doc-ears-fixer（医生耳朵固定器）

doc-ears-fixer

Purpose

Automated fix skill that reads the latest audit/review report and applies fixes to EARS (Easy Approach to Requirements Syntax) documents. This skill bridges the gap between doc-ears-reviewer/doc-ears-audit (which identify issues) and the corrected EARS, enabling iterative improvement cycles.

Layer: 3 (EARS Quality Improvement)

Upstream: EARS document, Audit/Review Report (EARS-NN.A_audit_report_vNNN.md preferred, EARS-NN.R_review_report_vNNN.md legacy), PRD (source requirements)

Downstream: Fixed EARS, Fix Report (EARS-NN.F_fix_report_vNNN.md)

When to Use This Skill

Use doc-ears-fixer when:

• After Review: Run after doc-ears-reviewer identifies issues
• Iterative Improvement: Part of Review -> Fix -> Review cycle
• Automated Pipeline: CI/CD integration for quality gates
• Batch Fixes: Apply fixes to multiple EARS based on review reports

Do NOT use when:

• No audit/review report exists (run doc-ears-audit or doc-ears-reviewer first)
• Creating new EARS (use doc-ears or doc-ears-autopilot)
• Only need validation (use doc-ears-validator)

Skill Dependencies

Workflow Overview

flowchart TD
  A[Input: EARS Path] --> B[Find Latest Audit/Review Report]
  B --> C{Report Found?}
  C -->|No| D[Run doc-ears-audit or doc-ears-reviewer First]
  C -->|Yes| E[Parse Report]

    E --> F[Categorize Issues]

    subgraph FixPhases["Fix Phases"]
        F --> F0[Phase 0: Fix Structure Violations]
        F0 --> G[Phase 1: Create Missing Files]
        G --> H[Phase 2: Fix Broken Links]
        H --> I[Phase 3: Fix Element IDs]
        I --> J[Phase 4: Fix Content Issues]
        J --> K[Phase 5: Update References]
        K --> K2[Phase 6: Handle Upstream Drift]
    end

    K2 --> L[Write Fixed EARS]
    L --> M[Generate Fix Report]
    M --> N{Re-run Review?}
    N -->|Yes| O[Invoke doc-ears-reviewer]
    O --> P{Score >= Threshold?}
    P -->|No, iterations < max| F
    P -->|Yes| Q[COMPLETE]
    N -->|No| Q

Fix Phases

Phase 0: Fix Structure Violations (CRITICAL)

Fixes EARS documents that are not in nested folders. This phase runs FIRST because all subsequent phases depend on correct folder structure.

Nested Folder Rule: ALL EARS MUST be in nested folders regardless of document size.

Required Structure:

Fix Actions:

Structure Fix Workflow:

def fix_ears_structure(ears_path: str) -> list[Fix]:
    """Fix EARS structure violations."""
    fixes = []

    filename = os.path.basename(ears_path)
    parent_folder = os.path.dirname(ears_path)

    # Extract EARS ID and slug from filename
    match = re.match(r'EARS-(\d+)_([^/]+)\.md', filename)
    if not match:
        return []  # Cannot auto-fix invalid filename

    ears_id = match.group(1)
    slug = match.group(2)
    expected_folder = f"EARS-{ears_id}_{slug}"

    # Check if already in nested folder
    if os.path.basename(parent_folder) != expected_folder:
        # Create nested folder
        new_folder = os.path.join(os.path.dirname(parent_folder), expected_folder)
        os.makedirs(new_folder, exist_ok=True)

        # Move file
        new_path = os.path.join(new_folder, filename)
        shutil.move(ears_path, new_path)
        fixes.append(f"Moved {ears_path} to {new_path}")

        # Update upstream links in moved file
        content = Path(new_path).read_text()
        updated_content = content.replace('../02_PRD/', '../../02_PRD/')
        updated_content = updated_content.replace('../01_BRD/', '../../01_BRD/')
        Path(new_path).write_text(updated_content)
        fixes.append(f"Updated relative links for nested folder structure")

    return fixes

Link Path Updates After Move:

Phase 1: Create Missing Files

Creates files that are referenced but don't exist.

