Ghost Busters Skill — Development skill for Claude Code
Claude Code skill for hunting bugs that never crash: a protocol, a catalogue of 12 silent-bug classes, production probes.
How to install Ghost Busters Skill
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open RakamakadoN/ghost-busters-skill and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Ghost Busters Skill does
Claude Code skill for hunting bugs that never crash: a protocol, a catalogue of 12 silent-bug classes, production probes.
Alternatives in Development
- /hunt — Active vulnerability hunting on a target 1.8k ★
- Engage.Crash — Crash → root cause → reachability → empirical exploitability verdict (native bugs) 348 ★
- Diagnosing Bugs — PHẢI dùng khi: đang stuck với bug khó reproduce, không biết debug từ đâu, đã fix nhiều lần mà bug vẫn quay lại 90 ★
README
Ghost Busters
**A Claude Code skill for hunting bugs that never crash.**
A ghost is a bug that does not fail. It shows 0 instead of an error, "success" instead of 0 rows, green CI instead of a broken screen, silence instead of an alarm. Stack traces do not find ghosts. Instruments do.
This skill is a hunting protocol distilled from roughly sixty production incidents in a CRM/ERP (React + TypeScript + Supabase/PostgreSQL + an external accounting system) over three months of pair-debugging between a solo developer and Claude. Every class in the catalogue was caught at least twice.
What's inside
ghost-busters/
├── SKILL.md the protocol: 8 phases + raid, report format, excuses table
├── references/
│ ├── ghost-classes.md 13 classes of silent bugs, probes, traps, real precedents
│ ├── probes.sql read-only production probes (Postgres / Supabase / pg_cron)
│ ├── hunt-report.md report template + a worked example
│ ├── scan-unbounded-reads.cjs AST scan: supabase-js reads without paging (class 3 signature)
│ ├── scan-select-columns.cjs AST scan: selected/filtered columns vs generated types (class 5)
│ └── lib/chains.cjs shared chain walker, with the analysis limits spelled out
├── tests/ scanner regression tests + fixtures, and hunt scenarios with criteria
├── CUSTOMIZE.md the six things that are yours to fill in
└── LICENSE MIT
The protocol in one screen
0 CALL symptom with a number, a screen, a period · which source answers this type of question
1 ARCHIVE engineering memory before the first hypothesis · rejected hypotheses first
2 INSTRUMENTS production read-only probes by class · "no access" = diagnosis is a hypothesis
3 TRAIL the single proven path from screen to writer to schedule · "why doesn't it know?"
4 TRAP a test that fails now and describes the class, not the instance
5 EXORCISM
Related Skills
Claude Shot Catalogue
Claude Agent Skill: turn a folder of rushes into a catalogue an editor can cut from — pairs sound to picture w
Claude Bug Bounty — Plugin Guide
This repo is a Claude Code plugin for professional bug bounty hunting across HackerOne, Bugcrowd, Intigriti, a
0x002132 Skills Bugbounty Hunt Auth Bypass
Hunting skill for auth bypass vulnerabilities. Built from 4 public bug bounty reports. Use when hunting auth b
0x002132 Skills Bugbounty Hunt Business Logic
Hunting skill for business logic vulnerabilities. Built from 7 public bug bounty reports. Use when hunting bus
Debug Blitz
Parallel debug assault — fires debug-detective + silent-failure-hunter + env-debugger simultaneously. Three ro
Cr Tune
Mine the CR ledger for disproved classes per critic and draft citation-backed tuning proposals (.coderabbit.ya
Related Agents
Framework Pitfall Reviewer
Reviews code changes for bug classes that survive normal review — framework-silent-override (e.g. retryUntil o
Bugs Edgecases Hunter
Verification Quad Agent 3. Adversarial red-team auditor hunting silent bugs, None/undefined crashes, unhandled
Harness Code Reviewer
Code reviewer — two-stage review against a pinned SHA: spec compliance first, then code quality, hunting fail-