maestro-flow 0.5.36 → 0.5.37

package/.claude/skills/scholar-anti-ai-writing/SKILL.md

@@ -1,173 +1,173 @@
----
-name: scholar-anti-ai-writing
-description: Remove AI writing patterns from academic prose. Detects and fixes inflated symbolism, promotional language, superficial analyses, vague attributions, AI vocabulary, and formulaic structures. Supports English and Chinese. Triggers on "remove AI patterns", "humanize text", "anti-AI polish", "去除AI写作痕迹", "人性化处理".
-allowed-tools: Read, Write, Edit, Bash, Glob, Grep, AskUserQuestion, TodoWrite
----
-
-# Scholar Anti-AI Writing
-
-Detect and eliminate AI writing patterns from academic prose. Based on Wikipedia's "Signs of AI writing" guide. Supports both English and Chinese text.
-
-**Core insight**: LLMs predict the most statistically likely outcome — creating detectable patterns. This skill identifies and rewrites those patterns while preserving meaning and adding authentic voice.
-
-## Pre-load (before execution)
-
-1. **Codebase docs**: If `.workflow/codebase/ARCHITECTURE.md` exists, read for project context
-2. **Specs**: `maestro spec load --category coding` — load coding conventions
-3. **Wiki knowledge**: `maestro search "academic writing research paper" --json` — top 5 entries as prior context
-4. All optional — proceed without if unavailable
-
-## Architecture Overview
-
-```
-User: "Remove AI patterns from my paper"
-         |
-         v
-┌──────────────────────────────────────────────────────────────┐
-│  SKILL.md (Orchestrator)                                      │
-│  Collect preferences → Dispatch phases → Track progress       │
-└──────────┬───────────────────────────────────────────────────┘
-           |
-   ┌───────┼───────────┬──────────────┐
-   v       v           v              v
-┌──────┐┌──────────┐┌──────────────┐
-│ P1   ││ P2       ││ P3           │
-│Detect││ Rewrite  ││ Validate     │
-│ &    ││ & Polish ││ & Score      │
-│Score ││          ││              │
-└──┬───┘└────┬─────┘└──────┬───────┘
-   │         │             │
-   v         v             v
- pattern   polished     final score
- report    prose        + diff report
-```
-
-## Key Design Principles
-
-1. **Pattern detection + soul injection**: Removing AI patterns is only half the job — add authentic voice.
-2. **Preserve meaning**: Core content and technical claims must remain intact.
-3. **Language-aware**: English and Chinese have different AI pattern signatures.
-4. **Scoring-driven**: Quantitative 5-dimension scoring (50-point scale) guides revision.
-5. **Academic tone**: Maintain scholarly register — humanize without becoming informal.
-
-## Interactive Preference Collection
-
-Before dispatching to any phase, collect these preferences:
-
-```
-Questions to ask the user:
-
-1. Input Source
-   "What text should I process? (file path, directory, or paste text)"
-   → inputSource
-
-2. Language
-   Options: English | Chinese | Bilingual (auto-detect per section)
-   → language
-
-3. Writing Context
-   Options: Academic Paper | Technical Report | Blog/Article | General Prose
-   → writingContext
-
-4. Aggressiveness
-   Options: Conservative (minimal changes) | Balanced (Recommended) | Aggressive (heavy rewrite)
-   → aggressiveness
-```
-
-Store responses as `antiAiPreferences` context for all phases.
-
-## Auto Mode Defaults
-
-When `workflowPreferences.autoYes === true`:
-- Process all .tex files in cwd
-- Auto-detect language
-- Academic Paper context, Balanced aggressiveness
-- Skip confirmation prompts
-
-## Execution Flow
-
-> **COMPACT DIRECTIVE**: Context compression MUST check TodoWrite phase status.
-> The phase currently marked `in_progress` is the active execution phase — preserve its FULL content.
-> Only compress phases marked `completed` or `pending`.
