maestro-flow 0.5.36 → 0.5.37

package/.claude/skills/skill-iter-tune/templates/eval-prompt.md

@@ -1,134 +1,134 @@
-# Evaluation Prompt Template
-
-Phase 03 使用此模板构造 maestro delegate 提示词，让 Gemini 按多维度评估 skill 质量。
-
-## Template
-
-```
-PURPOSE: Evaluate the quality of a workflow skill by examining both its definition files and the artifacts it produced when executed against a test scenario. Provide a structured multi-dimensional score with actionable improvement suggestions.
-
-SKILL DEFINITION:
-${skillContent}
-
-TEST SCENARIO:
-${testScenario.description}
-Requirements: ${testScenario.requirements}
-Success Criteria: ${testScenario.success_criteria}
-
-ARTIFACTS PRODUCED:
-${artifactsSummary}
-
-EVALUATION CRITERIA:
-${evaluationCriteria}
-
-${previousEvalContext}
-
-TASK:
-1. Read all skill definition files and produced artifacts carefully
-2. Score each dimension on 0-100 based on the evaluation criteria:
-   - Clarity (weight 0.20): Instructions unambiguous, well-structured, easy to follow
-   - Completeness (weight 0.25): All phases, edge cases, error handling covered
-   - Correctness (weight 0.25): Logic sound, data flow consistent, no contradictions
-   - Effectiveness (weight 0.20): Produces high-quality output for the test scenario
-   - Efficiency (weight 0.10): Minimal redundancy, appropriate context usage
-3. Calculate weighted composite score
-4. List top 3 strengths
-5. List top 3-5 weaknesses with specific file:section references
-6. Provide 3-5 prioritized improvement suggestions with concrete changes
-
-MODE: analysis
-
-EXPECTED OUTPUT FORMAT (strict JSON, no markdown wrapping):
-{
-  "composite_score": <number 0-100>,
-  "dimensions": [
-    { "name": "Clarity", "id": "clarity", "score": <0-100>, "weight": 0.20, "feedback": "<specific feedback>" },
-    { "name": "Completeness", "id": "completeness", "score": <0-100>, "weight": 0.25, "feedback": "<specific feedback>" },
-    { "name": "Correctness", "id": "correctness", "score": <0-100>, "weight": 0.25, "feedback": "<specific feedback>" },
-    { "name": "Effectiveness", "id": "effectiveness", "score": <0-100>, "weight": 0.20, "feedback": "<specific feedback>" },
-    { "name": "Efficiency", "id": "efficiency", "score": <0-100>, "weight": 0.10, "feedback": "<specific feedback>" }
-  ],
-  "strengths": ["<strength 1>", "<strength 2>", "<strength 3>"],
-  "weaknesses": ["<weakness 1 with file:section reference>", "..."],
-  "suggestions": [
-    {
-      "priority": "high|medium|low",
-      "target_file": "<relative path to skill file>",
-      "description": "<what to change>",
-      "rationale": "<why this improves quality>",
-      "code_snippet": "<optional: suggested replacement content>"
-    }
-  ],
-  "chain_scores": {
-    "<skill_name>": "<number 0-100, per-skill score — only present in chain mode>"
-  }
-}
-
-CONSTRAINTS:
-- Be rigorous and specific — reference exact file paths and sections
-- Each suggestion MUST include a target_file that maps to a skill file
-- Focus suggestions on highest-impact changes first
-- Do NOT suggest changes already addressed in previous iterations
-- Output ONLY the JSON object, no surrounding text or markdown
-```
-
-## Variable Substitution
-
-| Variable | Source | Description |
-|----------|--------|-------------|
-| `${skillContent}` | Same as execute-prompt.md | 完整 skill 文件内容 |
-| `${testScenario.*}` | iteration-state.json | 测试场景信息 |
-| `${artifactsSummary}` | Phase 03 reads artifacts/ dir | 产出物文件列表 + 内容摘要 |
-| `${evaluationCriteria}` | specs/evaluation-criteria.md | 评分标准全文 |
-| `${previousEvalContext}` | 历史迭代记录 | 前几轮评估摘要（避免重复建议） |
-| `${chainContext}` | Phase 03 constructs | chain 模式下的链上下文信息 |
-
-## previousEvalContext Construction
-
-```javascript
-// Build context from prior iterations to avoid repeating suggestions
-const previousEvalContext = state.iterations.length > 0
-  ? `PREVIOUS ITERATIONS (context for avoiding duplicate suggestions):
-${state.iterations.map(iter => `
-Iteration ${iter.round}: Score ${iter.evaluation?.score || 'N/A'}
-  Applied changes: ${iter.improvement?.changes_applied?.map(c => c.summary).join('; ') || 'none'}
-  Remaining weaknesses: ${iter.evaluation?.weaknesses?.slice(0, 3).join('; ') || 'none'}
-`).join('')}
-IMPORTANT: Focus on NEW issues or issues NOT adequately addressed in previous improvements.`
-  : '';
-```
-
-## chainContext Construction
-
-```javascript
-// Build chain context for evaluation (chain mode only)
-const chainContext = state.execution_mode === 'chain'
-  ? `CHAIN CONTEXT:
-This skill chain contains ${state.chain_order.length} skills executed in order:
-${state.chain_order.map((s, i) => `${i+1}. ${s}`).join('\n')}
-Current evaluation covers the entire chain output.
