@yeongjaeyou/claude-code-config 0.8.1 → 0.9.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -2,7 +2,7 @@
2
2
  name: python-pro
3
3
  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.
4
4
  tools: Read, Write, Edit, Bash
5
- model: sonnet
5
+ model: opus
6
6
  ---
7
7
 
8
8
  You are a Python expert specializing in clean, performant, and idiomatic Python code.
@@ -1,7 +1,7 @@
1
1
  ---
2
2
  name: web-researcher
3
3
  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)
4
- model: sonnet
4
+ model: opus
5
5
  ---
6
6
 
7
7
  # Web Research Expert Agent
@@ -40,7 +40,11 @@ Detect keywords in the prompt to determine execution mode:
40
40
  ## Sub-Agent Output Schema
41
41
 
42
42
  **MANDATORY**: Each platform search Task **MUST** return results in this exact structure.
43
- Sub-agents that fail to follow this schema will have their results rejected by the coordinator.
43
+
44
+ **On schema violation:**
45
+ - Missing required fields → retry (max 2 attempts)
46
+ - Invalid YAML → skip with warning, note in report
47
+ - Empty findings → accept with `confidence: 0.1`
44
48
 
45
49
  ### Required Fields (Quick & Deep)
46
50
 
@@ -82,8 +86,8 @@ suggested_followups:
82
86
  | Platform | Purpose | Tool |
83
87
  |----------|---------|------|
84
88
  | **Google** | General web search, recent discussions | WebSearch |
85
- | **GitHub** | Code, issues, PRs | `gh` CLI |
86
- | **Hugging Face** | ML models, datasets, Spaces | `huggingface_hub` API |
89
+ | **GitHub** | Code, issues, PRs | `/code-explorer` skill |
90
+ | **Hugging Face** | ML models, datasets, Spaces | `/code-explorer` skill |
87
91
  | **Reddit** | Community discussions, experiences | WebSearch |
88
92
  | **Stack Overflow** | Q&A, solutions | WebSearch |
89
93
  | **Context7** | Official library documentation | MCP |
@@ -156,6 +160,11 @@ Parallel execution (Task tool, run_in_background: true):
156
160
  └── Task: arXiv + general web (if needed)
157
161
  ```
158
162
 
163
+ **Execution policy:**
164
+ - Timeout: 300000ms (5min) per agent
165
+ - Min success: 3/5 agents required to proceed
166
+ - On partial failure: continue with warning in report
167
+
159
168
  **Each Task prompt MUST include:**
160
169
  ```
161
170
  Format your response according to the Sub-Agent Output Schema.
@@ -187,7 +196,7 @@ Return findings as a YAML block with ALL fields:
187
196
  - gaps, conflicts, suggested_followups (Deep Mode required)
188
197
  ```
189
198
 
190
- Each Task returns results following Sub-Agent Output Schema. **Reject non-compliant responses.**
199
+ Each Task returns results following Sub-Agent Output Schema. **On non-compliance: retry once, then skip with warning.**
191
200
 
192
201
  ---
193
202
 
@@ -227,6 +236,11 @@ Selective execution (1-2 platforms only):
227
236
  └── Task: {Second most suitable platform} (if needed)
228
237
  ```
229
238
 
239
+ **Round context to include:**
240
+ - Unresolved `gaps` list from Round 1
241
+ - Detected `conflicts` requiring verification
242
+ - Top 3 key findings (title + URL only)
243
+
230
244
  Repeat Round 1.5 analysis → Convergence check
231
245
 
232
246
  ---
@@ -271,17 +285,16 @@ Validation execution:
271
285
  | Condition | Value |
272
286
  |-----------|-------|
273
287
  | Max rounds | 3 |
274
- | Max total time | 10 min |
275
- | Token budget per round | 50k |
288
+ | Max total time | 15 min |
289
+ | Min successful agents | 3/5 per round |
276
290
 
277
291
  ### Soft Limits (Convergence Conditions)
278
292
 
279
293
  Terminate when any of the following is met:
280
294
 
281
- - **New information < 10%**: Compared to previous round
282
- - **All gaps resolved**: All identified gaps addressed
283
- - **Question fully covered**: All aspects of user question answerable
284
- - **High confidence**: Core answer confidence > 0.9
295
+ - **New information < 10%**: `(new_unique_urls / prev_total_urls) < 0.1`
296
+ - **All gaps resolved**: `len(unresolved_gaps) == 0`
297
+ - **High confidence**: `avg(agent_confidences) > 0.9`
285
298
 
