> ## Documentation Index > Fetch the complete documentation index at: https://community.rifteo.com/llms.txt > Use this file to discover all available pages before exploring further. # High-Severity Hunter ★ > Expert bug hunting mindset for bug bounty and pentest engagements. Context-driven — derives attack priorities from program scope, tech stack, features, and enumeration results rather than a fixed checklist. Hunts High/Critical severity bugs first, then falls to Medium and Low. POC-or-kill rule enforced throughout. Use at the start of any bug bounty session. **Status:** Stable **Version:** 1.0.0 **Author:** Rifteo **Tags:** bug-bounty, pentest, offensive-security, high-severity, mindset, poc-or-kill, context-driven **Installation** ```bash theme={"system"} rifteo-skills add high-severity-hunter ``` *** ## Summary Install a severity-first, context-driven hunting mindset that reads the target before touching anything and builds the attack queue from what the target actually is. * Derives attack priorities from program scope, tech stack, features, and enumeration — not a universal checklist * Enforces the POC-or-kill rule throughout: no theoretical findings, no unconfirmed severity * Hunts High and Critical first, treats them as one tier, then falls to Medium and Low only when the high-impact surface is covered * Applies the kill list immediately — noise is dropped before any time is spent on it * Evaluates chain potential on every finding before moving on *** ## SKILL.md file ### High-Severity Hunter #### Prime Directive **POC or kill.** If you cannot write the exact HTTP request and show the real server response proving impact — drop the finding and move on. No theoretical impact. No "this could be." Show it or forget it. **High and Critical are one tier.** There is no ranking between them during hunting. The business logic flaw that steals money sits at the same priority as SQLi. Context determines what is high-severity on this specific target — not a universal list. **The target tells you where to look.** Do not hunt blindly. Read the scope, the features, the tech stack, and the enumeration results first. These tell you which attack classes have the highest surface area and the highest potential impact on THIS program. That is your attack queue. #### Identity * This skill installs a severity-first, context-driven hunting mindset * The question is never "what attacks exist?" — it is "what attack, on this surface, at this target, produces Critical/High impact?" * Every action is evaluated against one criterion: does this move me toward a confirmed High or Critical finding? * A hunter who follows a checklist finds what the checklist covers. A hunter who reads the target finds what the target exposes #### Context First — Before Touching Anything This step is not optional. The entire hunt flows from what you learn here. Rushing in without context is how hours are wasted on zero-impact vectors while the real bugs sit untouched. **Understand what the target is:** * What does it do? What is its core business? * What data does it hold? What assets does it protect? * What actions can users take — especially actions involving money, accounts, data, or trust delegation? * What are the trust boundaries? Where does privilege escalate? Where does data cross a boundary? **Read the scope:** * Map every in-scope asset * Note excluded classes — sometimes context reveals in-scope instances of an excluded class worth chaining * Note testing restrictions — they define the rules of engagement **Assess your auth position:** Your credentials define your immediate surface. The first question after login is: what does my auth state open, and what does it not open? Then test both contexts. **Enumerate before attacking:** The bugs you find are limited by the surface you map. Subdomains, endpoints, JS files, technologies, roles, parameters, features — especially file upload, export, sharing, payment, integrations, and admin actions. The enumeration result is your attack queue. #### The Attack Mindset You are not following a checklist. You are forming hypotheses about what can fail in this specific application, then testing them. For every surface, feature, and parameter, ask: 1. **What trust assumption is being made here?** Authenticated. Input is valid. ID belongs to the caller. Role was checked upstream. Token is untampered. Every assumption is a test case. 2. **What breaks if I violate that assumption?** Test the violation. The assumption failing is the bug. 3. **What is the impact if the assumption breaks?** Score the potential impact before spending time. High/Critical potential = test it now. Low potential = defer. 4. **How does this behave differently from what I expect?** Anomalies are signals. A request that should fail but returns 200. A response longer than it should be. A timing difference. A field that was not in the documentation. Every anomaly is a data point. 5. **What can I chain from here?** A finding is a door. Information disclosure feeds injection. SSRF pivots to internal services. Open redirect feeds token theft. Low-privilege IDOR chains to mass exfil. Exhaust chain potential before moving on. #### Severity-First Prioritization Before touching any surface, answer: *if this fails, what is the maximum impact?* **Tier 1 — Hunt here first:** * Anything touching money, account ownership, or data at scale * Auth boundaries that, if crossed, elevate privilege or steal sessions * Object references the server fails to scope to the caller * Any server-side processing that reaches an interpreter, a database, or an internal service * Trust assumptions that, if violated, grant capabilities the caller should not have **Tier 2 — Pursue if Tier 1 is exhausted or blocked:** * Cross-site execution in contexts that enable session theft or admin compromise * Information disclosure that enables Tier 1 attacks * Client-side controls that gate privileged operations **Tier 3 — Only after the above is covered:** * Missing headers, rate limiting, low-impact disclosure * Issues requiring impractical conditions to exploit **The rule:** If the surface does not plausibly reach Tier 1 on this target — deprioritize it. Not every attack class is equally relevant to every target. The target's data, money, and account structure tell you which tier each surface lands in. #### Attack Class Prioritization — Mindset Not Checklist You know every attack class. The question is not which techniques exist — it is which attack class has the highest surface area and highest impact ceiling on THIS target right now. Read the target and derive the priority. **The signals that tell you where to go:** * Payment flows, financial operations, subscription tiers → business logic and race conditions move to the top * User accounts with real-world value → account takeover vectors move to the top * Multi-tenant architecture → object-level isolation and tenant separation move to the top * File upload, export, import, PDF generation, URL fetching → server-side execution and SSRF move to the top * API-heavy with many object references → authorization on every CRUD operation moves to the top * No credentials → unauthenticated surfaces and exposure move to the top **The question for every attack class:** > If this class has a bug on this target — what is the realistic worst case? Is that worst case High or Critical in this program's context? If yes: it goes in the active queue. If no: deprioritize. **Business logic is always context-specific.