@yeongjaeyou/claude-code-config 0.8.1 → 0.9.1

package/.claude/agents/python-pro.md

@@ -2,7 +2,7 @@
 name: python-pro
 description: Write idiomatic Python code with advanced features like decorators, generators, and async/await. Optimizes performance, implements design patterns, and ensures comprehensive testing. Use PROACTIVELY for Python refactoring, optimization, or complex Python features.
 tools: Read, Write, Edit, Bash
-model: sonnet
+model: opus
 ---
 
 You are a Python expert specializing in clean, performant, and idiomatic Python code.

package/.claude/agents/web-researcher.md

@@ -1,7 +1,7 @@
 ---
 name: web-researcher
 description: Use this agent when you need to conduct comprehensive research on technical topics across multiple platforms (Reddit, GitHub, Stack Overflow, Hugging Face, arXiv, etc.) and generate a synthesized report. (project)
-model: sonnet
+model: opus
 ---
 
 # Web Research Expert Agent
@@ -40,7 +40,11 @@ Detect keywords in the prompt to determine execution mode:
 ## Sub-Agent Output Schema
 
 **MANDATORY**: Each platform search Task **MUST** return results in this exact structure.
-Sub-agents that fail to follow this schema will have their results rejected by the coordinator.
+
+**On schema violation:**
+- Missing required fields → retry (max 2 attempts)
+- Invalid YAML → skip with warning, note in report
+- Empty findings → accept with `confidence: 0.1`
 
 ### Required Fields (Quick & Deep)
 
@@ -82,8 +86,8 @@ suggested_followups:
 | Platform | Purpose | Tool |
 |----------|---------|------|
 | **Google** | General web search, recent discussions | WebSearch |
-| **GitHub** | Code, issues, PRs | `gh` CLI |
-| **Hugging Face** | ML models, datasets, Spaces | `huggingface_hub` API |
+| **GitHub** | Code, issues, PRs | `/code-explorer` skill |
+| **Hugging Face** | ML models, datasets, Spaces | `/code-explorer` skill |
 | **Reddit** | Community discussions, experiences | WebSearch |
 | **Stack Overflow** | Q&A, solutions | WebSearch |
 | **Context7** | Official library documentation | MCP |
@@ -156,6 +160,11 @@ Parallel execution (Task tool, run_in_background: true):
 └── Task: arXiv + general web (if needed)
 ```
 
+**Execution policy:**
+- Timeout: 300000ms (5min) per agent
+- Min success: 3/5 agents required to proceed
+- On partial failure: continue with warning in report
+
 **Each Task prompt MUST include:**
 ```
 Format your response according to the Sub-Agent Output Schema.
@@ -187,7 +196,7 @@ Return findings as a YAML block with ALL fields:
 - gaps, conflicts, suggested_followups (Deep Mode required)
 ```
 
-Each Task returns results following Sub-Agent Output Schema. **Reject non-compliant responses.**
+Each Task returns results following Sub-Agent Output Schema. **On non-compliance: retry once, then skip with warning.**
 
 ---
 
@@ -227,6 +236,11 @@ Selective execution (1-2 platforms only):
 └── Task: {Second most suitable platform} (if needed)
 ```
 
+**Round context to include:**
+- Unresolved `gaps` list from Round 1
+- Detected `conflicts` requiring verification
+- Top 3 key findings (title + URL only)
+
 Repeat Round 1.5 analysis → Convergence check
 
 ---
@@ -271,17 +285,16 @@ Validation execution:
 | Condition | Value |
 |-----------|-------|
 | Max rounds | 3 |
-| Max total time | 10 min |
-| Token budget per round | 50k |
+| Max total time | 15 min |
+| Min successful agents | 3/5 per round |
 
 ### Soft Limits (Convergence Conditions)
 
 Terminate when any of the following is met:
 
-- **New information < 10%**: Compared to previous round
-- **All gaps resolved**: All identified gaps addressed
-- **Question fully covered**: All aspects of user question answerable
-- **High confidence**: Core answer confidence > 0.9
+- **New information < 10%**: `(new_unique_urls / prev_total_urls) < 0.1`
+- **All gaps resolved**: `len(unresolved_gaps) == 0`
+- **High confidence**: `avg(agent_confidences) > 0.9`
 