286
299
  ### Forced Termination
287
300
 
@@ -373,42 +386,26 @@ Handling:
373
386
 
374
387
  ## Platform-Specific Search Guides
375
388
 
376
- ### 1. GitHub Search (gh CLI)
377
-
378
- ```bash
379
- # Repository search
380
- gh search repos "object detection" --sort stars --limit 10
381
- gh search repos "gradio app" --language python --limit 5
382
-
383
- # Code search
384
- gh search code "Qwen2VL" --extension py
389
+ ### 1. GitHub Search
385
390
 
386
- # JSON output
387
- gh search repos "keyword" --limit 10 --json fullName,description,stargazersCount,url
388
- ```
391
+ Use `/code-explorer` skill for GitHub repository and code search.
389
392
 
390
- **Analysis order:**
391
- 1. README.md (usage)
392
- 2. Main entry point (app.py, main.py)
393
- 3. Dependencies (requirements.txt, pyproject.toml)
393
+ **Capabilities:**
394
+ - Repository search with quality filters (stars, language, date)
395
+ - Code search across repositories
396
+ - Long-tail keyword optimization
397
+ - Multi-query search patterns
394
398
 
395
399
  ---
396
400
 
397
401
  ### 2. Hugging Face Search
398
402
 
399
- ```python
400
- from huggingface_hub import HfApi
401
- api = HfApi()
402
-
403
- # Model search
404
- models = api.list_models(search="object detection", limit=10, sort="downloads")
403
+ Use `/code-explorer` skill for Hugging Face resources search.
405
404
 
406
- # Dataset search
407
- datasets = api.list_datasets(search="coco", limit=10, sort="downloads")
408
-
409
- # Spaces search
410
- spaces = api.list_spaces(search="gradio demo", limit=10, sort="likes")
411
- ```
405
+ **Capabilities:**
406
+ - Models, Datasets, Spaces search
407
+ - Download via `uvx hf` CLI
408
+ - Search quality principles (Long-tail, Multi-Query)
412
409
 
413
410
  ---
414
411
 
@@ -4,11 +4,12 @@ description: Consult multiple AI models and synthesize collective wisdom (LLM Co
4
4
 
5
5
  # LLM Council
6
6
 
7
- Inspired by Andrej Karpathy's LLM Council: query multiple AI models with the same question, anonymize their responses, and synthesize collective wisdom.
7
+ 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.
8
8
 
9
9
  **Core Philosophy:**
10
10
  - Collective intelligence > single expert opinion
11
11
  - Anonymization prevents model favoritism
12
+ - Multi-round deliberation resolves conflicts and fills gaps
12
13
  - Diverse perspectives lead to better answers
13
14
 
14
15
  ---
@@ -19,17 +20,51 @@ Inspired by Andrej Karpathy's LLM Council: query multiple AI models with the sam
19
20
 
20
21
  1. **No arguments**: `/council` - Prompt user for question
21
22
  2. **Question only**: `/council How should I structure this API?`
22
- 3. **With evaluate flag**: `/council --evaluate What's the best approach?`
23
- 4. **With deep reasoning**: `/council --deep What's the best architecture?`
24
- 5. **Combined**: `/council --deep --evaluate Complex design decision?`
23
+ 3. **Quick mode**: `/council --quick What's the best approach?`
25
24
 
26
25
  **Flags:**
27
- - `--evaluate`: Enable Stage 3 peer evaluation where each model ranks the others
28
- - `--deep`: Enable maximum reasoning depth (Codex: xhigh + gpt-5.1-codex-max)
26
+ - `--quick`: Quick mode - reduced reasoning depth, faster convergence
27
+
28
+ **Default behavior (no flags):**
29
+ - Maximum reasoning depth (Codex: xhigh + gpt-5.1-codex-max)
30
+ - Full multi-round deliberation (up to 3 rounds)
29
31
 
