Awesome Bug Bounty

Distilled knowledge base for bug bounty hunting and authorized security research. Paths below are relative to this skill's directory. Prefer them; only fetch source repos (Fallback table) when deeper detail is needed.

Operating rules

1. Read extended detail from knowledge/*.md before improvising:
   • knowledge/vuln-types.md — per-vuln hunt focus + example patterns
   • knowledge/payloads.md — payload/bypass cheat sheet by context
   • knowledge/business-logic.md — business logic + race condition playbooks
   • knowledge/methodology.md — recon/API methodology, best practices, non-duplicated engagement path, report template, wordlists
   • knowledge/tools.md — tool-choice matrix: proxies, AI-native hunters, Obscura, MCP stack, authz/API/bizlogic tools
   • knowledge/install.md — install commands + post-install setup (API keys, proxy CA, MCP registration)
2. For writeup links, full payload lists, or tool internals: fall back to source repos via registered opencode references (@awesome-bb-writeups, @bug-bounty-reference, @payloads-all-the-things, @hack-skills, @bizlogic, @aw-junaid-bug-bounty, @hackerone-reports, @autorizepro, @burp-api-security-suite, @awesome-bugbounty-tools, @obscura, @caido-skills) or the Fallback table URLs. If a reference isn't registered, use the GitHub URL.
3. Authorized testing only. Stay inside program scope and rules of engagement.
4. Evidence standard for reports: clear impact, minimal repro steps, PoC request-response, severity justification, fix guidance.

Operating profile gate (ask before acting)

When: at the first action-taking turn of a session/engagement — skip entirely if the user already stated preferences/rules in this conversation, or if engagements/<target>/profile.yaml (or an equivalent profile the user pointed to) already exists. Never re-ask within a session after answers are given.

How: ask once, using the question tool, a single question:

"Operating profile for this engagement?"

• Defaults — Stealth (Recommended) — in-place/Ollama LLM, no cloud keys, recon + verify layers, install core, MCP: pd-tools & obscura. Applied immediately, no further prompts.
• Configure… — answer the full set below (one question call, all items together).

Full set (only when Configure…):

1. Mode — Stealth / no API keys (Recommended) (local/offline tools only; HEXSTRIKE_API_KEY and proxy CA are local auth, allowed) · Balanced (in-place LLM + read-only keys like GITHUB_TOKEN, no paid LLM APIs) · Full (cloud API keys per tool)
2. LLM backend — In-place (opencode session model) · Ollama (http://localhost:11434/v1) · Cloud API keys · None — deterministic only
3. Tool layers (multiple) — Proxy (Burp/Caido/ZAP) · Recon stack · Verify (sqlmap/dalfox/interactsh) · Obscura browser · bizlogic · AI hunter · MCP/LLM-offensive suite
4. Install policy — Install core now · Install only what this engagement needs · Don't install — report gaps only
5. MCP surface (multiple) — pd-tools · hexstrike · obscura · ptai · mcp-bb · none

After answers: write them to engagements/<target>/profile.yaml (keys: mode, llm, layers[], install, mcp[]; env names only, never secrets), then obey for the rest of the engagement:

Profile choice	Enforced behavior
Mode Stealth	Never suggest/export cloud API keys; Caido AI plugins, Burp AI, AutorizePro AI off unless backend = Ollama; prefer offline/zero-dep tools (nuclei -duc, gau/wayback passive); hexstrike/Obscura only as local stdio MCP; active scanning rate-capped
Mode Balanced	In-place/Ollama only for LLM; read-only GitHub PAT allowed; no paid API suggest
Mode Full	Per-tool keys from knowledge/install.md §11 as needed; AI hunters may use cloud LLMs
LLM In-place	All AI features pointed at the session agent — zero key setup
LLM Ollama	Point tools at http://localhost:11434/v1; verify reachable before starting
LLM None	Deterministic only: ptai --no-llm, no AI triage steps in playbooks
Layer unchecked	Do not install, run, or mention that layer as a next step this engagement
Install Don't install	Health-check and report gaps (knowledge/install.md §10); never run install commands
MCP none / unchecked	Leave servers enabled: false in opencode.jsonc; drive tools via shell instead

Declining the gate or saying "use defaults" = Defaults / Stealth above, recorded without further prompting.

