@nahisaho/shikigami 1.41.0 → 1.44.0

package/AGENTS.md

@@ -32,7 +32,7 @@ VIOLATION = WORKFLOW FAILURE
 
 | Item | Value |
 |------|-------|
-| Version | 1.41.0 |
+| Version | 1.42.0 |
 | Agent Skills | 4 |
 | MCP Tools | 11 |
 | Frameworks | 53 |
@@ -146,14 +146,53 @@ FORBIDDEN: Web search, using search tools, or gathering information before proje
 
 ### Prohibition 3: save_research Required After search/visit
 
+**🚨🚨🚨 EVERY search/visit MUST be followed by save_research 🚨🚨🚨**
+
 ```
 WHEN: Executed search or visit MCP tool
-DO: IMMEDIATELY call save_research MCP tool (within the same response)
-EXAMPLE:
-  1. search(query: "AI market trends")
-  2. save_research(content: <search results>, source: "search", query: "AI market trends")
-FORBIDDEN: Proceeding to next process without saving
-CONSEQUENCE: Research data will be lost if not saved
+DO: IMMEDIATELY call save_research MCP tool (within the SAME response, not later)
+
+MANDATORY PATTERN - NO EXCEPTIONS:
+  search(...) → save_research(...) → next action
+  visit(...)  → save_research(...) → next action
+
+EXAMPLE 1 (Search):
+  1. search(query: ["AI市場動向", "AI market trends"])
+  2. save_research(
+       content: "<検索結果の全文をここに>",
+       source: "search",
+       query: "AI市場動向"
+     )
+
+EXAMPLE 2 (Visit):
+  1. visit(url: "https://example.com/article")
+  2. save_research(
+       content: "<ページ内容の全文をここに>",
+       source: "visit",
+       query: "https://example.com/article"
+     )
+
+FORBIDDEN (IMMEDIATE VIOLATION):
+  - Calling search then another search without save_research in between
+  - Calling visit then summarizing without save_research first
+  - Proceeding to analysis without saving raw data
+  - "I'll save it later" - NO! Save IMMEDIATELY
+
+CONSEQUENCE: Research data will be lost, workflow corrupted, unable to cite sources
+```
+
+**CORRECT FLOW**:
+```
+Agent: search(query: [...])
+       → [結果を受信]
+       → save_research(content: <全結果>, source: "search", query: <query>)
+       → "検索結果を保存しました。次に..."
+```
+
+**INCORRECT FLOW (VIOLATION)**:
+```
+Agent: search(query: [...])
+       → "検索結果によると..." ← ❌ save_research を呼んでいない！
 ```
 
 ### Prohibition 4: save_prompt Required After User Input
@@ -335,7 +374,7 @@ NEXT: Phase 1 (Prompt Optimization) - NOT Phase 3 (Search)
 
 ### Phase 1: Prompt Optimization
 
-**🚨🚨🚨 THIS PHASE IS MANDATORY - NEVER SKIP 🚨🚨🚨**
+** THIS PHASE IS MANDATORY - NEVER SKIP **
 
 ```
 WHEN: Phase 0 completed (project created and set_project called)
@@ -423,15 +462,34 @@ DO:
   LOOP:
     1. Think: Analyze knowledge gaps
     2. Search: search(query: ["Japanese query", "English query"])
-    3. IMMEDIATELY: save_research(content: <search_results>, source: "search", query: <query>)
+    3. 🚨 IMMEDIATELY: save_research(content: <FULL search_results>, source: "search", query: <query>)
+       → Do NOT proceed without this step!
     4. Visit: visit(url: <target_url>)
-    5. IMMEDIATELY: save_research(content: <page_content>, source: "visit", query: <url>)
+    5. 🚨 IMMEDIATELY: save_research(content: <FULL page_content>, source: "visit", query: <url>)
+       → Do NOT proceed without this step!
     6. Report: Organize collected information
     7. Action: Determine if additional search needed
 OUTPUT: Collected information summary, source list (URL + credibility)
 ```
 
-**CRITICAL**: Steps 3 and 5 are MANDATORY. Never skip save_research calls.
+**🚨 CRITICAL - save_research is MANDATORY after EVERY search/visit**:
+- Steps 3 and 5 are NOT optional
+- You MUST call save_research IMMEDIATELY after each search/visit
+- Do NOT analyze or summarize before saving raw data
+- Do NOT call multiple searches without saving each one
+
+**CORRECT Phase 3 Flow**:
+```
+search(query: [...])
+↓
+save_research(content: <全結果>, source: "search", ...)  ← 必須！
+↓
+visit(url: ...)
+↓
+save_research(content: <全内容>, source: "visit", ...)   ← 必須！
+↓
+[分析・次のアクション]
+```
 
 **Search Rules**:
 - Parallel JP/EN search required: `["Japanese query", "English query"]`
@@ -663,6 +721,104 @@ projects/pjXXXXX_Name_YYYYMMDD/
 