30
32
  ---
31
33
 
32
- ## Context Gathering (Before Stage 1)
34
+ ## Council Member Output Schema
35
+
36
+ All council members MUST return responses in this structured format:
37
+
38
+ ```yaml
39
+ council_member:
40
+ model: "opus" | "sonnet" | "codex" | "gemini"
41
+ response:
42
+ summary: "1-2 sentence core answer"
43
+ detailed_answer: "full response content"
44
+ confidence: 0.0-1.0
45
+ key_points:
46
+ - point: "key insight"
47
+ evidence: "file:line or reasoning"
48
+ code_references:
49
+ - file: "/absolute/path/to/file.py"
50
+ lines: "42-58"
51
+ context: "why this is relevant"
52
+ caveats:
53
+ - "potential limitation or edge case"
54
+ # Round 2+ additional fields
55
+ gaps:
56
+ - "aspect not fully addressed"
57
+ conflicts:
58
+ - "disagrees with [model] on [topic]: [reason]"
59
+ ```
60
+
61
+ **Schema enforcement:**
62
+ - Sub-agents that fail to follow this schema will have their results flagged
63
+ - Missing required fields trigger re-query in next round
64
+
65
+ ---
66
+
67
+ ## Context Gathering (Before Round 1)
33
68
 
34
69
  Before querying models, collect relevant context:
35
70
 
@@ -45,17 +80,14 @@ Model-specific guidelines (project root):
45
80
  - ./gemini.md (Gemini)
46
81
  ```
47
82
 
48
- **File Path Inclusion:**
49
- When the question involves specific files or images, provide **exact absolute paths** so each model can read them directly:
50
-
83
+ **File Path Inclusion (MANDATORY format):**
51
84
  ```
52
- Prompt addition:
53
- "Relevant files for this question:
54
- - /absolute/path/to/file.py (source code)
85
+ Relevant files for this question:
86
+ - /absolute/path/to/file.py:45-78 (authentication logic)
87
+ - /absolute/path/to/model.py:12-35 (User model definition)
55
88
  - /absolute/path/to/screenshot.png (UI reference)
56
- - /absolute/path/to/diagram.jpg (architecture)
57
89
 
58
- You can use Read tool to examine these files if needed."
90
+ Use your file access tools to READ these files directly.
59
91
  ```
60
92
 
61
93
  **Model-specific file access:**
@@ -84,124 +116,257 @@ Directories: node_modules/, __pycache__/, .git/
84
116
 
85
117
  ## Execution
86
118
 
87
- ### Stage 1: Collect Responses
119
+ ### Round 1: Collect Initial Responses
88
120
 
89
- Query all 4 models **in parallel** with identical prompt:
121
+ Query all 4 models **in parallel** using Task tool with sub-agents:
90
122
 
123
+ **Claude Opus:**
91
124
  ```
92
- Models:
93
- - Claude Opus → Task(model="opus", subagent_type="general-purpose")
94
- - Claude Sonnet → Task(model="sonnet", subagent_type="general-purpose")
95
- - Codex → mcp__codex__codex(sandbox="read-only", reasoningEffort="high")
96
- - Gemini → Bash: cat <<'EOF' | gemini -p -
97
- ```
125
+ Task(model="opus", subagent_type="general-purpose", run_in_background: true):
126
+ prompt: |
127
+ You are participating in an LLM Council deliberation as Claude Opus.
128
+
129
+ ## Guidelines
130
+ Read and follow ./CLAUDE.md project guidelines.
131
+ You have access to MCP tools. Use them actively to gather accurate information.
98
132
 
99
- **Reasoning Control:**
100
- | Model | Default | --deep |
101
- |-------|---------|--------|
102
- | Claude Opus | standard | same |
103
- | Claude Sonnet | standard | same |
104
- | Codex | `reasoningEffort="high"` | `xhigh` + `gpt-5.1-codex-max` |
105
- | Gemini | CLI control N/A | CLI control N/A |
133
+ ## Question
134
+ [USER_QUESTION]
106
135
 
107
- Note:
108
- - Codex `xhigh` only works with `gpt-5.1-codex-max` or `gpt-5.2`
109
- - Gemini CLI doesn't support thinking level control yet (Issue #6693)
136
+ ## Context Files (READ directly using exact paths)
137
+ [FILE_LIST_WITH_LINE_NUMBERS]
110
138
 
111
- **Prompt template for each model:**
139
+ ## Current Changes
140
+ [git diff summary]
141
+
142
+ ## Instructions
143
+ Provide your best answer following the Council Member Output Schema.
144
+ Be concise but thorough. Focus on accuracy and actionable insights.
145
+
146
+ ## Output (YAML format required)
147
+ [COUNCIL_MEMBER_SCHEMA]
112
148
  ```
113
- You are participating in an LLM Council deliberation.
114
149
 