 ### Forced Termination
 
@@ -373,42 +386,26 @@ Handling:
 
 ## Platform-Specific Search Guides
 
-### 1. GitHub Search (gh CLI)
-
-```bash
-# Repository search
-gh search repos "object detection" --sort stars --limit 10
-gh search repos "gradio app" --language python --limit 5
-
-# Code search
-gh search code "Qwen2VL" --extension py
+### 1. GitHub Search
 
-# JSON output
-gh search repos "keyword" --limit 10 --json fullName,description,stargazersCount,url
-```
+Use `/code-explorer` skill for GitHub repository and code search.
 
-**Analysis order:**
-1. README.md (usage)
-2. Main entry point (app.py, main.py)
-3. Dependencies (requirements.txt, pyproject.toml)
+**Capabilities:**
+- Repository search with quality filters (stars, language, date)
+- Code search across repositories
+- Long-tail keyword optimization
+- Multi-query search patterns
 
 ---
 
 ### 2. Hugging Face Search
 
-```python
-from huggingface_hub import HfApi
-api = HfApi()
-
-# Model search
-models = api.list_models(search="object detection", limit=10, sort="downloads")
+Use `/code-explorer` skill for Hugging Face resources search.
 
-# Dataset search
-datasets = api.list_datasets(search="coco", limit=10, sort="downloads")
-
-# Spaces search
-spaces = api.list_spaces(search="gradio demo", limit=10, sort="likes")
-```
+**Capabilities:**
+- Models, Datasets, Spaces search
+- Download via `uvx hf` CLI
+- Search quality principles (Long-tail, Multi-Query)
 
 ---
 

package/.claude/commands/council.md

@@ -4,11 +4,12 @@ description: Consult multiple AI models and synthesize collective wisdom (LLM Co
 
 # LLM Council
 
-Inspired by Andrej Karpathy's LLM Council: query multiple AI models with the same question, anonymize their responses, and synthesize collective wisdom.
+Inspired by Andrej Karpathy's LLM Council: query multiple AI models with the same question, anonymize their responses, and synthesize collective wisdom through multi-round deliberation.
 
 **Core Philosophy:**
 - Collective intelligence > single expert opinion
 - Anonymization prevents model favoritism
+- Multi-round deliberation resolves conflicts and fills gaps
 - Diverse perspectives lead to better answers
 
 ---
@@ -19,17 +20,51 @@ Inspired by Andrej Karpathy's LLM Council: query multiple AI models with the sam
 
 1. **No arguments**: `/council` - Prompt user for question
 2. **Question only**: `/council How should I structure this API?`
-3. **With evaluate flag**: `/council --evaluate What's the best approach?`
-4. **With deep reasoning**: `/council --deep What's the best architecture?`
-5. **Combined**: `/council --deep --evaluate Complex design decision?`
+3. **Quick mode**: `/council --quick What's the best approach?`
 
 **Flags:**
-- `--evaluate`: Enable Stage 3 peer evaluation where each model ranks the others
-- `--deep`: Enable maximum reasoning depth (Codex: xhigh + gpt-5.1-codex-max)
+- `--quick`: Quick mode - reduced reasoning depth, faster convergence
+
+**Default behavior (no flags):**
+- Maximum reasoning depth (Codex: xhigh + gpt-5.1-codex-max)
+- Full multi-round deliberation (up to 3 rounds)
 
 ---
 
-## Context Gathering (Before Stage 1)
+## Council Member Output Schema
+
+All council members MUST return responses in this structured format:
+
+```yaml
+council_member:
+  model: "opus" | "sonnet" | "codex" | "gemini"
+  response:
+    summary: "1-2 sentence core answer"
+    detailed_answer: "full response content"
+    confidence: 0.0-1.0
+    key_points:
+      - point: "key insight"
+        evidence: "file:line or reasoning"
+    code_references:
+      - file: "/absolute/path/to/file.py"
+        lines: "42-58"
+        context: "why this is relevant"
+    caveats:
+      - "potential limitation or edge case"
+  # Round 2+ additional fields
+  gaps:
+    - "aspect not fully addressed"
+  conflicts:
+    - "disagrees with [model] on [topic]: [reason]"
+```
+
+**Schema enforcement:**
+- Sub-agents that fail to follow this schema will have their results flagged
+- Missing required fields trigger re-query in next round
+
+---
+
+## Context Gathering (Before Round 1)
 