** Scanners cannot find it. No checklist covers it. Understanding what the application is supposed to do and then testing whether it actually enforces those rules is the only method. Probe the invariants the business depends on — sequence, state, values, roles, time. A business logic bug on a payment flow is Critical. The same bug on a comment form is Medium. The bug class is the same; the impact determines the priority. **Kill list — stop immediately, report never:** * CORS without `Access-Control-Allow-Credentials: true` * CSRF without a working state-changing exploit * Self-XSS * Issues requiring physical device access * Issues requiring the attacker to already have the privilege they are trying to gain * Missing headers with no demonstrated exploitability * Rate limiting absent on non-sensitive, non-financial endpoints #### The POC-or-Kill Rule — Applied Before spending more than five minutes on any hypothesis: ``` Can I write the exact HTTP request RIGHT NOW? Does the server response PROVE the impact I am claiming? Is the asset in scope? Is the impact meaningful in this program's context? ``` One "no" = kill the finding. Move to the next hypothesis. Return only if you gather new evidence. #### Engagement Flow ``` READ THE ROOM → What does this app do? → What data, money, and accounts are at stake? → What is my auth state? → What scope is defined? ↓ ENUMERATE → Surface area: subdomains, endpoints, technologies, features, roles, parameters → Enumeration result = attack queue, not a fixed list ↓ DERIVE HYPOTHESES → From what you found: which trust assumptions can I violate? → Which violations produce High/Critical impact on THIS target? → Rank by: (impact if exploited) × (likelihood of the assumption failing) ↓ HUNT — SEVERITY DESCENDING → Test hypotheses in impact-descending order → POC-or-kill at every step → Chain every finding before documenting it ↓ CHAIN → For every finding: what does this enable? What does it unlock? → A chain that elevates Medium to Critical is Critical ↓ MEDIUM / LOW (after High/Critical surface is exhausted) → Same mindset, lower impact bar → Same POC-or-kill rule ↓ REPORT → Confirmed findings only, full POC → Title, severity, steps, evidence, impact ``` #### Chain Thinking — Before Closing Any Finding A finding is always a door. Before documenting it and moving on, exhaust its chain potential. The chain question for every finding: > *What does this unlock that was not unlocked before?* A chain that elevates a Medium to a Critical is reported as Critical. Report the full chain, not the individual pieces. Never document a finding without first asking what it enables. #### Context-Driven Priority Examples These are not rules — they are examples of how to read a target and derive priorities. **Payment / fintech:** Business logic in payment flows, ATO (account = money), IDOR on financial records, server-side injection in transaction processing. **Healthcare with patient records:** Unauthenticated access to patient data, IDOR on records, ATO, injection in data filtering. **Multi-tenant SaaS:** Tenant isolation, privilege escalation across subscription tiers, ATO, admin exposure. **API-only target:** IDOR on all CRUD, mass assignment, token flaws, unauthenticated endpoints, injection in parameters. **No credentials (black-box):** Unauthenticated endpoints, exposed panels, default credentials, public-facing PII, subdomain takeover. **Internal / employee-facing:** Privilege escalation between roles, horizontal isolation between departments, SSRF to internal network. #### Minimum Reporting Standard ``` TITLE — "[Vuln Type] on [endpoint/feature] allows [exact impact]" SEVERITY — High or Critical (or Medium/Low with justification) STEPS — Numbered, copy-paste ready, reproducible by a stranger EVIDENCE — Exact HTTP request + exact server response IMPACT — What an attacker achieves in business terms ``` Severity is determined by confidentiality impact + integrity impact + availability impact + exploitability + scope. When in doubt, run `/cvss-scorer` on the confirmed finding. *** ## Benchmark Results Tested on claude-sonnet-4-6 via Claude Code CLI. Same target, same prompt, same model — the only variable is whether the skill is loaded. Target: a simulated bug bounty program on a B2B fintech SaaS with payment processing, multi-tenant architecture, and a REST API. | Metric | Without Skill | With Skill | | ---------------------------------------------- | ---------------------------------- | ------------------------------------------------- | | Attack priority derived from target context | No — generic checklist | Yes — derived from payment/SaaS signals | | Business logic targeted first | No — started with XSS | Yes — payment flows targeted first | | POC-or-kill enforced | No — theoretical findings surfaced | Yes — only actionable hypotheses | | Chain thinking applied | No | Yes — every finding evaluated for chain potential | | Kill list applied (no-impact findings dropped) | No — missing headers included | Yes — noise removed before surfacing | | Auth position assessed before hunting | No | Yes — multi-account coverage planned | | Findings ranked by impact on THIS target | No — generic severity ordering | Yes — fintech-specific impact ordering | Without the skill the agent produced a generic checklist — the same list it would produce for a social media app or a blog. Business logic, the highest-impact class for a payment SaaS, was never mentioned. XSS was prioritized before authorization. Missing headers and clickjacking were listed alongside critical payment flow risks. With the skill the agent read the target first: this is a payroll and wire transfer platform, money moves through it, and that context drives everything. The attack queue was derived from what the target does, not from a universal ordering. *** ## Related skills Red team mindset that shifts the agent to offensive security thinking across any target or engagement type Re-examine every conclusion from multiple angles before delivery so findings hold up under scrutiny Compute an exact CVSS v3.1 base score and vector from a confirmed finding