115
- ## Guidelines
116
- Read and follow your project guidelines before answering:
117
- - Claude models: Read ./CLAUDE.md
118
- - Codex: Read ./AGENTS.md
119
- - Gemini: Read ./gemini.md
150
+ **Claude Sonnet:**
151
+ ```
152
+ Task(model="sonnet", subagent_type="general-purpose", run_in_background: true):
153
+ prompt: [Same structure as Opus]
154
+ ```
120
155
 
121
- ## Question
122
- [USER_QUESTION]
156
+ **Codex:**
157
+ ```
158
+ Task(subagent_type="general-purpose", run_in_background: true):
159
+ prompt: |
160
+ You are participating in an LLM Council deliberation as Codex.
161
+
162
+ ## Tool Usage
163
+ Use mcp__codex__codex tool with:
164
+ - sandbox: "read-only"
165
+ - workingDirectory: "{PROJECT_ROOT}"
166
+ - reasoningEffort: "xhigh" (or "high" with --quick)
167
+ - model: "gpt-5.1-codex-max"
123
168
 
124
- ## Context
125
- Working Directory: [ABSOLUTE_PATH]
169
+ ## Guidelines
170
+ Read and follow ./AGENTS.md project guidelines.
171
+ You have access to MCP tools. Use them actively to gather accurate information.
126
172
 
127
- ## Relevant Files (READ these directly for accurate context)
128
- - [/absolute/path/to/file1.ext] - [brief description]
129
- - [/absolute/path/to/image.png] - [screenshot/diagram description]
130
- - [/absolute/path/to/file2.ext] - [brief description]
173
+ ## Question
174
+ [USER_QUESTION]
131
175
 
132
- Do NOT ask for file contents. Use your file access tools to READ them directly.
176
+ ## Context Files
177
+ [FILE_LIST_WITH_LINE_NUMBERS]
133
178
 
134
- ## Current Changes (if applicable)
135
- [git diff summary or key changes]
179
+ ## Instructions
180
+ Parse Codex's response and return structured YAML following the schema.
136
181
 
137
- ## Instructions
138
- Provide your best answer. Be concise but thorough.
139
- Focus on accuracy, practicality, and actionable insights.
182
+ ## Output (YAML format required)
183
+ [COUNCIL_MEMBER_SCHEMA]
140
184
  ```
141
185
 