 Before querying models, collect relevant context:
 
@@ -45,17 +80,14 @@ Model-specific guidelines (project root):
 - ./gemini.md (Gemini)
 ```
 
-**File Path Inclusion:**
-When the question involves specific files or images, provide **exact absolute paths** so each model can read them directly:
-
+**File Path Inclusion (MANDATORY format):**
 ```
-Prompt addition:
-"Relevant files for this question:
-- /absolute/path/to/file.py (source code)
+Relevant files for this question:
+- /absolute/path/to/file.py:45-78 (authentication logic)
+- /absolute/path/to/model.py:12-35 (User model definition)
 - /absolute/path/to/screenshot.png (UI reference)
-- /absolute/path/to/diagram.jpg (architecture)
 
-You can use Read tool to examine these files if needed."
+Use your file access tools to READ these files directly.
 ```
 
 **Model-specific file access:**
@@ -84,124 +116,257 @@ Directories: node_modules/, __pycache__/, .git/
 
 ## Execution
 
-### Stage 1: Collect Responses
+### Round 1: Collect Initial Responses
 
-Query all 4 models **in parallel** with identical prompt:
+Query all 4 models **in parallel** using Task tool with sub-agents:
 
+**Claude Opus:**
 ```
-Models:
-- Claude Opus    → Task(model="opus", subagent_type="general-purpose")
-- Claude Sonnet  → Task(model="sonnet", subagent_type="general-purpose")
-- Codex          → mcp__codex__codex(sandbox="read-only", reasoningEffort="high")
-- Gemini         → Bash: cat <<'EOF' | gemini -p -
-```
+Task(model="opus", subagent_type="general-purpose", run_in_background: true):
+  prompt: |
+    You are participating in an LLM Council deliberation as Claude Opus.
+
+    ## Guidelines
+    Read and follow ./CLAUDE.md project guidelines.
+    You have access to MCP tools. Use them actively to gather accurate information.
 
-**Reasoning Control:**
-| Model | Default | --deep |
-|-------|---------|--------|
-| Claude Opus | standard | same |
-| Claude Sonnet | standard | same |
-| Codex | `reasoningEffort="high"` | `xhigh` + `gpt-5.1-codex-max` |
-| Gemini | CLI control N/A | CLI control N/A |
+    ## Question
+    [USER_QUESTION]
 
-Note:
-- Codex `xhigh` only works with `gpt-5.1-codex-max` or `gpt-5.2`
-- Gemini CLI doesn't support thinking level control yet (Issue #6693)
+    ## Context Files (READ directly using exact paths)
+    [FILE_LIST_WITH_LINE_NUMBERS]
 