-
-### TodoWrite Setup
-
-```
-Anti-AI Writing:
-- [ ] Phase 1: Detect & Score — scan for AI patterns, generate initial score
-- [ ] Phase 2: Rewrite & Polish — rewrite flagged sections, add voice
-- [ ] Phase 3: Validate & Score — re-score, verify quality threshold
-```
-
-### Phase Sequence
-
-```
-Phase 1: Detect & Score
-   └─ Ref: phases/01-detect-score.md
-      ├─ Input: inputSource, language
-      └─ Output: patternReport (flagged passages + initial scores)
-
-Phase 2: Rewrite & Polish
-   └─ Ref: phases/02-rewrite-polish.md
-      ├─ Input: patternReport, writingContext, aggressiveness
-      └─ Output: polishedText (rewritten content)
-
-Phase 3: Validate & Score
-   └─ Ref: phases/03-validate-score.md
-      ├─ Input: polishedText, original text
-      └─ Output: finalReport (before/after scores, diff)
-```
-
-**Phase Reference Documents** (read on-demand when phase executes):
-
-| Phase | Document | Purpose | Compact |
-|-------|----------|---------|---------|
-| 1 | [phases/01-detect-score.md](phases/01-detect-score.md) | Scan patterns, score text | TodoWrite driven |
-| 2 | [phases/02-rewrite-polish.md](phases/02-rewrite-polish.md) | Rewrite and humanize | TodoWrite driven + sentinel |
-| 3 | [phases/03-validate-score.md](phases/03-validate-score.md) | Re-score, generate report | TodoWrite driven |
-
-**Compact Rules**:
-1. **TodoWrite `in_progress`** → preserve full content, do not compress
-2. **TodoWrite `completed`** → may compress to summary
-3. **sentinel fallback** → phases marked with sentinel contain compact sentinel; if only sentinel remains, **must immediately `Read()` to recover**
-
-## Core Rules
-
-1. **Never lose meaning**: Technical claims, data, and specific facts must survive rewriting.
-2. **Score before and after**: Every processed text gets a 5-dimension score (50-point scale).
-3. **Flag, don't force**: In Conservative mode, highlight issues but let the user decide.
-4. **Academic register**: For papers, maintain formal-but-natural tone. No slang or casual language.
-5. **Minimum threshold**: Target score >= 35/50 for academic papers, >= 40/50 for submission-ready.
-
-## Quick Scoring System (5 dimensions, 10 points each)
-
-| Dimension | Question | Target |
-|-----------|----------|--------|
-| **Directness** | Direct statements or announcements? | >= 7 |
-| **Rhythm** | Varied or metronomic? | >= 7 |
-| **Trust** | Respects reader intelligence? | >= 7 |
-| **Authenticity** | Sounds human? | >= 7 |
-| **Density** | Anything cuttable? | >= 7 |
-
-## Data Flow
-
-```
-Phase 1 ──patternReport──→ Phase 2
-Phase 2 ──polishedText──→ Phase 3
-Phase 1 ──originalText──→ Phase 3 (for comparison)
-```
-
-## Error Handling
-
-| Error | Action |
-|-------|--------|
-| File not found | Ask user for correct path |
-| Mixed languages in one file | Process each section in detected language |
-| Score below threshold after rewrite | Flag for manual review, suggest specific areas |
-| LaTeX commands broken during rewrite | Preserve LaTeX structure, only modify prose content |
-| User disagrees with changes | Offer side-by-side diff, allow per-section accept/reject |
-
-## Related Skills
-
-- **scholar-writing**: Phase 5 (Anti-AI Polish) uses this skill's patterns
-- **scholar-review**: Uses scoring to assess writing quality
+---
+name: scholar-anti-ai-writing
+description: Remove AI writing patterns from academic prose. Detects and fixes inflated symbolism, promotional language, superficial analyses, vague attributions, AI vocabulary, and formulaic structures. Supports English and Chinese. Triggers on "remove AI patterns", "humanize text", "anti-AI polish", "去除AI写作痕迹", "人性化处理".
+allowed-tools: Read, Write, Edit, Bash, Glob, Grep, AskUserQuestion, TodoWrite
+---
+
+# Scholar Anti-AI Writing
+
+Detect and eliminate AI writing patterns from academic prose. Based on Wikipedia's "Signs of AI writing" guide. Supports both English and Chinese text.
+
+**Core insight**: LLMs predict the most statistically likely outcome — creating detectable patterns. This skill identifies and rewrites those patterns while preserving meaning and adding authentic voice.