142
- **Important:**
143
- - Use `run_in_background: true` for Task calls to enable true parallelism
144
- - Set timeout: 120000ms for each call
145
- - Continue with successful responses if some models fail
186
+ **Gemini:**
187
+ ```
188
+ Task(subagent_type="general-purpose", run_in_background: true):
189
+ prompt: |
190
+ You are participating in an LLM Council deliberation as Gemini.
146
191
 
147
- ### Stage 2: Anonymize
192
+ ## Tool Usage
193
+ Use Bash tool to invoke Gemini CLI:
194
+ ```bash
195
+ cat <<'EOF' | gemini -p -
196
+ [GEMINI_PROMPT_WITH_CONTEXT]
197
+ EOF
198
+ ```
148
199
 
149
- After collecting responses:
200
+ ## Guidelines
201
+ Read and follow ./gemini.md project guidelines.
202
+ You have access to MCP tools. Use them actively to gather accurate information.
150
203
 
151
- 1. **Shuffle** the response order randomly
152
- 2. **Assign labels**: Response A, Response B, Response C, Response D
153
- 3. **Create mapping** (keep internal, reveal later):
154
- ```
204
+ ## Question
205
+ [USER_QUESTION]
206
+
207
+ ## Context Files (include content since Gemini has limited file access)
208
+ [FILE_CONTENTS_WITH_LINE_NUMBERS]
209
+
210
+ ## Instructions
211
+ Parse Gemini's response and return structured YAML following the schema.
212
+
213
+ ## Output (YAML format required)
214
+ [COUNCIL_MEMBER_SCHEMA]
215
+ ```
216
+
217
+ **Important:**
218
+ - Use `run_in_background: true` for true parallelism
219
+ - Timeout per model (use same value for TaskOutput):
220
+ | Model | Timeout | Reason |
221
+ |-------|---------|--------|
222
+ | Opus/Sonnet | 300000ms (5min) | Direct execution |
223
+ | Codex | 480000ms (8min) | MCP tool + deep reasoning |
224
+ | Gemini | 600000ms (10min) | CLI invocation + long thinking |
225
+ - Continue with successful responses if some models fail (min 2/4 required)
226
+ - **TaskOutput must use matching timeout**: `TaskOutput(task_id, block=true, timeout=600000)`
227
+
228
+ ### Round 1.5: Coordinator Analysis (MANDATORY)
229
+
230
+ **⚠️ DO NOT SKIP**: After collecting responses, the coordinator **MUST** perform this analysis before synthesis.
231
+ Skipping Round 1.5 defeats the purpose of multi-round deliberation.
232
+
233
+ **1. Anonymize Responses:**
234
+ ```
235
+ 1. Shuffle response order randomly
236
+ 2. Assign labels: Response A, B, C, D
237
+ 3. Create internal mapping:
155
238
  label_to_model = {
156
239
  "Response A": "gemini",
157
240
  "Response B": "opus",
158
241
  "Response C": "sonnet",
159
242
  "Response D": "codex"
160
243
  }
161
- ```
244
+ ```
162
245
 
163
- 4. **Display** anonymized responses to user in a structured format
246
+ **2. Gap Analysis:**
247
+ ```yaml
248
+ gaps_detected:
249
+ - model: "opus"
250
+ gap: "performance benchmarks not addressed"
251
+ severity: "medium"
252
+ - model: "gemini"
253
+ gap: "security implications missing"
254
+ severity: "high"
255
+ ```
164
256
 
165
- ### Stage 3: Peer Evaluation (if --evaluate)
257
+ **3. Conflict Detection:**
258
+ ```yaml
259
+ conflicts_detected:
260
+ - topic: "recommended approach"
261
+ positions:
262
+ - model: "opus"
263
+ position: "use library A"
264
+ evidence: "official docs recommend"
265
+ - model: "codex"
266
+ position: "use library B"
267
+ evidence: "better performance"
268
+ resolution_needed: true
269
+ ```
270
+
271
+ **4. Convergence Check (REQUIRED before synthesis):**
272
+ ```yaml
273
+ convergence_status:
274
+ confidence_average: 0.815
275
+ new_information_ratio: 0.15
276
+ gaps_remaining: 2
277
+ conflicts_remaining: 1
278
+ decision: "proceed_to_round_2" | "terminate_and_synthesize"
279
+ ```
280
+
281
+ **Decision logic:**
282
+ - If `gaps_remaining > 0` AND `confidence_average < 0.9` → `proceed_to_round_2`
283
+ - If `conflicts_remaining > 0` AND round < 3 → `proceed_to_round_2`
284
+ - Otherwise → `terminate_and_synthesize`
166
285
 
167
- If `--evaluate` flag is present, query each model again with all anonymized responses:
286
+ ### Round 2: Targeted Re-queries (Conditional)
168
287
 