-Please provide per-skill quality scores in an additional "chain_scores" field.`
-  : '';
-```
-
-## artifactsSummary Construction
-
-```javascript
-// Read manifest.json if available, otherwise list files
-const manifestPath = `${iterDir}/artifacts/manifest.json`;
-let artifactsSummary;
-
-if (fileExists(manifestPath)) {
-  const manifest = JSON.parse(Read(manifestPath));
-  artifactsSummary = manifest.artifacts.map(a =>
-    `- ${a.path}: ${a.description} (Phase ${a.phase})`
-  ).join('\n');
-} else {
-  // Fallback: list all files with first 200 lines each
-  const files = Glob(`${iterDir}/artifacts/**/*`);
-  artifactsSummary = files.map(f => {
-    const content = Read(f, { limit: 200 });
-    return `--- ${f.replace(iterDir + '/artifacts/', '')} ---\n${content}`;
-  }).join('\n\n');
-}
-```
+# Evaluation Prompt Template
+
+Phase 03 使用此模板构造 maestro delegate 提示词，让 Agy 按多维度评估 skill 质量。
+
+## Template
+
+```
+PURPOSE: Evaluate the quality of a workflow skill by examining both its definition files and the artifacts it produced when executed against a test scenario. Provide a structured multi-dimensional score with actionable improvement suggestions.
+
+SKILL DEFINITION:
+${skillContent}
+
+TEST SCENARIO:
+${testScenario.description}
+Requirements: ${testScenario.requirements}
+Success Criteria: ${testScenario.success_criteria}
+
+ARTIFACTS PRODUCED:
+${artifactsSummary}
+
+EVALUATION CRITERIA:
+${evaluationCriteria}
+
+${previousEvalContext}
+
+TASK:
+1. Read all skill definition files and produced artifacts carefully
+2. Score each dimension on 0-100 based on the evaluation criteria:
+   - Clarity (weight 0.20): Instructions unambiguous, well-structured, easy to follow
+   - Completeness (weight 0.25): All phases, edge cases, error handling covered
+   - Correctness (weight 0.25): Logic sound, data flow consistent, no contradictions
+   - Effectiveness (weight 0.20): Produces high-quality output for the test scenario
+   - Efficiency (weight 0.10): Minimal redundancy, appropriate context usage
+3. Calculate weighted composite score
+4. List top 3 strengths
+5. List top 3-5 weaknesses with specific file:section references
+6. Provide 3-5 prioritized improvement suggestions with concrete changes
+
+MODE: analysis
+
+EXPECTED OUTPUT FORMAT (strict JSON, no markdown wrapping):
+{
+  "composite_score": <number 0-100>,
+  "dimensions": [
+    { "name": "Clarity", "id": "clarity", "score": <0-100>, "weight": 0.20, "feedback": "<specific feedback>" },
+    { "name": "Completeness", "id": "completeness", "score": <0-100>, "weight": 0.25, "feedback": "<specific feedback>" },
+    { "name": "Correctness", "id": "correctness", "score": <0-100>, "weight": 0.25, "feedback": "<specific feedback>" },
+    { "name": "Effectiveness", "id": "effectiveness", "score": <0-100>, "weight": 0.20, "feedback": "<specific feedback>" },
+    { "name": "Efficiency", "id": "efficiency", "score": <0-100>, "weight": 0.10, "feedback": "<specific feedback>" }
+  ],
+  "strengths": ["<strength 1>", "<strength 2>", "<strength 3>"],
+  "weaknesses": ["<weakness 1 with file:section reference>", "..."],
+  "suggestions": [
+    {
+      "priority": "high|medium|low",
+      "target_file": "<relative path to skill file>",
+      "description": "<what to change>",
+      "rationale": "<why this improves quality>",
