repair-broken-references

When Use

Use this skill when project references have become stale:

• Documentation contains broken internal links
• External URLs return 404 errors
• Import statements reference moved or deleted modules
• Cross-references between files out of sync
• Files exist but never referenced anywhere

Do NOT use for refactoring module dependencies or redesigning information architecture. This skill repairs existing references, not restructures them.

Inputs

Parameter	Type	Required	Description
project_path	string	Yes	Absolute path to project root
check_external	boolean	No	Verify external URLs (default: true, slow)
fix_mode	enum	No	auto (fix obvious), report (document only), interactive (prompt)
orphan_threshold	integer	No	Days since last modified to flag as orphan (default: 180)

Steps

Step 1: Scan for Broken Internal Links

Find all markdown links pointing to non-existent files.

# Find all markdown files
find . -name "*.md" -type f > markdown_files.txt

# Extract all markdown links: [text](path)
grep -oP '\[.*?\]\(\K[^)]+' *.md | sort | uniq > all_links.txt

# For each link:
while read link; do
  # Skip external URLs (http/https)
  if [[ "$link" =~ ^https?:// ]]; then
    continue
  fi

  # Resolve relative path
  target=$(realpath -m "$link")

  # Check if target exists
  if [ ! -e "$target" ]; then
    echo "BROKEN: $link (referenced in $file)" >> broken_internal.txt
  fi
done < all_links.txt

Got: broken_internal.txt lists all broken internal references

If fail: realpath unavailable? Manually check each link

Step 2: Check External URLs

Verify external links still accessible (HTTP 200 response).

# Extract external URLs
grep -ohP 'https?://[^\s\)]+' *.md | sort | uniq > external_urls.txt

# Check each URL (rate-limit to avoid bans)
while read url; do
  status=$(curl -o /dev/null -s -w "%{http_code}" "$url")

  if [ "$status" -ge 400 ]; then
    echo "DEAD ($status): $url" >> dead_urls.txt
  fi

  sleep 0.5  # Rate limit
done < external_urls.txt

Got: dead_urls.txt lists URLs returning 4xx/5xx errors

If fail: curl unavailable or blocked? Use online link checker or skip

Note: Some URLs may return 403 due to bot detection but work in browsers. Manual review required.

Step 3: Find Broken Imports

Check that all import/require statements reference existing modules.

JavaScript/TypeScript:

# Find all import statements
grep -rh "^import.*from ['\"]" . | sed -E "s/.*from ['\"]([^'\"]+)['\"].*/\1/" > imports.txt

# For each import:
while read import; do
  # Skip node_modules and external packages
  if [[ "$import" =~ ^[./] ]]; then
    # Resolve to file path
    target="${import}.js"  # Try .js, .ts, .jsx, .tsx

    if [ ! -e "$target" ]; then
      echo "BROKEN IMPORT: $import" >> broken_imports.txt
    fi
  fi
done < imports.txt

Python:

# Find all import statements
grep -rh "^from .* import\|^import " . --include="*.py" | \
  sed -E "s/from ([^ ]+) import.*/\1/" | \
  sed -E "s/import ([^ ]+)/\1/" > imports.txt

# For each local import (starts with .)
# Check if module file exists

R:

# Find library() and source() calls
grep -rh "library(\\|source(" . --include="*.R" | \
  sed -E 's/.*library\("([^"]+)"\).*/\1/' > packages.txt

# For source() calls, check if file exists
# For library() calls, check if package installed
Rscript -e "installed.packages()[,'Package']" > installed_packages.txt

Got: broken_imports.txt lists all references to deleted/moved modules

If fail: Language-specific tool unavailable? Manually review recent refactoring commits

Step 4: Find Orphaned Files

Identify files that exist but are never referenced anywhere.

# Find all code files
find . -type f \( -name "*.js" -o -name "*.py" -o -name "*.R" \) > all_files.txt

# For each file:
while read file; do
  basename=$(basename "$file")

  # Search for references (import, require, source, href, link)
  refs=$(grep -r "$basename" . --exclude-dir=node_modules --exclude-dir=.git | wc -l)

  # If only 1 reference (itself):
  if [ "$refs" -le 1 ]; then
    # Check last modified date
    last_mod=$(git log -1 --format="%ci" "$file")

    # If modified more than orphan_threshold days ago
    # Flag as potential orphan
    echo "ORPHAN: $file (last modified: $last_mod)" >> orphans.txt
  fi
done < all_files.txt

Got: orphans.txt lists files not referenced elsewhere

If fail: git log fails? Use filesystem mtime instead

Note: Some files (e.g., CLI entry points, top-level scripts) are legitimately unreferenced but not orphans. Requires manual review.

