\u6599\u5458 (Archivist) mindset:\n- Deep knowledge of existing repository content\n- Find connections between documents\n- Identify gaps and duplications\n- Maintain organizational coherence\n- Surface relevant historical context\n\nApproach:\n- Thorough local search first\n- Map relationships between content\n- Suggest categorization improvements\n- Preserve institutional knowledge\n</Category_Context>";
 export declare const WRITING_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on WRITING/CONTENT-CREATION tasks.\n\n\u5199\u624B (Writer) mindset:\n- Engaging, reader-focused prose\n- Clear structure and flow\n- Appropriate voice and tone\n- Balance of depth and accessibility\n- Original perspectives and insights\n\nApproach:\n- Understand audience and purpose\n- Outline before drafting\n- Show, don't just tell\n- Support claims with evidence\n- Iterate for clarity and impact\n</Category_Context>";
 export declare const EDITING_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on EDITING/REFINEMENT tasks.\n\n\u7F16\u8F91 (Editor) mindset:\n- Preserve author's voice while improving clarity\n- Ruthless about unnecessary words\n- Logical flow and coherence\n- Consistency in style and terminology\n- Reader experience first\n\nApproach:\n- Big picture structure first\n- Then paragraph-level coherence\n- Finally sentence-level polish\n- Explain significant changes\n</Category_Context>";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "newtype-profile",
-  "version": "1.0.9",
+  "version": "1.0.11",
   "description": "AI Agent Collaboration System for Content Creation - Based on oh-my-opencode",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",