+
+## Pre-load (before execution)
+
+1. **Codebase docs**: If `.workflow/codebase/ARCHITECTURE.md` exists, read for project context
+2. **Specs**: `maestro load --type spec --category coding` — load coding conventions
+3. **Wiki knowledge**: `maestro search "academic writing research paper" --json` — top 5 entries as prior context
+4. All optional — proceed without if unavailable
+
+## Architecture Overview
+
+```
+User: "Remove AI patterns from my paper"
+         |
+         v
+┌──────────────────────────────────────────────────────────────┐
+│  SKILL.md (Orchestrator)                                      │
+│  Collect preferences → Dispatch phases → Track progress       │
+└──────────┬───────────────────────────────────────────────────┘
+           |
+   ┌───────┼───────────┬──────────────┐
+   v       v           v              v
+┌──────┐┌──────────┐┌──────────────┐
+│ P1   ││ P2       ││ P3           │
+│Detect││ Rewrite  ││ Validate     │
+│ &    ││ & Polish ││ & Score      │
+│Score ││          ││              │
+└──┬───┘└────┬─────┘└──────┬───────┘
+   │         │             │
+   v         v             v
+ pattern   polished     final score
+ report    prose        + diff report
+```
+
+## Key Design Principles
+
+1. **Pattern detection + soul injection**: Removing AI patterns is only half the job — add authentic voice.
+2. **Preserve meaning**: Core content and technical claims must remain intact.
+3. **Language-aware**: English and Chinese have different AI pattern signatures.
+4. **Scoring-driven**: Quantitative 5-dimension scoring (50-point scale) guides revision.
+5. **Academic tone**: Maintain scholarly register — humanize without becoming informal.
+
+## Interactive Preference Collection
+
+Before dispatching to any phase, collect these preferences:
+
+```
+Questions to ask the user:
+
+1. Input Source
+   "What text should I process? (file path, directory, or paste text)"
+   → inputSource
+
+2. Language
+   Options: English | Chinese | Bilingual (auto-detect per section)
+   → language
+
+3. Writing Context
+   Options: Academic Paper | Technical Report | Blog/Article | General Prose
+   → writingContext
+
+4. Aggressiveness
+   Options: Conservative (minimal changes) | Balanced (Recommended) | Aggressive (heavy rewrite)
+   → aggressiveness
+```
+
+Store responses as `antiAiPreferences` context for all phases.
+
+## Auto Mode Defaults
+
+When `workflowPreferences.autoYes === true`:
+- Process all .tex files in cwd
+- Auto-detect language
+- Academic Paper context, Balanced aggressiveness
+- Skip confirmation prompts
+
+## Execution Flow
+
+> **COMPACT DIRECTIVE**: Context compression MUST check TodoWrite phase status.
+> The phase currently marked `in_progress` is the active execution phase — preserve its FULL content.
+> Only compress phases marked `completed` or `pending`.
+
+### TodoWrite Setup
+
+```
+Anti-AI Writing:
+- [ ] Phase 1: Detect & Score — scan for AI patterns, generate initial score
+- [ ] Phase 2: Rewrite & Polish — rewrite flagged sections, add voice
+- [ ] Phase 3: Validate & Score — re-score, verify quality threshold
+```
+
+### Phase Sequence
+
+```
+Phase 1: Detect & Score
+   └─ Ref: phases/01-detect-score.md
+      ├─ Input: inputSource, language
+      └─ Output: patternReport (flagged passages + initial scores)
+
+Phase 2: Rewrite & Polish
+   └─ Ref: phases/02-rewrite-polish.md
+      ├─ Input: patternReport, writingContext, aggressiveness
+      └─ Output: polishedText (rewritten content)
+
+Phase 3: Validate & Score
+   └─ Ref: phases/03-validate-score.md
+      ├─ Input: polishedText, original text
+      └─ Output: finalReport (before/after scores, diff)
+```
+
+**Phase Reference Documents** (read on-demand when phase executes):
+
+| Phase | Document | Purpose | Compact |
+|-------|----------|---------|---------|
+| 1 | [phases/01-detect-score.md](phases/01-detect-score.md) | Scan patterns, score text | TodoWrite driven |
+| 2 | [phases/02-rewrite-polish.md](phases/02-rewrite-polish.md) | Rewrite and humanize | TodoWrite driven + sentinel |
+| 3 | [phases/03-validate-score.md](phases/03-validate-score.md) | Re-score, generate report | TodoWrite driven |
+
+**Compact Rules**:
+1. **TodoWrite `in_progress`** → preserve full content, do not compress
+2. **TodoWrite `completed`** → may compress to summary
+3. **sentinel fallback** → phases marked with sentinel contain compact sentinel; if only sentinel remains, **must immediately `Read()` to recover**
+
+## Core Rules
+
+1. **Never lose meaning**: Technical claims, data, and specific facts must survive rewriting.