+      "code_snippet": "<optional: suggested replacement content>"
+    }
+  ],
+  "chain_scores": {
+    "<skill_name>": "<number 0-100, per-skill score — only present in chain mode>"
+  }
+}
+
+CONSTRAINTS:
+- Be rigorous and specific — reference exact file paths and sections
+- Each suggestion MUST include a target_file that maps to a skill file
+- Focus suggestions on highest-impact changes first
+- Do NOT suggest changes already addressed in previous iterations
+- Output ONLY the JSON object, no surrounding text or markdown
+```
+
+## Variable Substitution
+
+| Variable | Source | Description |
+|----------|--------|-------------|
+| `${skillContent}` | Same as execute-prompt.md | 完整 skill 文件内容 |
+| `${testScenario.*}` | iteration-state.json | 测试场景信息 |
+| `${artifactsSummary}` | Phase 03 reads artifacts/ dir | 产出物文件列表 + 内容摘要 |
+| `${evaluationCriteria}` | specs/evaluation-criteria.md | 评分标准全文 |
+| `${previousEvalContext}` | 历史迭代记录 | 前几轮评估摘要（避免重复建议） |
+| `${chainContext}` | Phase 03 constructs | chain 模式下的链上下文信息 |
+
+## previousEvalContext Construction
+
+```javascript
+// Build context from prior iterations to avoid repeating suggestions
+const previousEvalContext = state.iterations.length > 0
+  ? `PREVIOUS ITERATIONS (context for avoiding duplicate suggestions):
+${state.iterations.map(iter => `
+Iteration ${iter.round}: Score ${iter.evaluation?.score || 'N/A'}
+  Applied changes: ${iter.improvement?.changes_applied?.map(c => c.summary).join('; ') || 'none'}
+  Remaining weaknesses: ${iter.evaluation?.weaknesses?.slice(0, 3).join('; ') || 'none'}
+`).join('')}
+IMPORTANT: Focus on NEW issues or issues NOT adequately addressed in previous improvements.`
+  : '';
+```
+
+## chainContext Construction
+
+```javascript
+// Build chain context for evaluation (chain mode only)
+const chainContext = state.execution_mode === 'chain'
+  ? `CHAIN CONTEXT:
+This skill chain contains ${state.chain_order.length} skills executed in order:
+${state.chain_order.map((s, i) => `${i+1}. ${s}`).join('\n')}
+Current evaluation covers the entire chain output.
+Please provide per-skill quality scores in an additional "chain_scores" field.`
+  : '';
+```
+
+## artifactsSummary Construction
+
+```javascript
+// Read manifest.json if available, otherwise list files
+const manifestPath = `${iterDir}/artifacts/manifest.json`;
+let artifactsSummary;
+
+if (fileExists(manifestPath)) {
+  const manifest = JSON.parse(Read(manifestPath));
+  artifactsSummary = manifest.artifacts.map(a =>
+    `- ${a.path}: ${a.description} (Phase ${a.phase})`
+  ).join('\n');
+} else {
+  // Fallback: list all files with first 200 lines each
+  const files = Glob(`${iterDir}/artifacts/**/*`);
+  artifactsSummary = files.map(f => {
+    const content = Read(f, { limit: 200 });
+    return `--- ${f.replace(iterDir + '/artifacts/', '')} ---\n${content}`;
+  }).join('\n\n');
+}
+```

package/.claude/skills/skill-iter-tune/templates/execute-prompt.md

@@ -1,97 +1,97 @@
-# Execute Prompt Template
-
-Phase 02 使用此模板构造 maestro delegate 提示词，让 Claude 模拟执行 skill 并产出所有预期产物。
-
-## Template
-
-```
-PURPOSE: Simulate executing the following workflow skill against a test scenario. Produce all expected output artifacts as if the skill were invoked with the given input. This is for evaluating skill quality.