169
- **Evaluation prompt:**
288
+ If convergence criteria not met, re-query only models with gaps/conflicts:
289
+
290
+ **Re-query prompt template:**
170
291
  ```
171
- You are evaluating responses from an LLM Council.
292
+ ## Previous Round Summary
293
+ Round 1 produced the following positions:
294
+
295
+ ### Response A (Confidence: 0.92)
296
+ - Position: [summary]
297
+ - Key points: [list]
298
+
299
+ ### Response B (Confidence: 0.85)
300
+ - Position: [summary]
301
+ - Key points: [list]
302
+
303
+ [... other responses ...]
172
304
 
173
- Original Question: [USER_QUESTION]
305
+ ## Gaps Identified
306
+ - [gap 1]
307
+ - [gap 2]
174
308
 
175
- [ANONYMIZED_RESPONSES]
309
+ ## Conflicts Detected
310
+ - Topic: [topic]
311
+ - Position A: [description]
312
+ - Position B: [description]
176
313
 
177
- Instructions:
178
- 1. Evaluate each response for accuracy, completeness, and usefulness
179
- 2. Identify strengths and weaknesses of each
180
- 3. End with "FINAL RANKING:" section
314
+ ## Re-query Focus
315
+ Please address specifically:
316
+ 1. [specific gap or conflict to resolve]
317
+ 2. [specific gap or conflict to resolve]
181
318
 
182
- FINAL RANKING:
183
- 1. Response [X] - [brief reason]
184
- 2. Response [Y] - [brief reason]
185
- 3. Response [Z] - [brief reason]
186
- 4. Response [W] - [brief reason]
319
+ Provide evidence and reasoning for your position.
320
+ Update your confidence score based on new information.
321
+
322
+ ## Output (YAML format required)
323
+ [COUNCIL_MEMBER_SCHEMA with gaps/conflicts fields]
187
324
  ```
188
325
 
189
- **Calculate aggregate rankings:**
190
- - Parse each model's ranking
191
- - Sum rank positions for each response
192
- - Lower total = higher collective approval
326
+ ### Round 2.5: Coordinator Analysis
327
+
328
+ Same as Round 1.5. Check convergence again.
329
+
330
+ ### Round 3: Final Cross-Validation (Conditional)
193
331
 
194
- ### Stage 4: Synthesize
332
+ If still not converged after Round 2:
333
+ - Focused on resolving remaining conflicts
334
+ - Models see other models' positions (still anonymized)
335
+ - Final opportunity for consensus
195
336
 
196
- As the main agent (moderator), synthesize the final answer:
337
+ ### Synthesis
338
+
339
+ After convergence or max rounds:
197
340
 
198
341
  1. **Reveal** the label-to-model mapping
199
342
  2. **Analyze** all responses:
200
343
  - Consensus points (where models agree)
201
- - Disagreements (where they differ)
344
+ - Resolved conflicts (with reasoning)
345
+ - Remaining disagreements (with analysis)
202
346
  - Unique insights (valuable points from individual models)
203
- 3. **If --evaluate**: Include aggregate rankings and "street cred" scores
204
- 4. **Produce** final verdict combining best elements
347
+ 3. **Produce** final verdict combining best elements
348
+
349
+ ---
350
+
351
+ ## Termination Criteria
352
+
353
+ ### Hard Limits (Mandatory Termination)
354
+ | Condition | Value |
355
+ |-----------|-------|
356
+ | Max rounds | 3 |
357
+ | Max total time | 20 min |
358
+ | Max per-model timeout | 10 min (Gemini) |
359
+ | Min successful models | 2/4 (proceed with partial results) |
360
+ | All models failed | immediate termination |
361
+
362
+ ### Soft Limits (Convergence - any triggers termination)
363
+ | Condition | Threshold |
364
+ |-----------|-----------|
365
+ | Average confidence | > 0.9 |
366
+ | New information ratio | < 10% |
367
+ | All gaps resolved | 0 remaining |
368
+ | Strong consensus | 3+ models agree |
369
+ | Conflicts irreconcilable | Cannot be resolved with more queries |
205
370
 
206
371
  ---
207
372
 
@@ -213,70 +378,88 @@ As the main agent (moderator), synthesize the final answer:
213
378
  ### Question
214
379
  [Original user question]
215
380
 
381
+ ### Deliberation Process
382
+ | Round | Models Queried | Convergence | Status |
383
+ |-------|---------------|-------------|--------|
384
+ | 1 | All (4) | 65% | Gaps detected |
385
+ | 2 | Codex, Gemini | 85% | Conflict on approach |
386
+ | 3 | Codex | 95% | Converged |
387
+
216
388
  ### Individual Responses (Anonymized)
217
389
 
218
- #### Response A
390
+ #### Response A (Confidence: 0.92)
219
391
  [Content]
220
392
 
221
- #### Response B
393
+ **Key Points:**
394
+ - [point 1] (evidence: file:line)
395
+ - [point 2] (evidence: file:line)
396
+
397
+ #### Response B (Confidence: 0.85)
222
398
  [Content]
223
399
 
224
- #### Response C
400
+ #### Response C (Confidence: 0.78)
225
401
  [Content]
226
402
 