+2. **Score before and after**: Every processed text gets a 5-dimension score (50-point scale).
+3. **Flag, don't force**: In Conservative mode, highlight issues but let the user decide.
+4. **Academic register**: For papers, maintain formal-but-natural tone. No slang or casual language.
+5. **Minimum threshold**: Target score >= 35/50 for academic papers, >= 40/50 for submission-ready.
+
+## Quick Scoring System (5 dimensions, 10 points each)
+
+| Dimension | Question | Target |
+|-----------|----------|--------|
+| **Directness** | Direct statements or announcements? | >= 7 |
+| **Rhythm** | Varied or metronomic? | >= 7 |
+| **Trust** | Respects reader intelligence? | >= 7 |
+| **Authenticity** | Sounds human? | >= 7 |
+| **Density** | Anything cuttable? | >= 7 |
+
+## Data Flow
+
+```
+Phase 1 ──patternReport──→ Phase 2
+Phase 2 ──polishedText──→ Phase 3
+Phase 1 ──originalText──→ Phase 3 (for comparison)
+```
+
+## Error Handling
+
+| Error | Action |
+|-------|--------|
+| File not found | Ask user for correct path |
+| Mixed languages in one file | Process each section in detected language |
+| Score below threshold after rewrite | Flag for manual review, suggest specific areas |
+| LaTeX commands broken during rewrite | Preserve LaTeX structure, only modify prose content |
+| User disagrees with changes | Offer side-by-side diff, allow per-section accept/reject |
+
+## Related Skills
+
+- **scholar-writing**: Phase 5 (Anti-AI Polish) uses this skill's patterns
+- **scholar-review**: Uses scoring to assess writing quality

package/.claude/skills/scholar-anti-ai-writing/phases/01-detect-score.md

@@ -1,126 +1,126 @@
-# Phase 1: Detect & Score
-
-## Objective
-
-Scan the input text for AI writing patterns across four categories and generate an initial quality score on five dimensions. This phase produces a pattern report that drives Phase 2 rewrites.
-
-## Execution
-
-### Step 1.1: Load Input
-
-1. Accept input as file path (.tex, .md) or pasted text
-2. Read full content into working memory
-3. Detect primary language:
-   - **English**: ASCII-dominant, Latin script
-   - **Chinese**: CJK character majority
-   - **Mixed**: Both scripts present -- note ratio and mark bilingual sections
-4. If LaTeX (.tex): parse preamble separately, focus scanning on body content only (between `\begin{document}` and `\end{document}`)
-
-### Step 1.2: Pattern Scan
-
-Scan text against reference pattern files. For each match, record: passage text, line/paragraph location, pattern ID, category, and severity (high/medium/low).