-
-SKILL CONTENT:
-${skillContent}
-
-TEST SCENARIO:
-Description: ${testScenario.description}
-Input Arguments: ${testScenario.input_args}
-Requirements: ${testScenario.requirements}
-Success Criteria: ${testScenario.success_criteria}
-
-TASK:
-1. Study the complete skill structure (SKILL.md + all phase files)
-2. Follow the skill's execution flow sequentially (Phase 1 → Phase N)
-3. For each phase, produce the artifacts that phase would generate
-4. Write all output artifacts to the current working directory
-5. Create a manifest.json listing all produced artifacts with descriptions
-
-MODE: write
-
-CONTEXT: @**/*
-
-EXPECTED:
-- All artifacts the skill would produce for this test scenario
-- Each artifact in its correct relative path
-- A manifest.json at root: { "artifacts": [{ "path": "...", "description": "...", "phase": N }] }
-
-CONSTRAINTS:
-- Follow the skill execution flow exactly — do not skip or reorder phases
-- Produce realistic, high-quality output (not placeholder content)
-- If the skill requires user interaction (AskUserQuestion), use reasonable defaults
-- If the skill invokes external tools/CLIs, document what would be called but produce expected output directly
-```
-
-## Variable Substitution
-
-| Variable | Source | Description |
-|----------|--------|-------------|
-| `${skillContent}` | Phase 02 reads all skill files | 完整 SKILL.md + phase 文件内容，用 markdown headers 分隔 |
-| `${testScenario.description}` | iteration-state.json | 用户描述的测试场景 |
-| `${testScenario.input_args}` | iteration-state.json | 模拟传给 skill 的参数 |
-| `${testScenario.requirements}` | iteration-state.json | 质量要求列表 |
-| `${testScenario.success_criteria}` | iteration-state.json | 成功标准定义 |
-
-## Chain Mode Extension
-
-When running in chain mode, the template is invoked once per skill in `chain_order`. Each invocation includes:
-
-### Additional Variable
-
-| Variable | Source | Description |
-|----------|--------|-------------|
-| `${previousChainOutput}` | Phase 02 chain loop | 前序 skill 的 artifacts 摘要 (chain 模式下非首个 skill) |
-
-### Chain Prompt Modification
-
-When `execution_mode === 'chain'`, the prompt includes:
-
-```
-PREVIOUS CHAIN OUTPUT (from upstream skill "${previousSkillName}"):
-${previousChainOutput}
-
-IMPORTANT: Use the above output as input context for this skill's execution.
-```
-
-This section is only added for skills at position 2+ in the chain. The first skill in the chain receives no upstream context.
-
-## skillContent Construction
-
-```javascript
-// Read only executable skill files and format with consistent headers
-const skillMd = Read(`${skill.path}/SKILL.md`);
-const phaseFiles = Glob(`${skill.path}/phases/*.md`).map(f => ({
-  relativePath: f.replace(skill.path + '/', ''),
-  content: Read(f)
-}));
-const specFiles = Glob(`${skill.path}/specs/*.md`).map(f => ({
-  relativePath: f.replace(skill.path + '/', ''),
-  content: Read(f)
-}));
-
-const skillContent = `
-### File: SKILL.md
-${skillMd}
-
-${phaseFiles.map(f => `### File: ${f.relativePath}\n${f.content}`).join('\n\n')}
-
-${specFiles.length > 0 ? specFiles.map(f => `### File: ${f.relativePath}\n${f.content}`).join('\n\n') : ''}
-`.trim();
-```
+# Execute Prompt Template
+
+Phase 02 使用此模板构造 maestro delegate 提示词，让 Claude 模拟执行 skill 并产出所有预期产物。
+
+## Template
+
+```
+PURPOSE: Simulate executing the following workflow skill against a test scenario. Produce all expected output artifacts as if the skill were invoked with the given input. This is for evaluating skill quality.
