h1-playbooks

Attack-pattern playbooks distilled from the top-paid HackerOne reports per vulnerability class (from the reddelexc/hackerone-reports corpus, 14.9k reports). Use when PLANNING a hunt for a specific bug

SKILL.md

GitHub

h1-playbooks — Where each bug class actually pays

Each references/<class>.md holds the class's top-15 paid reports (bounty, program, title, link) mined from the corpus. Read the relevant playbook BEFORE hunting a class, and re-read the linked top reports when a hunt stalls — the winning patterns repeat across programs.

The pattern matrix (what the corpus says pays, per class)

IDOR (idor.md, $175k paid) — the money is in bulk/object-transfer flows, not single records: import/export features, project/file transfers, share-link regeneration, admin "act as user" tools. Ordinary CRUD IDORs cluster at $1–2k; import-flow IDORs hit $20k (GitLab). Hunt: every "copy/move/share/import" feature first.

Stored XSS (xss-stored.md, $427k) — pays via privileged context: payloads landing in admin panels, support inboxes, file names, email subjects rendered by staff. Reflected is commodity; stored-in-privileged-view is the tier jump. Hunt: any field a higher-privileged user will render.

RCE/Command+Code Injection (rce-injection.md, $836k — the richest class) — the big payouts are template/SSTI, deserialization, sandbox escapes in sandboxed eval features (code playgrounds, math evaluators, markdown/latex renderers), not classic shell injection. Hunt: anything that evaluates user input as code "safely".

Privilege escalation / broken access control (priv-esc.md, $415k) — winners are role/flag tampering in profile-update mass-assignment (adding role/isAdmin to editable objects) and vertical hops via feature flags. Hunt: profile/settings update endpoints — replay with role keys appended.

Info disclosure (info-disclosure.md, $411k) — top payouts are API responses leaking other tenants' data via mutable identifiers and debug/internal endpoints exposed without auth — not stack traces. Hunt: every response field that isn't yours; every /internal, /debug, /admin route unauthenticated.

Auth/MFA (auth-mfa.md, $295k) — pays for OTP flow logic: missing attempt counters, response-state tampering (changing verified:false→true), OTP reuse across sessions, login-state race. Password-reset token leakage also lands here. Hunt: the OTP verify request body and response handling.

Business logic (business-logic.md, $106k, ceiling $12k) — almost all high payouts are money movement: double payouts, currency/amount tampering at the payment-provider handoff, coupon/promo reuse and self-referral. Hunt: the exact request sent to the payment provider, and every discount redemption.

SSRF (ssrf.md, $171k) — big ones are URL-fetching features (webhooks, import-from-URL, image-by-URL, PDF generators) escalated to cloud metadata. Blind SSRF alone pays little; metadata read pays. Hunt: every field the server fetches.

SQLi (sqli.md, $174k) — modern payouts concentrate in filter/sort parameters, JSON/GQL inputs, and legacy report endpoints rather than login forms. Hunt: ?sort=, ?filter=, array params.

File read / traversal (file-read-traversal.md, $130k) — winners read server config/source with secrets (/proc/self/environ, app source with DB creds), not /etc/passwd. Proof of impact = a secret, not a file.

ATO/OAuth (ato-oauth.md, 142 paid, title-mined) — recurring patterns: redirect_uri partial-match bypasses, state-parameter absence, token leakage via referrer/postMessage, email-change-without-verification account merging, and subdomain takeover of logout/redirect hosts.

GraphQL (graphql.md, 29 paid) — winning angles: introspection-gated-but-aliasable brute force, mutation authz gaps (the UI hides it, the API doesn't), batching rate-limit bypass for OTP/login brute force.

Usage in the chain

  1. Recon (recon-engine) → surface known → pick classes by feature present (import feature → read idor.md; webhook → ssrf.md; OTP → auth-mfa.md).
  2. Hunt with the matching bb-* skill; on stall, open this skill's playbook and diff your target's surface against the top-15 patterns.
  3. Before reporting, calibrate severity with h1-reports --program/--vuln and bb-disclosures; write with vuln-report.

Regenerate playbooks after h1-reports --refresh by re-running the generator (pattern: filter corpus by class → top-15 by bounty → markdown).