-
-#### Category A: Content Patterns
-
-Reference: `references/patterns-english.md` sections 1-6, `references/patterns-chinese.md` sections 1-6
-
-| ID | Pattern | Severity |
-|----|---------|----------|
-| C1 | Undue emphasis on significance | High |
-| C2 | Undue emphasis on notability | Medium |
-| C3 | Superficial -ing analyses | High |
-| C4 | Promotional language | High |
-| C5 | Vague attributions | Medium |
-| C6 | Formulaic challenges sections | Medium |
-
-#### Category B: Language Patterns
-
-Reference: `references/patterns-english.md` sections 7-12, `references/patterns-chinese.md` sections 7-12
-
-| ID | Pattern | Severity |
-|----|---------|----------|
-| L1 | AI vocabulary (Additionally, crucial, delve, enhance, landscape, etc.) | High |
-| L2 | Copula avoidance (serves as, stands as) | Medium |
-| L3 | Negative parallelisms | Medium |
-| L4 | Rule of three | Low |
-| L5 | Elegant variation (synonym cycling) | Medium |
-| L6 | False ranges | Low |
-
-#### Category C: Style Patterns
-
-Reference: `references/patterns-english.md` sections 13-15, `references/patterns-chinese.md` sections 13-15
-
-| ID | Pattern | Severity |
-|----|---------|----------|
-| S1 | Em dash overuse | Medium |
-| S2 | Boldface overuse | Low |
-| S3 | Inline-header vertical lists | Medium |
-
-#### Category D: Communication & Filler Patterns
-
-Reference: `references/patterns-english.md` sections 16-20, `references/patterns-chinese.md` sections 16-20
-
-| ID | Pattern | Severity |
-|----|---------|----------|
-| F1 | Collaborative artifacts (I hope this helps, Let me know) | High |
-| F2 | Knowledge cutoff disclaimers | High |
-| F3 | Filler phrases | Medium |
-| F4 | Excessive hedging | Medium |
-| F5 | Generic positive conclusions | Medium |
-
-### Step 1.3: Score
-
-Rate the text on a 1-10 scale for each dimension (10 = best). Score per section if the text has clear sections, then compute overall scores.
-
-| Dimension | 1 (worst) | 10 (best) |
-|-----------|-----------|-----------|
-| **Directness** | Every point announced before stated; throat-clearing openers everywhere | Claims made directly without pre-announcement or softening |
-| **Rhythm** | Every sentence same length and structure; metronomic cadence | Varied sentence lengths; short punchy lines mixed with longer ones |
-| **Trust** | Reader hand-held through every inference; over-explained | Facts stated; reader trusted to draw conclusions |
-| **Authenticity** | Reads like template-generated content; no human voice | Sounds like a specific person wrote it; has opinions and texture |
-| **Density** | Significant cuttable filler; redundant phrases throughout | Every word earns its place; nothing cuttable without losing meaning |
-
-**Scoring rules**:
-- Score each section independently first
-- Weight section scores by word count for overall score
-- Total possible = 50 (5 dimensions x 10 max)
-- Record specific passages that caused low scores as evidence
-
-## Output
-
-```
-patternReport:
-  language: <detected language>
-  sections:
-    - name: <section identifier>
-      wordCount: <n>
-      flaggedPassages:
-        - text: "<flagged passage>"
-          location: <line or paragraph ref>
-          patternId: <C1-C6, L1-L6, S1-S3, F1-F5>
-          category: <content|language|style|filler>
-          severity: <high|medium|low>
-          suggestion: "<brief fix direction>"
-      scores:
-        directness: <1-10>
-        rhythm: <1-10>
-        trust: <1-10>
-        authenticity: <1-10>
-        density: <1-10>
-        total: <5-50>
-  overall:
-    totalPatterns: <count>
-    bySeverity: { high: <n>, medium: <n>, low: <n> }
-    byCategory: { content: <n>, language: <n>, style: <n>, filler: <n> }
-    scores:
-      directness: <weighted avg>
-      rhythm: <weighted avg>
-      trust: <weighted avg>
-      authenticity: <weighted avg>
-      density: <weighted avg>
-      total: <weighted avg>
-```
-
-## Next Phase
-
-Pass `patternReport` to **Phase 2: Rewrite & Polish** for targeted corrections.
+# Phase 1: Detect & Score
+
+## Objective
+
+Scan the input text for AI writing patterns across four categories and generate an initial quality score on five dimensions. This phase produces a pattern report that drives Phase 2 rewrites.
+
+## Execution
+
+### Step 1.1: Load Input
+
+1. Accept input as file path (.tex, .md) or pasted text
+2. Read full content into working memory
+3. Detect primary language:
+   - **English**: ASCII-dominant, Latin script
+   - **Chinese**: CJK character majority
+   - **Mixed**: Both scripts present -- note ratio and mark bilingual sections
+4. If LaTeX (.tex): parse preamble separately, focus scanning on body content only (between `\begin{document}` and `\end{document}`)
+
+### Step 1.2: Pattern Scan
+
+Scan text against reference pattern files. For each match, record: passage text, line/paragraph location, pattern ID, category, and severity (high/medium/low).