+
+SKILL CONTENT:
+${skillContent}
+
+TEST SCENARIO:
+Description: ${testScenario.description}
+Input Arguments: ${testScenario.input_args}
+Requirements: ${testScenario.requirements}
+Success Criteria: ${testScenario.success_criteria}
+
+TASK:
+1. Study the complete skill structure (SKILL.md + all phase files)
+2. Follow the skill's execution flow sequentially (Phase 1 → Phase N)
+3. For each phase, produce the artifacts that phase would generate
+4. Write all output artifacts to the current working directory
+5. Create a manifest.json listing all produced artifacts with descriptions
+
+MODE: write
+
+CONTEXT: @**/*
+
+EXPECTED:
+- All artifacts the skill would produce for this test scenario
+- Each artifact in its correct relative path
+- A manifest.json at root: { "artifacts": [{ "path": "...", "description": "...", "phase": N }] }
+
+CONSTRAINTS:
+- Follow the skill execution flow exactly — do not skip or reorder phases
+- Produce realistic, high-quality output (not placeholder content)
+- If the skill requires user interaction (AskUserQuestion), use reasonable defaults
+- If the skill invokes external tools/CLIs, document what would be called but produce expected output directly
+```
+
+## Variable Substitution
+
+| Variable | Source | Description |
+|----------|--------|-------------|
+| `${skillContent}` | Phase 02 reads all skill files | 完整 SKILL.md + phase 文件内容，用 markdown headers 分隔 |
+| `${testScenario.description}` | iteration-state.json | 用户描述的测试场景 |
+| `${testScenario.input_args}` | iteration-state.json | 模拟传给 skill 的参数 |
+| `${testScenario.requirements}` | iteration-state.json | 质量要求列表 |
+| `${testScenario.success_criteria}` | iteration-state.json | 成功标准定义 |
+
+## Chain Mode Extension
+
+When running in chain mode, the template is invoked once per skill in `chain_order`. Each invocation includes:
+
+### Additional Variable
+
+| Variable | Source | Description |
+|----------|--------|-------------|
+| `${previousChainOutput}` | Phase 02 chain loop | 前序 skill 的 artifacts 摘要 (chain 模式下非首个 skill) |
+
+### Chain Prompt Modification
+
+When `execution_mode === 'chain'`, the prompt includes:
+
+```
+PREVIOUS CHAIN OUTPUT (from upstream skill "${previousSkillName}"):
+${previousChainOutput}
+
+IMPORTANT: Use the above output as input context for this skill's execution.
+```
+
+This section is only added for skills at position 2+ in the chain. The first skill in the chain receives no upstream context.
+
+## skillContent Construction
+
+```javascript
+// Read only executable skill files and format with consistent headers
+const skillMd = Read(`${skill.path}/SKILL.md`);
+const phaseFiles = Glob(`${skill.path}/phases/*.md`).map(f => ({
+  relativePath: f.replace(skill.path + '/', ''),
+  content: Read(f)
+}));
+const specFiles = Glob(`${skill.path}/specs/*.md`).map(f => ({
+  relativePath: f.replace(skill.path + '/', ''),
+  content: Read(f)
+}));
+
+const skillContent = `
+### File: SKILL.md
+${skillMd}
+
+${phaseFiles.map(f => `### File: ${f.relativePath}\n${f.content}`).join('\n\n')}
+
+${specFiles.length > 0 ? specFiles.map(f => `### File: ${f.relativePath}\n${f.content}`).join('\n\n') : ''}
+`.trim();
+```

package/.claude/skills/skill-simplify/SKILL.md

@@ -1,63 +1,63 @@
----
-name: skill-simplify
-description: SKILL.md simplification with functional integrity verification. Analyze redundancy, optimize content, check no functionality lost. Triggers on "simplify skill", "optimize skill", "skill-simplify".
-allowed-tools: AskUserQuestion, Read, Write, Edit, Bash, Glob, Grep
----
-
-# Skill Simplify
-
-Three-phase pipeline: analyze functional inventory, apply optimization rules, verify integrity.
