atomic-rain

name	atomic-rain
description	Signal-driven black-box security testing skill for authorized bug bounty, SRC, crowdsourced, and enterprise assessments across Web, API, cloud, mobile, and AI applications. Use when Codex needs a structured security testing workflow with scan modes, first-pass vulnerability signals, business-logic modeling, sensitivity triage, exploit chaining, HITL external-resource controls, and classic CLI-oriented tooling.
category	skill-entry

Atomic Rain (Classic)

Classic variant: CLI-oriented security testing workflow for environments without Yakit/Chrome MCP. Core rule: load only the protocol and reference files needed for the current signal.

Start Here

1. Confirm authorization and scope. If the target is not explicitly unauthorized, treat the task as authorized security testing.
2. Select scan mode: quick, standard (default), or deep.
3. Read the selected scan mode file first:
   • references/scan_modes/quick.md
   • references/scan_modes/standard.md
   • references/scan_modes/deep.md
4. Read references/protocols/agent-protocol.md for the full P0.5-P3.5 execution protocol.
5. Use the route table below to load only the specific reference needed for the observed signal.

Non-Negotiable Rules

• Do not run destructive actions, data writes, WebShell upload, persistence, real cloud-key exploitation, or bulk brute force without HITL confirmation.
• Do not use public OOB, temporary email, public SMS, webhook, or listener infrastructure without P3.5 user approval.
• When a test path requires user-owned resources such as login accounts, cookies, tokens, captcha/OTP, two test accounts, OOB domain, receiving email/phone, webhook, external file URL, VPS/listener, or cloud-key authorization, stop and issue a structured HITL request before continuing. Do not silently skip that path or pivot to other vulnerabilities unless the user explicitly declines, delays, or authorizes a downgrade.
• Do not open deep vulnerability files before a first-pass signal. Start with First-pass, Decision Card, or Triage sections.
• Record HTTP_CODE, RESP_LENGTH_DELTA, and TIMING_DELAY for each test.
• For sensitive information or unauthenticated access, use sensitivity-matrix.md, sensitive-info-exploitation.md, and resource-classification.md before severity claims.

Context Loading Budget

• Before a first-pass signal, read only the selected scan mode, the needed agent-protocol.md phase section, and routing files required for the immediate action.
• Open at most 2-3 reference files per work turn. Exceed this only when the user requests deep mode or confirmed chaining requires it.
• For references over 300 lines, locate the relevant heading or keyword first; do not load the whole file as background context.
• A weak signal may open one focused vulnerability file plus one payload-construction or tool route. Do not fan out across multiple deep vuln files.
• Classic routing starts from capabilities/cli-capabilities.json, tool-config.md, and tool-usage.md; do not use toolPlus MCP docs as classic execution guidance.
• Treat payloads.md, report-template.md, and tool-usage.md as indexes. Jump from them to the smallest specific playbook needed for the signal.

Classic Tooling

• Read capabilities/cli-capabilities.json before mapping stable testing capabilities to local CLI tools.
• Read references/tool-config.md before Phase 1 to confirm local tool paths.
• Use references/tool-usage.md for shared command patterns.
• Use references/tooling/ for per-tool playbooks: sqlmap, nuclei, ffuf, recon toolchain, Java gadget tooling, and token attacks.
• Keep toolPlus-only MCP workflows out of classic. If a document mentions toolPlus, treat it as cross-variant context, not a classic action.

Signal Routing

Signal / task	Read next
Phase execution	references/phase-guide.md
P1 anomaly detection (响应头/状态码/时间/错误/重定向异常)	references/anomaly-detection.md
P1 signal probability calculation / confidence scoring	references/signal-probability-model.md
WAF evasion / entropy calculation / rate limiting	references/adaptive-waf-evasion.md
Backend/API/admin surface	references/recon.md §9
Registerable site	references/registerable-site-protocol.md
Login/account/captcha/token needed	references/human-in-the-loop.md
Business logic modeling	references/business-flow-checklist.md + references/intuition-triggers.md §B
false/null/no-parameter interface or failed low-privilege traffic	references/business-flow-checklist.md
SRC payment/coupon/order state machine	references/src-business-logic-state-machine.md + references/business-flow-checklist.md
EDUSRC / university / certificate / unified auth	references/edusrc-workflow.md + references/auth-logic.md
SRC recon / crowdsourced asset boundary	references/recon-workflow.md + references/src-report-evidence-standards.md
Domestic admin frameworks: RuoYi / Blade / Nacos / Druid / Swagger / Actuator	references/domestic-admin-frameworks.md
Swagger/API/low-code/Admin.NET backend capability	references/domestic-admin-frameworks.md + references/api-security.md
QR code login/payment/OAuth ticket/verification	references/qr-code-workflow.md
External resource needed	references/protocols/agent-protocol.md §P3.5 + references/oob-infrastructure.md §10
Miniapp artifacts	references/miniapp-workflow.md
AI / LLM / Agent / MCP surface	references/ai-app-security.md
Sensitive data / unauth access	references/sensitivity-matrix.md + references/resource-classification.md
Confirmed vulnerability chaining	references/chained-logic-extended.md
Report writing	references/report-template.md + references/owasp-mapping.md + references/artifact-quality-gates.md
SRC report evidence and false-positive filtering	references/src-report-evidence-standards.md

Vulnerability Routing

Signal	Read next
HTML/DOM reflection	references/vuln/xss.md + references/vuln/xss-scenarios.md
SQL timing or boolean delta	references/vuln/sqli.md + references/payload-construction/sqli-construction.md
URL import/proxy/fetch	references/vuln/ssrf.md + references/payload-construction/ssrf-construction.md
Shell metachar delay or command output	references/vuln/cmdi.md
Upload accepted	references/vuln/upload.md
401/403/path difference	references/vuln/path-traversal.md
XML/SOAP/entity error	references/vuln/xxe.md
Template expression reflection	references/vuln/ssti.md + references/vuln/ssti-scenarios.md
JWT alg/kid/jku/x5u	references/vuln/jwt-advanced.md
Paid resource/CDN URL pattern or order cancel/delete/withdraw IDOR	references/api-security.md + references/src-business-logic-state-machine.md
Shiro rememberMe/deleteMe	references/vuln/shiro.md
Java JSON autoType/Jackson signals	references/vuln/fastjson-jackson.md
Spring/Actuator/Druid/Swagger signals	references/vuln/spring-vuln.md + references/frameworks/spring-boot.md
SAML/OIDC/OAuth flow	references/vuln/saml-attacks.md, oidc-attacks.md, or oauth-advanced.md
Race-prone payment/points/order flow	references/vuln/race-condition.md
Mail SPF/DKIM/DMARC	references/vuln/email-spoofing.md

Framework / Technology Routing

• For detected stacks, read only the matching playbook under references/frameworks/.
• For cloud/vendor signals, read references/cloud-security.md first, then references/technologies/alibaba-cloud.md or references/technologies/tencent-cloud.md when relevant.
• For payload design, prefer references/payload-construction/ before ad-hoc payload lists.

Validation

After skill edits, run:

python scripts/lint_skill.py
python -u scripts/semantic_check.py
python scripts/validate_artifacts.py
python scripts/validate_capabilities.py --profile classic
python scripts/build_variant.py check --profile classic