Scope:

EARS Glossary Template:

---
title: "EARS-00: Requirements Glossary"
tags:
  - ears
  - glossary
  - reference
custom_fields:
  document_type: glossary
  artifact_type: EARS-REFERENCE
  layer: 3
---

# EARS-00: Requirements Glossary

Common terminology used across all EARS Requirements Documents.

## EARS Pattern Terms

| Term | Definition | Context |
|------|------------|---------|
| Ubiquitous | Always active requirement | Pattern type |
| Event-Driven | Triggered by specific event | Pattern type |
| Unwanted Behavior | Exception handling requirement | Pattern type |
| State-Driven | Active in specific state | Pattern type |
| Optional Feature | Conditionally included requirement | Pattern type |
| Complex | Multi-condition requirement | Pattern type |

## Requirement Keywords

| Term | Definition | Context |
|------|------------|---------|
| SHALL | Mandatory requirement | Normative |
| SHOULD | Recommended requirement | Advisory |
| MAY | Optional requirement | Permissive |
| MUST | Absolute requirement | Critical |

## Domain Terms

<!-- Add project-specific terminology below -->

| Term | Definition | Context |
|------|------------|---------|
| [Term] | [Definition] | [Where used] |

Pattern Placeholder Template:

---
title: "EARS Pattern: [Pattern Name]"
tags:
  - ears
  - pattern-spec
  - reference
custom_fields:
  document_type: pattern-spec
  status: placeholder
  created_by: doc-ears-fixer
---

# EARS Pattern: [Pattern Name]

> **Status**: Placeholder - Requires completion

## 1. Pattern Overview

[TODO: Document pattern overview]

## 2. Pattern Template

[TODO: Define EARS pattern template]

## 3. Pattern Examples

| Example ID | Requirement | Pattern Type |
| --- | --- | --- |
| EX-XX-01 | [Example requirement] | [Type] |

## 4. Constraints

[TODO: Document pattern constraints]

---

*Created by doc-ears-fixer as placeholder. Complete this document to resolve broken link issues.*

Phase 2: Fix Broken Links

Updates links to point to correct locations.

Fix Actions:

Path Resolution Logic:

def fix_link_path(ears_location: str, target_path: str) -> str:
    """Calculate correct relative path based on EARS location."""

    # Monolithic EARS: docs/03_EARS/EARS-01.md
    # Sectioned EARS: docs/03_EARS/EARS-01_slug/EARS-01.3_section.md

    if is_sectioned_ears(ears_location):
        # Need to go up one more level
        return "../" + calculate_relative_path(ears_location, target_path)
    else:
        return calculate_relative_path(ears_location, target_path)

PRD Link Fix:

Phase 3: Fix Element IDs

Converts invalid element IDs to correct format.

Conversion Rules:

Type Code Mapping (EARS-specific valid codes: 25, 26):

Regex Patterns:

# Find element IDs with invalid type codes for EARS
invalid_ears_type_01 = r'EARS\.(\d{2})\.01\.(\d{2})'
replacement_01 = r'EARS.\1.25.\2'

invalid_ears_type_22 = r'EARS\.(\d{2})\.22\.(\d{2})'
replacement_22 = r'EARS.\1.26.\2'

# Find legacy patterns
legacy_req = r'###\s+REQ-(\d+):'
legacy_con = r'###\s+CON-(\d+):'
legacy_pat = r'###\s+PAT-(\d+):'

Phase 4: Fix Content Issues

Addresses placeholders, incomplete content, and EARS-specific syntax issues.

Fix Actions:

Auto-Replacements:

replacements = {
    'YYYY-MM-DDTHH:MM:SS': datetime.now().strftime('%Y-%m-%dT%H:%M:%S'),
    'YYYY-MM-DD': datetime.now().strftime('%Y-%m-%d'),
    'MM/DD/YYYY': datetime.now().strftime('%m/%d/%Y'),
    '[Current date]': datetime.now().strftime('%Y-%m-%dT%H:%M:%S'),
    '[System Name]': extract_system_name_from_metadata(),
}

EARS Pattern Validation:

Phase 5: Update References

Ensures traceability and cross-references are correct.

Fix Actions:

Traceability Format:

<!-- Traceability to PRD -->
@trace: PRD-01.22.01 -> EARS-01.25.01

<!-- Reference to upstream -->
@ref: PRD-01 Section 3 (../02_PRD/PRD-01.md#3-feature-requirements)

Phase 6: Handle Upstream Drift (Auto-Merge)

Addresses issues where upstream PRD documents have changed since EARS creation. This phase implements a tiered auto-merge system based on drift percentage.