-**Prompt template for each model:**
+    ## Current Changes
+    [git diff summary]
+
+    ## Instructions
+    Provide your best answer following the Council Member Output Schema.
+    Be concise but thorough. Focus on accuracy and actionable insights.
+
+    ## Output (YAML format required)
+    [COUNCIL_MEMBER_SCHEMA]
 ```
-You are participating in an LLM Council deliberation.
 
-## Guidelines
-Read and follow your project guidelines before answering:
-- Claude models: Read ./CLAUDE.md
-- Codex: Read ./AGENTS.md
-- Gemini: Read ./gemini.md
+**Claude Sonnet:**
+```
+Task(model="sonnet", subagent_type="general-purpose", run_in_background: true):
+  prompt: [Same structure as Opus]
+```
 
-## Question
-[USER_QUESTION]
+**Codex:**
+```
+Task(subagent_type="general-purpose", run_in_background: true):
+  prompt: |
+    You are participating in an LLM Council deliberation as Codex.
+
+    ## Tool Usage
+    Use mcp__codex__codex tool with:
+    - sandbox: "read-only"
+    - workingDirectory: "{PROJECT_ROOT}"
+    - reasoningEffort: "xhigh"  (or "high" with --quick)
+    - model: "gpt-5.1-codex-max"
 
-## Context
-Working Directory: [ABSOLUTE_PATH]
+    ## Guidelines
+    Read and follow ./AGENTS.md project guidelines.
+    You have access to MCP tools. Use them actively to gather accurate information.
 
-## Relevant Files (READ these directly for accurate context)
-- [/absolute/path/to/file1.ext] - [brief description]
-- [/absolute/path/to/image.png] - [screenshot/diagram description]
-- [/absolute/path/to/file2.ext] - [brief description]
+    ## Question
+    [USER_QUESTION]
 
-Do NOT ask for file contents. Use your file access tools to READ them directly.
+    ## Context Files
+    [FILE_LIST_WITH_LINE_NUMBERS]
 
-## Current Changes (if applicable)
-[git diff summary or key changes]
+    ## Instructions
+    Parse Codex's response and return structured YAML following the schema.
 
-## Instructions
-Provide your best answer. Be concise but thorough.
-Focus on accuracy, practicality, and actionable insights.
+    ## Output (YAML format required)
+    [COUNCIL_MEMBER_SCHEMA]
 ```
 
-**Important:**
-- Use `run_in_background: true` for Task calls to enable true parallelism
-- Set timeout: 120000ms for each call
-- Continue with successful responses if some models fail
+**Gemini:**
+```
+Task(subagent_type="general-purpose", run_in_background: true):
+  prompt: |
+    You are participating in an LLM Council deliberation as Gemini.
 
-### Stage 2: Anonymize
+    ## Tool Usage
+    Use Bash tool to invoke Gemini CLI:
+    ```bash
+    cat <<'EOF' | gemini -p -
+    [GEMINI_PROMPT_WITH_CONTEXT]
+    EOF
+    ```
 
-After collecting responses:
+    ## Guidelines
+    Read and follow ./gemini.md project guidelines.
+    You have access to MCP tools. Use them actively to gather accurate information.
 
-1. **Shuffle** the response order randomly
-2. **Assign labels**: Response A, Response B, Response C, Response D
-3. **Create mapping** (keep internal, reveal later):
-   ```
+    ## Question
+    [USER_QUESTION]
+
+    ## Context Files (include content since Gemini has limited file access)
+    [FILE_CONTENTS_WITH_LINE_NUMBERS]
+
+    ## Instructions
+    Parse Gemini's response and return structured YAML following the schema.
+
+    ## Output (YAML format required)
+    [COUNCIL_MEMBER_SCHEMA]
+```
+
+**Important:**
+- Use `run_in_background: true` for true parallelism
+- Timeout per model (use same value for TaskOutput):
+  | Model | Timeout | Reason |
+  |-------|---------|--------|
+  | Opus/Sonnet | 300000ms (5min) | Direct execution |
+  | Codex | 480000ms (8min) | MCP tool + deep reasoning |
+  | Gemini | 600000ms (10min) | CLI invocation + long thinking |
+- Continue with successful responses if some models fail (min 2/4 required)
+- **TaskOutput must use matching timeout**: `TaskOutput(task_id, block=true, timeout=600000)`
+
+### Round 1.5: Coordinator Analysis (MANDATORY)
+
+**⚠️ DO NOT SKIP**: After collecting responses, the coordinator **MUST** perform this analysis before synthesis.
+Skipping Round 1.5 defeats the purpose of multi-round deliberation.
+
+**1. Anonymize Responses:**
+```
+1. Shuffle response order randomly
+2. Assign labels: Response A, B, C, D
+3. Create internal mapping:
    label_to_model = {
      "Response A": "gemini",
      "Response B": "opus",
      "Response C": "sonnet",
      "Response D": "codex"
    }
-   ```
+```
 
-4. **Display** anonymized responses to user in a structured format
+**2. Gap Analysis:**
+```yaml
+gaps_detected:
+  - model: "opus"
+    gap: "performance benchmarks not addressed"
+    severity: "medium"
+  - model: "gemini"
+    gap: "security implications missing"
+    severity: "high"
+```
 