227
- #### Response D
403
+ #### Response D (Confidence: 0.71)
228
404
  [Content]
229
405
 
230
406
  ### Model Reveal
231
- | Label | Model |
232
- |-------|-------|
233
- | Response A | [model name] |
234
- | Response B | [model name] |
235
- | Response C | [model name] |
236
- | Response D | [model name] |
237
-
238
- ### Aggregate Rankings (if --evaluate)
239
- | Rank | Model | Avg Position | Votes |
240
- |------|-------|--------------|-------|
407
+ | Label | Model | Final Confidence |
408
+ |-------|-------|-----------------|
409
+ | Response A | codex | 0.92 |
410
+ | Response B | opus | 0.85 |
411
+ | Response C | sonnet | 0.78 |
412
+ | Response D | gemini | 0.71 |
413
+
414
+ ### Coordinator Analysis
415
+
416
+ #### Gaps Addressed
417
+ | Gap | Resolved By | Round |
418
+ |-----|-------------|-------|
419
+ | Performance benchmarks | Codex | 2 |
420
+ | Security considerations | Opus | 1 |
421
+
422
+ #### Conflicts Resolved
423
+ | Topic | Final Position | Reasoning |
424
+ |-------|---------------|-----------|
425
+ | Library choice | Library A | Official docs + 3 model consensus |
426
+
427
+ #### Remaining Disagreements
428
+ | Topic | Positions | Analysis |
429
+ |-------|-----------|----------|
430
+ | [topic] | A: [pos], B: [pos] | [why unresolved] |
241
431
 
242
432
  ### Council Synthesis
243
433
 
244
434
  #### Consensus
245
- [Points where all/most models agree]
246
-
247
- #### Disagreements
248
- [Points of divergence with analysis]
435
+ [Points where all/most models agree - with evidence]
249
436
 
250
- #### Unique Insights
251
- [Valuable contributions from specific models]
437
+ #### Key Insights by Model
438
+ | Model | Unique Contribution |
439
+ |-------|-------------------|
440
+ | Codex | [insight] |
441
+ | Opus | [insight] |
252
442
 