6.0.1 Hash Validation Fixes

FIX-H001: Invalid Hash Placeholder

Trigger: Hash contains placeholder instead of SHA-256

Fix:

sha256sum <upstream_file_path> | cut -d' ' -f1

Update cache with: sha256:<64_hex_output>

FIX-H002: Missing Hash Prefix

Trigger: 64 hex chars but missing sha256: prefix

Fix: Prepend sha256: to value

FIX-H003: Upstream File Not Found

Trigger: Cannot compute hash (file missing)

Fix: Set drift_detected: true, add to manual review

Upstream: PRD (Layer 2) Downstream: BDD (Layer 4) ID Pattern: EARS.NN.xxxx (e.g., EARS.01.2513)

Drift Issue Codes (from doc-ears-reviewer Check #9)

Tiered Auto-Merge System

Tier Classification:

Change Percentage Calculation

def calculate_drift_percentage(
    upstream_current: str,
    upstream_cached: str,
    referenced_sections: list[str]
) -> float:
    """
    Calculate drift percentage between cached and current upstream.

    Args:
        upstream_current: Current PRD content
        upstream_cached: Cached PRD content from last sync
        referenced_sections: List of PRD sections referenced by EARS

    Returns:
        Drift percentage (0.0 - 100.0)
    """
    # Extract referenced sections only
    current_sections = extract_sections(upstream_current, referenced_sections)
    cached_sections = extract_sections(upstream_cached, referenced_sections)

    # Calculate line-based diff
    diff = unified_diff(cached_sections.splitlines(), current_sections.splitlines())

    added_lines = sum(1 for line in diff if line.startswith('+') and not line.startswith('+++'))
    removed_lines = sum(1 for line in diff if line.startswith('-') and not line.startswith('---'))
    total_lines = max(len(cached_sections.splitlines()), 1)

    # Weighted calculation: additions weighted 1.0, removals weighted 1.5
    drift_score = (added_lines * 1.0 + removed_lines * 1.5) / total_lines * 100

    return min(drift_score, 100.0)

def determine_tier(drift_percentage: float) -> int:
    """Determine merge tier based on drift percentage."""
    if drift_percentage < 5.0:
        return 1
    elif drift_percentage <= 15.0:
        return 2
    else:
        return 3

Tier 1: Auto-Merge (< 5% Drift)

Trigger Conditions:

• Minor additions to PRD sections
• Threshold/limit updates
• Clarification text additions
• No structural changes

Auto-Merge Actions:

ID Generation for New Requirements:

def generate_next_ears_id(ears_doc: str, type_code: str = "25") -> str:
    """
    Generate next sequential EARS ID.

    ID Pattern: EARS.NN.xxxx
    - NN: Document number (01-99)
    - TT: Type code (25=Statement, 26=Constraint)
    - SS: Sequence number (01-99)

    Args:
        ears_doc: Path to EARS document
        type_code: Type code ("25" for statement, "26" for constraint)

    Returns:
        Next available EARS ID
    """
    # Extract document number from filename
    doc_num = extract_doc_number(ears_doc)  # e.g., "01" from "EARS-01.md"

    # Find highest existing sequence for this type
    pattern = rf'EARS\.{doc_num}\.{type_code}\.(\d{{2}})'
    existing_ids = re.findall(pattern, read_file(ears_doc))

    if existing_ids:
        max_seq = max(int(seq) for seq in existing_ids)
        next_seq = str(max_seq + 1).zfill(2)
    else:
        next_seq = "01"

    return f"EARS.{doc_num}.{type_code}.{next_seq}"

Version Update:

# Tier 1: Increment patch version
# 1.0.0 -> 1.0.1
current_version = "1.0.0"
new_version = increment_version(current_version, "patch")  # "1.0.1"

Tier 2: Auto-Merge with Changelog (5-15% Drift)

Trigger Conditions:

• Multiple new features in PRD
• Section reorganization (minor)
• Requirement refinements
• New constraint categories

Auto-Merge Actions:

Changelog Entry Format:

## Changelog Entry (Auto-Merge Tier 2)

**Date**: 2026-02-10T16:00:00
**Source**: PRD-01.md (v1.2.0 -> v1.3.0)
**Drift Percentage**: 8.3%

### Additions

| New ID | Statement Summary | Source PRD Section |
|--------|-------------------|-------------------|
| EARS.01.2513 | System shall support batch processing | PRD-01 Section 5.2 |
| EARS.01.2514 | System shall provide progress notifications | PRD-01 Section 5.3 |
| EARS.01.2605 | Response time constraint for batch ops | PRD-01 Section 7.1 |

### Modifications

| ID | Previous | Updated | Reason |
|----|----------|---------|--------|
| EARS.01.2507 | "within 100ms" | "within 50ms" | Performance threshold tightened |

### No Deletions (See No-Deletion Policy)

Version Update:

# Tier 2: Increment minor version
# 1.0.1 -> 1.1.0
current_version = "1.0.1"
new_version = increment_version(current_version, "minor")  # "1.1.0"

Tier 3: Archive and Regenerate (> 15% Drift)

Trigger Conditions:

• Major PRD restructuring
• Feature scope changes
• Fundamental requirement changes
• More than 15% content drift

Archive Process:

1. Create Archive Manifest:

---
title: "EARS-NN Archive Manifest"
tags:
  - ears
  - archive
  - drift-management
custom_fields:
  archive_date: "2026-02-10T16:00:00"
  archived_version: "1.1.0"
  drift_percentage: 23.5
  trigger: "Tier 3 - Major Drift"
  upstream_source: "PRD-01.md v2.0.0"
---

# EARS-NN Archive Manifest

## Archive Details

| Field | Value |
|-------|-------|
| Archive Date | 2026-02-10T16:00:00 |
| Archived Version | 1.1.0 |
| Archive Location | `archive/EARS-01_v1.1.0_20260210/` |
| Drift Percentage | 23.5% |
| Reason | Major PRD restructuring |

## Archived Files

| File | Original Location | Archive Location |
|------|-------------------|------------------|
| EARS-01.md | docs/03_EARS/ | archive/EARS-01_v1.1.0_20260210/ |
| EARS-01.1_core.md | docs/03_EARS/EARS-01_slug/ | archive/EARS-01_v1.1.0_20260210/ |
| EARS-01.2_requirements.md | docs/03_EARS/EARS-01_slug/ | archive/EARS-01_v1.1.0_20260210/ |

## Regeneration Instructions

1. Run `/doc-ears-autopilot PRD-01` to generate new EARS
2. New version will be 2.0.0
3. Review archived statements for potential reuse
4. Update downstream BDD references

1. Move to Archive:

def archive_ears_document(ears_path: str, version: str, drift_pct: float) -> str:
    """
    Archive EARS document before regeneration.

    Args:
        ears_path: Path to EARS document
        version: Current version being archived
        drift_pct: Drift percentage that triggered archive

    Returns:
        Archive directory path
    """
    timestamp = datetime.now().strftime('%Y%m%d')
    archive_dir = f"archive/EARS-{doc_num}_v{version}_{timestamp}/"

    # Create archive directory
    os.makedirs(archive_dir, exist_ok=True)

    # Copy all EARS files
    for ears_file in glob(f"{ears_dir}/EARS-{doc_num}*"):
        shutil.copy2(ears_file, archive_dir)

    # Generate archive manifest
    generate_archive_manifest(archive_dir, version, drift_pct)

    return archive_dir

1. Trigger Regeneration:

# Invoke doc-ears-autopilot with major version increment
regeneration_params = {
    "source_prd": "PRD-01",
    "version_override": "2.0.0",
    "archive_reference": archive_dir,
    "preserve_custom_statements": True
}

Version Update:

# Tier 3: Increment major version
# 1.1.0 -> 2.0.0
current_version = "1.1.0"
new_version = increment_version(current_version, "major")  # "2.0.0"

No-Deletion Policy

Critical: EARS requirements are NEVER deleted during auto-merge. Items removed from upstream PRD are marked as deprecated.