-### Stage 3: Peer Evaluation (if --evaluate)
+**3. Conflict Detection:**
+```yaml
+conflicts_detected:
+  - topic: "recommended approach"
+    positions:
+      - model: "opus"
+        position: "use library A"
+        evidence: "official docs recommend"
+      - model: "codex"
+        position: "use library B"
+        evidence: "better performance"
+    resolution_needed: true
+```
+
+**4. Convergence Check (REQUIRED before synthesis):**
+```yaml
+convergence_status:
+  confidence_average: 0.815
+  new_information_ratio: 0.15
+  gaps_remaining: 2
+  conflicts_remaining: 1
+  decision: "proceed_to_round_2" | "terminate_and_synthesize"
+```
+
+**Decision logic:**
+- If `gaps_remaining > 0` AND `confidence_average < 0.9` → `proceed_to_round_2`
+- If `conflicts_remaining > 0` AND round < 3 → `proceed_to_round_2`
+- Otherwise → `terminate_and_synthesize`
 
-If `--evaluate` flag is present, query each model again with all anonymized responses:
+### Round 2: Targeted Re-queries (Conditional)
 
-**Evaluation prompt:**
+If convergence criteria not met, re-query only models with gaps/conflicts:
+
+**Re-query prompt template:**
 ```
-You are evaluating responses from an LLM Council.
+## Previous Round Summary
+Round 1 produced the following positions:
+
+### Response A (Confidence: 0.92)
+- Position: [summary]
+- Key points: [list]
+
+### Response B (Confidence: 0.85)
+- Position: [summary]
+- Key points: [list]
+
+[... other responses ...]
 
-Original Question: [USER_QUESTION]
+## Gaps Identified
+- [gap 1]
+- [gap 2]
 
-[ANONYMIZED_RESPONSES]
+## Conflicts Detected
+- Topic: [topic]
+  - Position A: [description]
+  - Position B: [description]
 
-Instructions:
-1. Evaluate each response for accuracy, completeness, and usefulness
-2. Identify strengths and weaknesses of each
-3. End with "FINAL RANKING:" section
+## Re-query Focus
+Please address specifically:
+1. [specific gap or conflict to resolve]
+2. [specific gap or conflict to resolve]
 
-FINAL RANKING:
-1. Response [X] - [brief reason]
-2. Response [Y] - [brief reason]
-3. Response [Z] - [brief reason]
-4. Response [W] - [brief reason]
+Provide evidence and reasoning for your position.
+Update your confidence score based on new information.
+
+## Output (YAML format required)
+[COUNCIL_MEMBER_SCHEMA with gaps/conflicts fields]
 ```
 
-**Calculate aggregate rankings:**
-- Parse each model's ranking
-- Sum rank positions for each response
-- Lower total = higher collective approval
+### Round 2.5: Coordinator Analysis
+
+Same as Round 1.5. Check convergence again.
+
+### Round 3: Final Cross-Validation (Conditional)
 
-### Stage 4: Synthesize
+If still not converged after Round 2:
+- Focused on resolving remaining conflicts
+- Models see other models' positions (still anonymized)
+- Final opportunity for consensus
 
-As the main agent (moderator), synthesize the final answer:
+### Synthesis
+
+After convergence or max rounds:
 
 1. **Reveal** the label-to-model mapping
 2. **Analyze** all responses:
    - Consensus points (where models agree)
-   - Disagreements (where they differ)
+   - Resolved conflicts (with reasoning)
+   - Remaining disagreements (with analysis)
    - Unique insights (valuable points from individual models)
-3. **If --evaluate**: Include aggregate rankings and "street cred" scores
-4. **Produce** final verdict combining best elements
+3. **Produce** final verdict combining best elements
+
+---
+
+## Termination Criteria
+
+### Hard Limits (Mandatory Termination)
+| Condition | Value |
+|-----------|-------|
+| Max rounds | 3 |
+| Max total time | 20 min |
+| Max per-model timeout | 10 min (Gemini) |
+| Min successful models | 2/4 (proceed with partial results) |
+| All models failed | immediate termination |
+
+### Soft Limits (Convergence - any triggers termination)
+| Condition | Threshold |
+|-----------|-----------|
+| Average confidence | > 0.9 |
+| New information ratio | < 10% |
+| All gaps resolved | 0 remaining |
+| Strong consensus | 3+ models agree |
+| Conflicts irreconcilable | Cannot be resolved with more queries |
 