Engage flow (impact-first)

0. Profile gate — once per engagement as above; persist profile.yaml; all later steps obey it.
1. Scope — write the scope manifest (knowledge/methodology.md → non-duplicated path): in-scope hosts, exclusions, rate limits, test windows. Every tool consumes this one file.
2. Recon / attack-surface map — single deduped workspace: assets → live hosts → site/API/auth surfaces (union + uro/sort -u, no per-tool re-enumeration); render SPAs with Obscura/Playwright; note in-scope MCP/LLM features.
3. Route by surface — highest-impact path first (auth bypass > ATO > RCE > SSRF/IDOR > XSS > info leaks).
4. Deep playbooks — read matching knowledge/*.md section; escalate to fallback repos only if uncovered. Fan-out tools read the shared surfaces, never re-scan them.
5. Report — dedupe findings by METHOD+host+path+param+class; impact-first writeup per template; one submission per issue.

Category router (symptom → knowledge → fallback)

Symptom / surface	Knowledge	Primary fallback
XSS, SQLi, SSTI, RCE, LFI/upload, SSRF, CSRF/CORS, smuggling, takeover, cache, host header, 401/403, SAML, most vuln classes	knowledge/vuln-types.md (+ knowledge/payloads.md for injection/WAF contexts)	PayloadsAllTheThings, Awesome-Bugbounty-Writeups, bug-bounty-reference, hackerone-reports docs/tops_*
IDOR/BOLA, API recon, GraphQL, mass assignment	knowledge/vuln-types.md, knowledge/methodology.md, knowledge/tools.md	AutorizePro, BurpAPISecuritySuite, hack-skills api-sec
Auth bypass, 2FA/MFA, OAuth/JWT, ATO	knowledge/vuln-types.md	bug-bounty-reference, hack-skills auth-sec
Business logic, race conditions	knowledge/business-logic.md	hack-skills, bizlogic, PayloadsAllTheThings, hackerone-reports TOPRACECONDITION
Recon, wordlists, engagement workspace, report template, SPA browsing	knowledge/methodology.md + knowledge/tools.md (Obscura)	aw-junaid/bug-bounty, obscura
Proxy / scanner / AI hunter / MCP tool choice, agent orchestration	knowledge/tools.md	awesome-bugbounty-tools, caido-skills, hexstrike-ai, pd-tools-mcp
Install commands, API keys, proxy CA, MCP registration	knowledge/install.md	upstream repo README
Mode / LLM / layers / MCP enablement	Operating profile gate → profile.yaml	knowledge/install.md §11–12
MCP server or LLM app testing (prompt injection, tool poisoning)	knowledge/tools.md, knowledge/payloads.md	MCPScan, mcpsec, mcpwn, AI Scanner

Covered classes are indexed in knowledge/vuln-types.md (XSS → MCP abuse).

Fallback repositories

Reference alias	Repository	Role
@awesome-bb-writeups	devanshbatham/Awesome-Bugbounty-Writeups	Writeups indexed by bug type
@bug-bounty-reference	ngalongc/bug-bounty-reference	Writeups by bug nature (XSSI, OAuth, money, business logic)
@payloads-all-the-things	swisskyrepo/PayloadsAllTheThings	64+ vuln categories: payloads, bypasses, methodology
@hack-skills	yaklang/hack-skills	100+ agent skills; master/category routers
@bizlogic	ekomsSavior/bizlogic	Business-logic heuristic scanner (9 checks)
@aw-junaid-bug-bounty	aw-junaid/bug-bounty	Methodologies, cheatsheets, wordlists, report templates
@hackerone-reports	reddelexc/hackerone-reports	Top disclosed HackerOne reports by bug type + program
@autorizepro	WuliRuler/AutorizePro	Burp authz-enforcement tester + AI FP reduction
@burp-api-security-suite	Teycir/BurpAPISecuritySuite	Burp API suite: recon, 15 attack types, BOLA/IDOR
@awesome-bugbounty-tools	vavkamil/awesome-bugbounty-tools	Curated tool index by phase (incl. AI Agents)
@obscura	h4ckf0r0day/obscura	Rust headless browser for AI agents: CDP + MCP, stealth, SPA rendering
@caido-skills	caido/skills	Caido Client SDK + AI skill (AI-native Burp alternative)

Deep research: fetch the corresponding reference path (e.g. @payloads-all-the-things → SQL Injection/README.md) rather than guessing payloads.