Deprecation Marker Format:

### EARS.01.2505 [DEPRECATED]

> **Deprecation Date**: 2026-02-10
> **Reason**: Removed from PRD-01 v1.3.0 (Section 4.2 deleted)
> **Downstream Impact**: BDD-01.2 scenarios affected

**Original Statement**:
When the user initiates export, the system shall generate CSV output within 5 seconds.

**Status**: Deprecated - Do not implement. Retained for traceability.

Deprecation Tracking:

def mark_deprecated(
    ears_id: str,
    reason: str,
    upstream_ref: str,
    downstream_refs: list[str]
) -> None:
    """
    Mark EARS requirement as deprecated instead of deleting.

    Args:
        ears_id: EARS ID to deprecate (e.g., "EARS.01.2505")
        reason: Reason for deprecation
        upstream_ref: PRD reference that triggered deprecation
        downstream_refs: List of downstream BDD references affected
    """
    deprecation_block = f"""
### {ears_id} [DEPRECATED]

> **Deprecation Date**: {datetime.now().strftime('%Y-%m-%d')}
> **Reason**: {reason}
> **Upstream Reference**: {upstream_ref}
> **Downstream Impact**: {', '.join(downstream_refs) if downstream_refs else 'None identified'}

**Status**: Deprecated - Do not implement. Retained for traceability.
"""
    # Insert deprecation block, preserve original statement

Enhanced Drift Cache

After processing drift issues, update .drift_cache.json:

{
  "ears_document": "EARS-01",
  "ears_version": "1.1.0",
  "ears_updated": "2026-02-10T16:00:00",
  "drift_reviewed": "2026-02-10T16:00:00",
  "upstream": {
    "document": "PRD-01.md",
    "version": "1.3.0",
    "last_sync": "2026-02-10T16:00:00"
  },
  "downstream": {
    "documents": ["BDD-01.md"],
    "notification_sent": "2026-02-10T16:05:00"
  },
  "section_hashes": {
    "PRD-01#section-3": "a1b2c3d4e5f6...",
    "PRD-01#section-5": "g7h8i9j0k1l2...",
    "PRD-01#section-7": "m3n4o5p6q7r8..."
  },
  "merge_history": [
    {
      "date": "2026-02-10T16:00:00",
      "tier": 2,
      "drift_percentage": 8.3,
      "version_before": "1.0.1",
      "version_after": "1.1.0",
      "additions": ["EARS.01.2513", "EARS.01.2514", "EARS.01.2605"],
      "modifications": ["EARS.01.2507"],
      "deprecations": []
    }
  ],
  "acknowledged_drift": [
    {
      "document": "PRD-01.md",
      "section": "section-8",
      "acknowledged_date": "2026-02-08",
      "reason": "Informational section - no EARS impact"
    }
  ]
}

Drift Acknowledgment Workflow

When drift is flagged but no EARS update is needed:

1. Run /doc-ears-fixer EARS-01 --acknowledge-drift
2. Fixer prompts: "Review drift for PRD-01.md Section 8?"
3. User confirms no EARS changes needed
4. Fixer adds to acknowledged_drift array
5. Future reviews skip this drift until upstream changes again

Downstream Notification

After any auto-merge, notify downstream BDD documents:

def notify_downstream(ears_id: str, changes: list[dict]) -> None:
    """
    Create notification for downstream BDD documents.

    Args:
        ears_id: EARS document ID (e.g., "EARS-01")
        changes: List of changes made during auto-merge
    """
    notification = {
        "type": "UPSTREAM_CHANGE",
        "source": ears_id,
        "timestamp": datetime.now().isoformat(),
        "changes": changes,
        "action_required": "Run /doc-bdd-fixer to sync"
    }

    # Write to .downstream_notifications.json
    write_notification(f"docs/04_BDD/.downstream_notifications.json", notification)

Command Usage

Basic Usage

# Fix EARS based on latest review
/doc-ears-fixer EARS-01

# Fix with explicit review report (legacy)
/doc-ears-fixer EARS-01 --review-report EARS-01.R_review_report_v001.md

# Fix with explicit audit report (preferred)
/doc-ears-fixer EARS-01 --review-report EARS-01.A_audit_report_v001.md

# Fix and re-run review
/doc-ears-fixer EARS-01 --revalidate

# Fix with iteration limit
/doc-ears-fixer EARS-01 --revalidate --max-iterations 3

Options

Report Selection Precedence:

1. Select latest report by timestamp.
2. If timestamps are equal, prefer EARS-NN.A_audit_report_vNNN.md over EARS-NN.R_review_report_vNNN.md.