 ---
 
@@ -213,70 +378,88 @@ As the main agent (moderator), synthesize the final answer:
 ### Question
 [Original user question]
 
+### Deliberation Process
+| Round | Models Queried | Convergence | Status |
+|-------|---------------|-------------|--------|
+| 1 | All (4) | 65% | Gaps detected |
+| 2 | Codex, Gemini | 85% | Conflict on approach |
+| 3 | Codex | 95% | Converged |
+
 ### Individual Responses (Anonymized)
 
-#### Response A
+#### Response A (Confidence: 0.92)
 [Content]
 
-#### Response B
+**Key Points:**
+- [point 1] (evidence: file:line)
+- [point 2] (evidence: file:line)
+
+#### Response B (Confidence: 0.85)
 [Content]
 
-#### Response C
+#### Response C (Confidence: 0.78)
 [Content]
 
-#### Response D
+#### Response D (Confidence: 0.71)
 [Content]
 
 ### Model Reveal
-| Label | Model |
-|-------|-------|
-| Response A | [model name] |
-| Response B | [model name] |
-| Response C | [model name] |
-| Response D | [model name] |
-
-### Aggregate Rankings (if --evaluate)
-| Rank | Model | Avg Position | Votes |
-|------|-------|--------------|-------|
+| Label | Model | Final Confidence |
+|-------|-------|-----------------|
+| Response A | codex | 0.92 |
+| Response B | opus | 0.85 |
+| Response C | sonnet | 0.78 |
+| Response D | gemini | 0.71 |
+
+### Coordinator Analysis
+
+#### Gaps Addressed
+| Gap | Resolved By | Round |
+|-----|-------------|-------|
+| Performance benchmarks | Codex | 2 |
+| Security considerations | Opus | 1 |
+
+#### Conflicts Resolved
+| Topic | Final Position | Reasoning |
+|-------|---------------|-----------|
+| Library choice | Library A | Official docs + 3 model consensus |
+
+#### Remaining Disagreements
+| Topic | Positions | Analysis |
+|-------|-----------|----------|
+| [topic] | A: [pos], B: [pos] | [why unresolved] |
 
 ### Council Synthesis
 
 #### Consensus
-[Points where all/most models agree]
-
-#### Disagreements
-[Points of divergence with analysis]
+[Points where all/most models agree - with evidence]
 
-#### Unique Insights
-[Valuable contributions from specific models]
+#### Key Insights by Model
+| Model | Unique Contribution |
+|-------|-------------------|
+| Codex | [insight] |
+| Opus | [insight] |
 