-
-**Phase Reference Documents** (read on-demand):
-
-| Phase | Document | Purpose |
-|-------|----------|---------|
-| 1 | [phases/01-analysis.md](phases/01-analysis.md) | Extract functional inventory, identify redundancy, validate pseudo-code format |
-| 2 | [phases/02-optimize.md](phases/02-optimize.md) | Apply simplification rules, fix format issues |
-| 3 | [phases/03-check.md](phases/03-check.md) | Verify functional integrity, validate format |
-
-## 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 "skill design optimization" --json` — top 5 entries as prior context
-4. All optional — proceed without if unavailable
-
-## Input Processing
-
-```javascript
-const targetPath = input.trim()
-const targetFile = targetPath.endsWith('.md') ? targetPath : `${targetPath}/SKILL.md`
-const originalContent = Read(targetFile)
-const originalLineCount = originalContent.split('\n').length
-```
-
-## TodoWrite Pattern
-
-```javascript
-TodoWrite({ todos: [
-  { content: `Phase 1: Analyzing ${targetFile}`, status: "in_progress", activeForm: "Extracting functional inventory" },
-  { content: "Phase 2: Optimize", status: "pending" },
-  { content: "Phase 3: Integrity Check", status: "pending" }
-]})
-```
-
-## Core Rules
-
-1. **Preserve ALL functional elements**: Code blocks with logic, agent calls, data structures, routing, error handling, input/output specs
-2. **Only reduce descriptive content**: Flowcharts, verbose comments, duplicate sections, examples that repeat logic
-3. **Never summarize algorithm logic**: If-else branches, function bodies, schemas must remain verbatim
-4. **Classify code blocks**: Distinguish `functional` (logic, routing, schemas) from `descriptive` (ASCII art, examples, display templates) — only descriptive blocks may be deleted
-5. **Merge equivalent variants**: Single/multi-perspective templates differing only by a parameter → one template with variant comment
-6. **Fix format issues**: Nested backtick template literals in code fences → convert to prose; hardcoded option lists → flag for dynamic generation; workflow handoff references → ensure execution steps present
-7. **Validate pseudo-code**: Check bracket matching, variable consistency, structural completeness
-8. **Quantitative verification**: Phase 3 counts must match Phase 1 counts for functional categories; descriptive block decreases are expected
-
-## Error Handling
-
-| Error | Resolution |
-|-------|------------|
-| Target file not found | Report error, stop |
-| Check FAIL (missing functional elements) | Show delta, revert to original, report which elements lost |
-| Check WARN (descriptive decrease or merge) | Show delta with justification |
-| Format issues found | Report in check, fix in Phase 2 |
+---
+name: skill-simplify
+description: SKILL.md simplification with functional integrity verification. Analyze redundancy, optimize content, check no functionality lost. Triggers on "simplify skill", "optimize skill", "skill-simplify".
+allowed-tools: AskUserQuestion, Read, Write, Edit, Bash, Glob, Grep
+---
+
+# Skill Simplify
+
+Three-phase pipeline: analyze functional inventory, apply optimization rules, verify integrity.
+
+**Phase Reference Documents** (read on-demand):
+
+| Phase | Document | Purpose |
+|-------|----------|---------|
+| 1 | [phases/01-analysis.md](phases/01-analysis.md) | Extract functional inventory, identify redundancy, validate pseudo-code format |
+| 2 | [phases/02-optimize.md](phases/02-optimize.md) | Apply simplification rules, fix format issues |
+| 3 | [phases/03-check.md](phases/03-check.md) | Verify functional integrity, validate format |
+
+## 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 "skill design optimization" --json` — top 5 entries as prior context
+4. All optional — proceed without if unavailable
+
+## Input Processing
+
+```javascript
+const targetPath = input.trim()
+const targetFile = targetPath.endsWith('.md') ? targetPath : `${targetPath}/SKILL.md`
+const originalContent = Read(targetFile)
+const originalLineCount = originalContent.split('\n').length
+```
+
+## TodoWrite Pattern
+
+```javascript
+TodoWrite({ todos: [
+  { content: `Phase 1: Analyzing ${targetFile}`, status: "in_progress", activeForm: "Extracting functional inventory" },
+  { content: "Phase 2: Optimize", status: "pending" },
+  { content: "Phase 3: Integrity Check", status: "pending" }
+]})
+```
+
+## Core Rules
+
+1. **Preserve ALL functional elements**: Code blocks with logic, agent calls, data structures, routing, error handling, input/output specs
+2. **Only reduce descriptive content**: Flowcharts, verbose comments, duplicate sections, examples that repeat logic
+3. **Never summarize algorithm logic**: If-else branches, function bodies, schemas must remain verbatim
+4. **Classify code blocks**: Distinguish `functional` (logic, routing, schemas) from `descriptive` (ASCII art, examples, display templates) — only descriptive blocks may be deleted
+5. **Merge equivalent variants**: Single/multi-perspective templates differing only by a parameter → one template with variant comment
+6. **Fix format issues**: Nested backtick template literals in code fences → convert to prose; hardcoded option lists → flag for dynamic generation; workflow handoff references → ensure execution steps present
+7. **Validate pseudo-code**: Check bracket matching, variable consistency, structural completeness
+8. **Quantitative verification**: Phase 3 counts must match Phase 1 counts for functional categories; descriptive block decreases are expected
+
+## Error Handling
+
+| Error | Resolution |
+|-------|------------|
+| Target file not found | Report error, stop |
+| Check FAIL (missing functional elements) | Show delta, revert to original, report which elements lost |
+| Check WARN (descriptive decrease or merge) | Show delta with justification |
+| Format issues found | Report in check, fix in Phase 2 |