+
+#### Category A: Content Patterns
+
+Reference: `references/patterns-english.md` sections 1-6, `references/patterns-chinese.md` sections 1-6
+
+| ID | Pattern | Severity |
+|----|---------|----------|
+| C1 | Undue emphasis on significance | High |
+| C2 | Undue emphasis on notability | Medium |
+| C3 | Superficial -ing analyses | High |
+| C4 | Promotional language | High |
+| C5 | Vague attributions | Medium |
+| C6 | Formulaic challenges sections | Medium |
+
+#### Category B: Language Patterns
+
+Reference: `references/patterns-english.md` sections 7-12, `references/patterns-chinese.md` sections 7-12
+
+| ID | Pattern | Severity |
+|----|---------|----------|
+| L1 | AI vocabulary (Additionally, crucial, delve, enhance, landscape, etc.) | High |
+| L2 | Copula avoidance (serves as, stands as) | Medium |
+| L3 | Negative parallelisms | Medium |
+| L4 | Rule of three | Low |
+| L5 | Elegant variation (synonym cycling) | Medium |
+| L6 | False ranges | Low |
+
+#### Category C: Style Patterns
+
+Reference: `references/patterns-english.md` sections 13-15, `references/patterns-chinese.md` sections 13-15
+
+| ID | Pattern | Severity |
+|----|---------|----------|
+| S1 | Em dash overuse | Medium |
+| S2 | Boldface overuse | Low |
+| S3 | Inline-header vertical lists | Medium |
+
+#### Category D: Communication & Filler Patterns
+
+Reference: `references/patterns-english.md` sections 16-20, `references/patterns-chinese.md` sections 16-20
+
+| ID | Pattern | Severity |
+|----|---------|----------|
+| F1 | Collaborative artifacts (I hope this helps, Let me know) | High |
+| F2 | Knowledge cutoff disclaimers | High |
+| F3 | Filler phrases | Medium |
+| F4 | Excessive hedging | Medium |
+| F5 | Generic positive conclusions | Medium |
+
+### Step 1.3: Score
+
+Rate the text on a 1-10 scale for each dimension (10 = best). Score per section if the text has clear sections, then compute overall scores.
+
+| Dimension | 1 (worst) | 10 (best) |
+|-----------|-----------|-----------|
+| **Directness** | Every point announced before stated; throat-clearing openers everywhere | Claims made directly without pre-announcement or softening |
+| **Rhythm** | Every sentence same length and structure; metronomic cadence | Varied sentence lengths; short punchy lines mixed with longer ones |
+| **Trust** | Reader hand-held through every inference; over-explained | Facts stated; reader trusted to draw conclusions |
+| **Authenticity** | Reads like template-generated content; no human voice | Sounds like a specific person wrote it; has opinions and texture |
+| **Density** | Significant cuttable filler; redundant phrases throughout | Every word earns its place; nothing cuttable without losing meaning |
+
+**Scoring rules**:
+- Score each section independently first
+- Weight section scores by word count for overall score
+- Total possible = 50 (5 dimensions x 10 max)
+- Record specific passages that caused low scores as evidence
+
+## Output
+
+```
+patternReport:
+  language: <detected language>
+  sections:
+    - name: <section identifier>
+      wordCount: <n>
+      flaggedPassages:
+        - text: "<flagged passage>"
+          location: <line or paragraph ref>
+          patternId: <C1-C6, L1-L6, S1-S3, F1-F5>
+          category: <content|language|style|filler>
+          severity: <high|medium|low>
+          suggestion: "<brief fix direction>"
+      scores:
+        directness: <1-10>
+        rhythm: <1-10>
+        trust: <1-10>
+        authenticity: <1-10>
+        density: <1-10>
+        total: <5-50>
+  overall:
+    totalPatterns: <count>
+    bySeverity: { high: <n>, medium: <n>, low: <n> }
+    byCategory: { content: <n>, language: <n>, style: <n>, filler: <n> }
+    scores:
+      directness: <weighted avg>
+      rhythm: <weighted avg>
+      trust: <weighted avg>
+      authenticity: <weighted avg>
+      density: <weighted avg>
+      total: <weighted avg>
+```
+
+## Next Phase
+
+Pass `patternReport` to **Phase 2: Rewrite & Polish** for targeted corrections.