 ### Final Verdict
-[Synthesized answer combining collective wisdom]
+[Synthesized answer combining collective wisdom with confidence level and caveats]
+
+### Code References
+| File | Lines | Context |
+|------|-------|---------|
+| /path/to/file.py | 45-78 | Authentication logic |
 ```
 
 ---
 
 ## Error Handling
 
-- **Model timeout**: Continue with successful responses, note failures
-- **All models fail**: Report error, suggest retry
-- **Parse failure in rankings**: Use fallback regex extraction
-- **Empty response**: Exclude from synthesis, note in output
-
----
-
-## Examples
-
-```bash
-# Basic council consultation
-/council What's the best way to implement caching in this API?
-
-# With peer evaluation for important decisions
-/council --evaluate Should we use microservices or monolith for this project?
-
-# Architecture review
-/council --evaluate Review the current authentication flow and suggest improvements
-```
+| Error | Response |
+|-------|----------|
+| Model timeout | Continue with successful responses, note failures |
+| All models fail | Report error, suggest retry |
+| Parse failure | Use fallback extraction, flag for re-query |
+| Empty response | Exclude from synthesis, note in output |
+| Schema violation | Flag and request re-query in next round |
 
 ---
 
@@ -284,35 +467,50 @@ As the main agent (moderator), synthesize the final answer:
 
 Use `AskUserQuestion` tool when clarification is needed:
 
-**Before Stage 1:**
+**Before Round 1:**
 - Question is ambiguous or too broad
 - Missing critical context (e.g., "review this code" but no file specified)
 - Multiple interpretations possible
 
-**During Execution:**
-- Conflicting requirements detected
-- Need to confirm scope (e.g., "Should I include performance considerations?")
+**During Deliberation:**
+- Strong disagreement between models that cannot be resolved
+- New information discovered that changes the question scope
 
 **After Synthesis:**
-- Models strongly disagree and user input would help decide
+- Remaining disagreements require user input to decide
 - Actionable next steps require user confirmation
 
 **Example questions:**
 ```
 - "Your question mentions 'the API' - which specific endpoint or service?"
-- "Should the council focus on: (1) Code quality, (2) Architecture, (3) Performance, or (4) All aspects?"
 - "Models disagree on X vs Y approach. Which aligns better with your constraints?"
+- "Should the council prioritize performance or maintainability?"
 ```
 
 **Important:** Never assume or guess when context is unclear. Ask first, then proceed.
 
 ---
 
+## Examples
+
+```bash
+# Standard council consultation (full multi-round, max reasoning)
+/council What's the best way to implement caching in this API?
+
+# Quick mode for simpler questions
+/council --quick Should we use tabs or spaces for indentation?
+
+# Architecture review
+/council Review the current authentication flow and suggest improvements
+```
+
+---
+
 ## Guidelines
 
 - Respond in the same language as the user's question
 - No emojis in code or documentation
 - If context is needed, gather it before querying models
-- For code-related questions, include relevant file snippets in the prompt
+- For code-related questions, include relevant file snippets with line numbers
 - Respect `@CLAUDE.md` project conventions
 - **Never assume unclear context - use AskUserQuestion to clarify**

package/README.md

@@ -64,7 +64,7 @@ A collection of custom slash commands, agents, and skills for Claude Code CLI.
 | `/ask-deepwiki` | Deep query GitHub repositories via DeepWiki MCP |
 | `/ask-codex` | Request code review via Codex MCP (with Claude cross-check) |
 | `/ask-gemini` | Request code review via Gemini CLI (with Claude cross-check) |
-| `/council` | Consult multiple AI models and synthesize collective wisdom |
+| `/council` | Consult multiple AI models with multi-round deliberation |
 
 ### GitHub Workflow Commands (`/gh/`)
 
@@ -89,8 +89,8 @@ A collection of custom slash commands, agents, and skills for Claude Code CLI.
 
 | Agent | Description |
 |-------|-------------|
-| `web-researcher` | Multi-platform tech research (Reddit, GitHub, SO, HF, arXiv, etc.) |
-| `python-pro` | Python advanced features expert (decorators, generators, async/await) |
+| `web-researcher` | Multi-platform tech research with opus model (Reddit, GitHub, HF, arXiv, etc.) |
+| `python-pro` | Python expert with opus model (decorators, generators, async/await, optimization) |
 
 ## Skills
 
@@ -247,8 +247,10 @@ cp node_modules/@yeongjaeyou/claude-code-config/.mcp.json .
 
 ### `/council` - LLM Council
 - Query multiple AI models (Opus, Sonnet, Codex, Gemini) in parallel
+- Multi-round deliberation with gap analysis and conflict resolution
 - Anonymize responses for unbiased evaluation
-- Synthesize collective wisdom into consensus
+- Convergence-based termination (confidence > 0.9, consensus reached)
+- Use `--quick` flag for faster single-round responses
 
 ### `web-researcher` Agent
 - Multi-platform search: GitHub (`gh` CLI), Hugging Face, Reddit, SO, arXiv

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yeongjaeyou/claude-code-config",
-  "version": "0.8.1",
+  "version": "0.9.1",
   "description": "Claude Code CLI custom commands, agents, and skills",
   "bin": {
     "claude-code-config": "./bin/cli.js"