index

Index Skill

Build the symbol index. Run once before using fetch or search, and re-run when code changes significantly.

Pre-flight

# Check config exists
cat .claude/codemunch/config.json 2>/dev/null || echo "NOT_CONFIGURED"
# If missing, run detect-lsp skill first

# Create index directory
mkdir -p .claude/codemunch

# Add to gitignore
grep -q ".claude/codemunch" .gitignore 2>/dev/null || echo ".claude/codemunch" >> .gitignore

---

Engine A — LSP Indexing (primary, when available)

LSP doesn't have a batch "index all files" API, but we can use it for on-demand precise symbol lookup. For the index build, use textDocument/documentSymbol on each file.

For each source file in the project:

# Get list of source files (respects .gitignore)
rg --files --type [lang] 2>/dev/null || \
  find . -name "*.[ext]" \
    -not -path "*/node_modules/*" \
    -not -path "*/.git/*" \
    -not -path "*/target/*" \
    -not -path "*/__pycache__/*" \
    -not -path "*/.claude/codemunch/*"

Use Claude Code's built-in LSP tool (/lsp) to call textDocument/documentSymbol for each file. This returns all symbols with:

• Name
• Kind (function, class, method, variable, constant, etc.)
• Range (start line, end line)
• Detail (type signature for typed languages)

Filter noise kinds: Skip noise kinds: constant, property, variable, enumerator. These inflate the index 15x without adding useful navigation value. Only keep symbols where kind is one of: function, method, class, interface, type, enum, namespace.

Store each kept symbol as:

{
  "name": "validateToken",
  "kind": "function",
  "file": "src/auth/tokens.ts",
  "start_line": 142,
  "end_line": 163,
  "signature": "validateToken(token: string): Promise<User | null>",
  "container": "AuthService"
}

---

Engine B — ctags Indexing (fallback)

When LSP is unavailable for a language:

# Universal ctags — generates precise symbol table
universal-ctags \
  --recurse \
  --output-format=json \
  --fields=+n+e+K+S \
  --extras=+q \
  --exclude=node_modules \
  --exclude=.git \
  --exclude=target \
  --exclude=__pycache__ \
  --exclude=.claude/codemunch \
  --languages=[detected_languages] \
  -f .claude/codemunch/ctags.json \
  . 2>/dev/null

# Parse ctags output into codemunch index format
# ctags JSON fields: name, path, pattern, kind, line, end, signature
# Filter: skip noise kinds (constant, property, variable, enumerator)
# Only keep: function, method, class, interface, type, enum, namespace