ghostbuster

name	ghostbuster
model	opus
effort	high
context	fork
argument-hint	[directory to scope, or leave blank for full codebase]
allowed-tools	Read, Glob, Grep, Bash(git log:*), Bash(git diff:*), Bash(git blame:*), Bash(wc:*), Agent, mcp__tilth__*
description	Dead-code forensics and spec cross-reference. Finds unreachable, orphaned, missing, or dormant code and classifies it DEAD, ZOMBIE, GHOST, or DORMANT. Use when the user says "find dead code", "what's unused", "what can I delete", "stale specs", "spec drift", "orphaned implementations", "find zombie code", or asks what's wired up vs sitting unused. Do NOT use for code-quality review (/age) or NIH/reinvented-wheel detection (/nih-audit).

/ghostbuster — Dead Code Forensics

Find what's expired. Cross-reference against specs. Triage the remains.

Scope: $ARGUMENTS (or repo root if blank)

Phase 1: Discovery

Detect languages from file extensions in scope. Build a file inventory:

Glob: {scope}/**/*.{ts,tsx,js,jsx,py,rs,go,sh,bash}

Filter out test files, node_modules/, vendor/, target/, dist/, build/.

Phase 2: Spec & Doc Collection

Search broadly — specs and documentation both reference code symbols:

Glob: **/specs/**/*.md
Glob: **/.claude/specs/*.md
Glob: **/SPEC.md
Glob: **/spec.md
Glob: **/CLAUDE.md
Glob: **/README.md
Glob: **/CONTRIBUTING.md
Glob: **/docs/**/*.md

Read each file and extract symbol references (backtick-wrapped identifiers, code blocks, prose references to functions/types/endpoints). Build a lookup of {symbol → [file:line]}.

Phase 3: Dead Code Scan

Spawn the ghostbuster agent:

Agent(
  subagent_type="ghostbuster",
  model="sonnet",
  prompt="Run dead code forensics.
    Scope: <$ARGUMENTS or repo root>
    Languages: <detected languages>
    Slug: <slug>",
  run_in_background=false
)

The agent writes the full JSON report to $TMPDIR/ghostbuster-{slug}.json and returns a structured summary (max 2000 chars). Read the full report for detailed findings.

Phase 4: Synthesize & Present

Parse the agent's findings. Present the full report to the user:

Category explanations with examples

For each category found, give one concrete example from the findings:

DEAD — Code with zero callers and no spec mention.

Example: src/utils/legacy-parser.ts:parseLegacyFormat — 0 references, untouched since August. Safe to delete.

ZOMBIE — Spec says it should exist, but nothing calls it at runtime.

Example: src/domains/billing/invoice-export.ts:exportToQuickbooks — mentioned in billing-integrations.md but has 0 callers. Incomplete implementation or abandoned?

GHOST — Spec references a symbol that doesn't exist in the codebase.

Example: .claude/specs/billing-integrations.md:67 references validateTaxRules — deleted 6 months ago. Stale spec.

DORMANT — Code has callers, but the root of the call chain is dead.

Example: src/utils/tax-formatter.ts:formatTaxLine — only caller is validateTaxRules which is itself dead. Entire chain can go.

Actionable summary

Group findings by recommended action:

1. Delete now — DEAD findings with confidence >= 75
2. Human triage — ZOMBIE findings (spec vs reality mismatch)
3. Update specs — GHOST findings (stale references)
4. Delete chain — DORMANT findings (list the full chain)

Offer next steps

• "Delete the DEAD code? I can remove the N files/functions with confidence >= 75."
• "Update the specs to remove GHOST references?"
• "Run /xray on ZOMBIE modules to understand what's missing?"

What This Skill Never Does

• Auto-delete code without user confirmation
• Judge whether dead code is intentional (feature flags, emergency rollback paths)
• Modify specs — it flags mismatches, the human decides
• Scan for code quality issues — that's /age or /simplify
• Detect reinvented wheels — that's /nih-audit

Gotchas

• Dynamic dispatch (trait impls, interface implementations, duck typing) can hide callers — confidence is capped at 95 for this reason
• Recently touched code (< 2 weeks) gets a confidence penalty — it may be WIP
• Shell functions sourced via . script.sh need text search for reference counting
• Specs use varied formats for symbol references — the agent casts a wide net but may miss prose-only references