The method-first
bug bounty agent.

12 skills and a master router that turn any AI agent into a disciplined hunter — verify, then report. PoC or nothing.

┌─ huntpack ─────────────────────────────────────────────┐
│ $ huntpack --assess example.com                        │
│                                                          │
│   → understand   application-archetypes e-commerce ✓     │
│   → model        8 assets · 3 actors · 2 walls           │
│   → hunt         business-logic price-bypass → CONFIRMED │
│   → validate     re-tested clean 4/5 killed            │
│   → report       written EN · PoC attached               │
│                                                          │
│   verdict: PoC or nothing.                              │
└──────────────────────────────────────────────────────────┘

One method, six stages.

Not a payload pack — a decision pipeline that holds an agent honest from first click to final report.

  1. 01Understandarchetypes · application model
  2. 02Reconsurface mapping
  3. 03Hunttargeted exploitation
  4. 04Disciplinescoped · honest
  5. 05Validateproof, no theory
  6. 06Reportchain + write-up

The arsenal.

Thirteen scoped skills — one dispatcher, twelve specialists. Each is a decision matrix you can load into any agent.

// router

  • routerreads any request, routes to the right specialist, enforces the method

// understand

  • application-archetypesclassify any target — e-commerce, fintech, SaaS — and get its ready-made attack plan
  • application-model-buildermap assets, actors, permissions and trust boundaries before a single payload

// recon

  • web2-reconsubdomains, live hosts, URLs, JS analysis, continuous monitoring

// hunt

  • api-securityREST, OAuth and JWT patterns — every one mapped as signal to test to confirm
  • business-logic-hunterexecution matrix for the highest-paying bug class: price, race, abuse
  • web2-vuln-classes24 bug classes — root causes, bypass tables, exploit techniques, paid examples

// discipline

  • domain-scoped-huntingdeclared boundaries, pivot hints, honest exit evidence

// validate

  • false-positive-killerre-tests from a clean state, kills theoretical bugs before triage
  • triage-validationthe 8-question gate — one wrong answer kills the report

// report

  • attack-path-builderchains low and medium findings into one verified critical path
  • report-writingstructured reports with PoC, impact and remediation — EN and AR

Ask it like you mean it.

The router reads intent, not keywords. A vague question becomes a scoped, multi-skill dispatch — everything comes back evidence-gated.

┌─ huntpack · routing ──────────────────────────────────┐
│ $ "is my payment flow abusable?"                    │
│   → archetypes (fintech) → app-model → business-logic  │
│ $ "enumerate the attack surface"                   │
│   → recon → surface-map → live-host sweep            │
│ $ "is this finding real?"                           │
│   → fp-killer → triage-gate → report                     │
└──────────────────────────────────────────────────────┘

Get started in 30 seconds.

Drop it into any agent that reads Agent Skills (SKILL.md with name + description): Claude Code, OpenCode, and more.

claude code
/plugin marketplace add Aether-DZ/huntpack
/plugin install huntpack
manual
git clone https://github.com/Aether-DZ/huntpack
cp -r huntpack/skills/* ~/.claude/skills/

View on GitHub full install guide →