Fix Types

Output Artifacts

Fix Report

Nested Folder Rule: ALL EARS use nested folders (EARS-NN_{slug}/) regardless of size. Fix reports are stored alongside the EARS document in the nested folder.

File Naming: EARS-NN.F_fix_report_vNNN.md

Location: Inside the EARS nested folder: docs/03_EARS/EARS-NN_{slug}/

Structure:

---
title: "EARS-NN.F: Fix Report v001"
tags:
  - ears
  - fix-report
  - quality-assurance
custom_fields:
  document_type: fix-report
  artifact_type: EARS-FIX
  layer: 3
  parent_doc: EARS-NN
  source_review: EARS-NN.A_audit_report_v001.md
  fix_date: "YYYY-MM-DDTHH:MM:SS"
  fix_tool: doc-ears-fixer
  fix_version: "1.0"
---

# EARS-NN Fix Report v001

## Summary

| Metric | Value |
|--------|-------|
| Source Review | EARS-NN.A_audit_report_v001.md (or legacy `EARS-NN.R_review_report_v001.md`) |
| Issues in Review | 12 |
| Issues Fixed | 10 |
| Issues Remaining | 2 (manual review required) |
| Files Created | 2 |
| Files Modified | 4 |

## Files Created

| File | Type | Location |
|------|------|----------|
| EARS-00_GLOSSARY.md | Requirements Glossary | docs/03_EARS/ |
| EARS-01_PATTERN_LIB.md | Pattern Library Placeholder | docs/03_EARS/ |

## Fixes Applied

| # | Issue Code | Issue | Fix Applied | File |
|---|------------|-------|-------------|------|
| 1 | REV-L001 | Broken glossary link | Created EARS-00_GLOSSARY.md | EARS-01.3_constraints.md |
| 2 | REV-L004 | Broken PRD reference | Updated path to ../02_PRD/PRD-01.md | EARS-01.1_core.md |
| 3 | REV-N004 | Element type 01 invalid | Converted to type 25 | EARS-01.1_core.md |
| 4 | REV-L003 | Absolute path used | Converted to relative | EARS-01.2_requirements.md |

## Issues Requiring Manual Review

| # | Issue Code | Issue | Location | Reason |
|---|------------|-------|----------|--------|
| 1 | REV-E001 | Missing EARS keyword | EARS-01.2:L45 | Requirement syntax needed |
| 2 | REV-E002 | Invalid pattern structure | EARS-01.2:L120 | Pattern refinement required |

## Validation After Fix

| Metric | Before | After | Delta |
|--------|--------|-------|-------|
| Review Score | 92 | 97 | +5 |
| Errors | 2 | 0 | -2 |
| Warnings | 4 | 1 | -3 |

## Next Steps

1. Complete EARS-01_PATTERN_LIB.md placeholder
2. Address missing EARS keywords in flagged requirements
3. Run `/doc-ears-reviewer EARS-01` to verify fixes

Integration with Autopilot

This skill is invoked by doc-ears-autopilot in the Review -> Fix cycle:

flowchart LR
    subgraph Phase5["Phase 5: Review & Fix Cycle"]
        A[doc-ears-reviewer] --> B{Score >= 85?}
        B -->|No| C[doc-ears-fixer]
        C --> D{Iteration < Max?}
        D -->|Yes| A
        D -->|No| E[Flag for Manual Review]
        B -->|Yes| F[PASS]
    end

Autopilot Integration Points:

Error Handling

Recovery Actions

Backup Strategy

Before applying any fixes:

1. Create backup in tmp/backup/EARS-NN_YYYYMMDD_HHMMSS/
2. Copy all EARS files to backup location
3. Apply fixes to original files
4. If error during fix, restore from backup

Related Skills

Version History

Implementation Plan Consistency (IPLAN-004)

• Treat plan-derived outputs as valid source mode and verify intent preservation from implementation plan scope/objectives.
• Validate upstream autopilot precedence assumption: --iplan > --ref > --prompt.
• Flag objective/scope conflicts between plan context and artifact output as blocking issues requiring clarification.
• Do not introduce legacy fallback paths such as docs-v2.0/00_REF.