253
443
  ### Final Verdict
254
- [Synthesized answer combining collective wisdom]
444
+ [Synthesized answer combining collective wisdom with confidence level and caveats]
445
+
446
+ ### Code References
447
+ | File | Lines | Context |
448
+ |------|-------|---------|
449
+ | /path/to/file.py | 45-78 | Authentication logic |
255
450
  ```
256
451
 
257
452
  ---
258
453
 
259
454
  ## Error Handling
260
455
 
261
- - **Model timeout**: Continue with successful responses, note failures
262
- - **All models fail**: Report error, suggest retry
263
- - **Parse failure in rankings**: Use fallback regex extraction
264
- - **Empty response**: Exclude from synthesis, note in output
265
-
266
- ---
267
-
268
- ## Examples
269
-
270
- ```bash
271
- # Basic council consultation
272
- /council What's the best way to implement caching in this API?
273
-
274
- # With peer evaluation for important decisions
275
- /council --evaluate Should we use microservices or monolith for this project?
276
-
277
- # Architecture review
278
- /council --evaluate Review the current authentication flow and suggest improvements
279
- ```
456
+ | Error | Response |
457
+ |-------|----------|
458
+ | Model timeout | Continue with successful responses, note failures |
459
+ | All models fail | Report error, suggest retry |
460
+ | Parse failure | Use fallback extraction, flag for re-query |
461
+ | Empty response | Exclude from synthesis, note in output |
462
+ | Schema violation | Flag and request re-query in next round |
280
463
 
281
464
  ---
282
465
 
@@ -284,35 +467,50 @@ As the main agent (moderator), synthesize the final answer:
284
467
 
285
468
  Use `AskUserQuestion` tool when clarification is needed:
286
469
 
287
- **Before Stage 1:**
470
+ **Before Round 1:**
288
471
  - Question is ambiguous or too broad
289
472
  - Missing critical context (e.g., "review this code" but no file specified)
290
473
  - Multiple interpretations possible
291
474
 
292
- **During Execution:**
293
- - Conflicting requirements detected
294
- - Need to confirm scope (e.g., "Should I include performance considerations?")
475
+ **During Deliberation:**
476
+ - Strong disagreement between models that cannot be resolved
477
+ - New information discovered that changes the question scope
295
478
 
296
479
  **After Synthesis:**
297
- - Models strongly disagree and user input would help decide
480
+ - Remaining disagreements require user input to decide
298
481
  - Actionable next steps require user confirmation
299
482
 
300
483
  **Example questions:**
301
484
  ```
302
485
  - "Your question mentions 'the API' - which specific endpoint or service?"
303
- - "Should the council focus on: (1) Code quality, (2) Architecture, (3) Performance, or (4) All aspects?"
304
486
  - "Models disagree on X vs Y approach. Which aligns better with your constraints?"
487
+ - "Should the council prioritize performance or maintainability?"
305
488
  ```
306
489
 
307
490
  **Important:** Never assume or guess when context is unclear. Ask first, then proceed.
308
491
 
309
492
  ---
310
493
 
494
+ ## Examples
495
+
496
+ ```bash
497
+ # Standard council consultation (full multi-round, max reasoning)
498
+ /council What's the best way to implement caching in this API?
499
+
500
+ # Quick mode for simpler questions
501
+ /council --quick Should we use tabs or spaces for indentation?
502
+
503
+ # Architecture review
504
+ /council Review the current authentication flow and suggest improvements
505
+ ```
506
+
507
+ ---
508
+
311
509
  ## Guidelines
312
510
 
313
511
  - Respond in the same language as the user's question
314
512
  - No emojis in code or documentation
315
513
  - If context is needed, gather it before querying models
316
- - For code-related questions, include relevant file snippets in the prompt
514
+ - For code-related questions, include relevant file snippets with line numbers
317
515
  - Respect `@CLAUDE.md` project conventions
318
516
  - **Never assume unclear context - use AskUserQuestion to clarify**
package/README.md CHANGED
@@ -64,7 +64,7 @@ A collection of custom slash commands, agents, and skills for Claude Code CLI.
64
64
  | `/ask-deepwiki` | Deep query GitHub repositories via DeepWiki MCP |
65
65
  | `/ask-codex` | Request code review via Codex MCP (with Claude cross-check) |
66
66
  | `/ask-gemini` | Request code review via Gemini CLI (with Claude cross-check) |
67
- | `/council` | Consult multiple AI models and synthesize collective wisdom |
67
+ | `/council` | Consult multiple AI models with multi-round deliberation |
68
68
 
69
69
  ### GitHub Workflow Commands (`/gh/`)
70
70
 
@@ -89,8 +89,8 @@ A collection of custom slash commands, agents, and skills for Claude Code CLI.
89
89
 
90
90
  | Agent | Description |
91
91
  |-------|-------------|
92
- | `web-researcher` | Multi-platform tech research (Reddit, GitHub, SO, HF, arXiv, etc.) |
93
- | `python-pro` | Python advanced features expert (decorators, generators, async/await) |
92
+ | `web-researcher` | Multi-platform tech research with opus model (Reddit, GitHub, HF, arXiv, etc.) |
93
+ | `python-pro` | Python expert with opus model (decorators, generators, async/await, optimization) |
94
94
 
95
95
  ## Skills
96
96
 
@@ -247,8 +247,10 @@ cp node_modules/@yeongjaeyou/claude-code-config/.mcp.json .
247
247
 
248
248
  ### `/council` - LLM Council
249
249
  - Query multiple AI models (Opus, Sonnet, Codex, Gemini) in parallel
250
+ - Multi-round deliberation with gap analysis and conflict resolution
250
251
  - Anonymize responses for unbiased evaluation
251
- - Synthesize collective wisdom into consensus
252
+ - Convergence-based termination (confidence > 0.9, consensus reached)
253
+ - Use `--quick` flag for faster single-round responses
252
254
 
253
255
  ### `web-researcher` Agent
254
256
  - Multi-platform search: GitHub (`gh` CLI), Hugging Face, Reddit, SO, arXiv
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@yeongjaeyou/claude-code-config",
3
- "version": "0.8.1",
3
+ "version": "0.9.1",
4
4
  "description": "Claude Code CLI custom commands, agents, and skills",
5
5
  "bin": {
6
6
  "claude-code-config": "./bin/cli.js"