 ---
 
+## Information Conflict Handling
+
+```
+WHEN: Multiple sources provide conflicting information on the same fact
+DO:
+  1. Mark the fact with "(?)" in the report
+  2. Note all conflicting values with their sources
+  3. Identify the most credible source based on Credibility Scoring
+  4. Add conflicting facts to "Information Conflicts" section at report end
+```
+
+**Conflict Notation Format**:
+```
+[Fact]: [Value](?) 
+  - Source A ([Name]): [Value A]
+  - Source B ([Name]): [Value B]
+  - Source C ([Name]): [Value C]
+  → Most credible: [Source] ([Reason])
+```
+
+**Example**:
+```
+市場規模: 500億円(?) 
+  - Source A (日経新聞): 500億円
+  - Source B (業界団体レポート): 480億円  
+  - Source C (調査会社X): 520億円
+  → Most credible: 日経新聞 (Major media, published 2025-12)
+```
+
+**Report Section Template**:
+```markdown
+## Information Conflicts
+
+| Fact | Values | Most Credible | Reason |
+|------|--------|---------------|--------|
+| 市場規模 | 480-520億円 | 500億円 (日経) | Major media, latest data |
+| 成長率 | 3.2%-4.1% | 3.5% (業界団体) | Primary source |
+```
+
+**Conflict Severity Levels**:
+| Deviation | Severity | Action |
+|-----------|----------|--------|
+| <5% | Low | Note in footnote |
+| 5-20% | Medium | Mark with (?), add to Conflicts section |
+| >20% | High | Mark with (??), investigate further, consider excluding unreliable sources |
+
+### Content/Opinion Conflicts
+
+```
+WHEN: Sources provide contradictory statements or opinions on the same topic
+DO:
+  1. Mark with "(⚡)" for factual contradictions
+  2. Mark with "(💭)" for opinion/interpretation differences
+  3. Present all viewpoints with their sources
+  4. Add to "Content Conflicts" section at report end
+```
+
+**Conflict Type Markers**:
+| Marker | Type | Description |
+|--------|------|-------------|
+| (?) | Numeric | Values differ by 5-20% |
+| (??) | Numeric (High) | Values differ by >20% |
+| (⚡) | Factual | Contradictory facts/statements |
+| (💭) | Opinion | Different interpretations/predictions |
+
+**Example (Factual Contradiction)**:
+```
+原因(⚡): 
+  - Source A (調査報告書): 「システム障害が主因」
+  - Source B (業界紙): 「人的ミスが主因」
+  → No consensus - both viewpoints presented
+```
+
+**Example (Opinion/Interpretation Difference)**:
+```
+将来予測(💭):
+  - Source A (アナリストX): 「市場は2030年まで拡大する」
+  - Source B (業界団体): 「市場は縮小傾向にある」
+  → Different interpretations noted - reader should evaluate
+```
+
+**Content Conflicts Section Template**:
+```markdown
+## Content Conflicts
+
+### Factual Contradictions (⚡)
+| Topic | Viewpoint A | Viewpoint B | Sources |
+|-------|-------------|-------------|----------|
+| 障害原因 | システム障害 | 人的ミス | 調査報告書 vs 業界紙 |
+
+### Opinion Differences (💭)
+| Topic | Viewpoint A | Viewpoint B | Sources |
+|-------|-------------|-------------|----------|
+| 市場予測 | 拡大継続 | 縮小傾向 | アナリストX vs 業界団体 |
+```
+
+---
+
 ## Academic Search Rules
 
 ```
@@ -715,4 +871,4 @@ DO: Auto-detect Shift_JIS/EUC-JP/ISO-2022-JP and convert to UTF-8
 ---
 
 **Last Updated**: 2026-01-28  
-**Version**: 1.41.0
+**Version**: 1.44.0

package/CHANGELOG.md

@@ -5,6 +5,47 @@ All notable changes to SHIKIGAMI will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.44.0] - 2026-01-28
+
+### Added
+
+- **AGENTS.md: Content/Opinion Conflict Handling**
+  - New markers for non-numeric conflicts:
+    - `(⚡)` for factual contradictions (e.g., different causes stated)
+    - `(💭)` for opinion/interpretation differences (e.g., predictions)
+  - Unified Conflict Type Markers table: (?), (??), (⚡), (💭)
+  - Content Conflicts section template with Factual/Opinion subsections
+  - Examples for both factual contradictions and opinion differences
+
+## [1.43.0] - 2026-01-28
+
+### Added
+
+- **AGENTS.md: Information Conflict Handling**
+  - New section for handling conflicting information across sources
+  - Mark facts with "(?)\" when sources disagree
+  - Conflict notation format with source comparison
+  - "Information Conflicts" table template for reports
+  - Conflict severity levels: Low (<5%), Medium (5-20%), High (>20%)
+  - High severity "(??)" marking for >20% deviations
+
+## [1.42.0] - 2026-01-28
+
+### Added
+
+- **AGENTS.md: Enforce save_research after every search/visit**
+  - Prohibition 3: Added 🚨🚨🚨 warning header
+  - Added MANDATORY PATTERN section with clear flow
+  - Added detailed EXAMPLE 1 (Search) and EXAMPLE 2 (Visit) with full parameters
+  - Added CORRECT/INCORRECT flow examples
+  - Added explicit prohibition against "I'll save it later"
+
+- **AGENTS.md: Phase 3 save_research enforcement**
+  - Added 🚨 markers to steps 3 and 5
+  - Added "Do NOT proceed without this step!" warnings
+  - Added CORRECT Phase 3 Flow diagram
+  - Expanded CRITICAL section with specific prohibitions
+
 ## [1.41.0] - 2026-01-28
 
 ### Added

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nahisaho/shikigami",
-  "version": "1.41.0",
+  "version": "1.44.0",
   "description": "GitHub Copilot Agent Skills for Deep Research & Consulting - AI-Powered Research Assistant with 50+ Consulting Frameworks",
   "keywords": [
     "github-copilot",