Step 5: Fix Internal Links

Repair broken internal references using one of three strategies:

Strategy 1: Find Moved Files

# For each broken link, search for file by name
while read broken_link; do
  filename=$(basename "$broken_link")

  # Search for file in project
  found=$(find . -name "$filename" | head -1)

  if [ -n "$found" ]; then
    # Update link to new path
    old_path="$broken_link"
    new_path="$found"

    # Use Edit tool to replace in all markdown files
    echo "FIX: $old_path -> $new_path"
  fi
done < broken_internal.txt

Strategy 2: Create Redirect Stub

# If file was deleted intentionally, create redirect stub
echo "# Moved" > "$broken_link"
echo "This content moved to [new location](new_path.md)" >> "$broken_link"

Strategy 3: Remove Dead Link

# If content no longer exists, remove link (keep text)
# Replace [text](broken_link) with text (plain)

Got: All broken internal links either fixed, redirected, or removed

If fail: Automated fix breaks context? Escalate for manual review

Step 6: Fix Broken Imports

Update import statements to reference correct paths after moves.

JavaScript Example:

// Before (broken)
import { helper } from './utils/helper';

// After (fixed — file moved to lib/)
import { helper } from './lib/helper';

For each broken import:

1. Locate moved module (similar to Step 5)
2. Update import path in all files referencing it
3. Run linter/type checker to verify fix

Got: All imports resolve correctly; no module-not-found errors

If fail: Module truly deleted? Escalate to determine if functionality still needed

Step 7: Document Orphaned Files

For files flagged as orphans, determine disposition:

1. Keep: Legitimately unreferenced (entry points, scripts, templates)
2. Archive: Old code no longer needed but preserve history
3. Delete: Dead code with no value

# Orphaned Files Review

| File | Last Modified | Recommendation | Reason |
|------|---------------|----------------|--------|
| scripts/old_deploy.sh | 2024-01-05 | Archive | Replaced by CI/CD |
| src/legacy_api.js | 2023-06-12 | Delete | API v1 fully deprecated |
| bin/cli.py | 2025-12-01 | Keep | CLI entry point (unreferenced by design) |

Got: Orphan review document created. Automated decisions flagged for human approval.

If fail: (N/A — document even if no clear disposition)

Step 8: Generate Repair Report

Summarize all broken references and fixes applied.

# Reference Repair Report

**Date**: YYYY-MM-DD
**Project**: <project_name>
**Fix Mode**: auto | report | interactive

## Broken Internal Links

- Total: X
- Fixed: Y
- Redirected: Z
- Escalated: W

Details:
- [file.md](file.md) line 45: Fixed broken link to moved doc
- [another.md](another.md) line 12: Created redirect stub

## Dead External URLs

- Total: X
- Fixed (wayback machine): Y
- Removed: Z

Details:
- https://example.com/old-page (404) → Removed
- https://api.old.com/docs (gone) → Replaced with new docs

## Broken Imports

- Total: X
- Fixed: Y
- Escalated: Z

Details:
- src/main.js line 3: Updated import path after refactor

## Orphaned Files

- Total: X
- Kept: Y
- Archived: Z
- Escalated for review: W

See ORPHAN_REVIEW.md for full analysis.

## Validation

- [x] All tests pass after fixes
- [x] Linter reports no module-not-found errors
- [x] Dead links documented in report

Got: Report saved to REFERENCE_REPAIR_REPORT.md

If fail: (N/A — generate report regardless)

Checks

After repairs:

• No broken internal links in documentation
• Dead external URLs documented (not all fixable)
• All imports resolve correctly
• Orphaned files reviewed and dispositioned
• Tests pass after import fixes
• Linter reports no unresolved references
• Git history preserved (used git mv for any moves)

Pitfalls

1. Automatic URL Fixes Break Context: Replacing dead links with web.archive.org URLs may not be what author intended. Some links better removed.

2. Over-Aggressive Orphan Deletion: Entry points, CLI scripts, templates often unreferenced by design. Don't delete without review.

3. Import Path Assumptions: Assume all relative imports use same base path. Different module systems (CommonJS, ES6, TypeScript) handle paths differently.

4. External URL False Positives: Some sites block curl/bots but work fine in browsers. Always manually verify dead URLs.

5. Circular Reference Traps: File A imports B, B imports A. Update one breaks other. Needs simultaneous fix.

6. Ignoring Fragment Identifiers: Fix [link](#section) needs checking if #section anchor exists, not just if file exists.

7. Wrong R binary on hybrid systems: On WSL or Docker, Rscript may resolve to cross-platform wrapper instead of native R. Check with which Rscript && Rscript --version. Prefer native R binary (e.g., /usr/local/bin/Rscript on Linux/WSL) for reliability. See Setting Up Your Environment for R path configuration.

See Also

• clean-codebase — Remove dead code after confirming orphans
• tidy-project-structure — Reorganize files (may create broken references)
• escalate-issues — Route complex reference issues to specialists
• compliance/documentation-audit — Comprehensive documentation review
• web-dev/link-checker — Advanced external URL validation