| name | bot-management-signals |
| description | Design bot-management signal pipelines: signal taxonomy, collection points, risk scoring, action thresholds (allow / challenge / throttle / block), feedback loops, privacy, and false-positive controls. Use when bot score design, device or TLS fingerprints as risk signals, bot management rules, challenge step-up policy, or anti-automation signal architecture for owned applications and authorized assessments.
|
Bot Management Signals Design
Design how signals become decisions for bot and abuse management: which
inputs to collect, how to score them, which actions fire at which thresholds,
and how humans recover from false positives. Prefer the platform’s existing
CDN/WAF/bot product and app middleware over inventing a parallel opaque scorer.
Scope And Authorization
- Design/implement on systems you own or are contracted to harden.
- Control-gap measurement (UI-only challenges, bypassable scores) is not
fraud enablement — hand CAPTCHA/rate probes to
captcha-bypass-research /
rate-limit-bypass-testing under clear scope only.
- Minimize data collection; redact fingerprints, cookies, and device IDs in docs.
- Tune thresholds in staging/synthetic traffic; no production floods against
shared users or challenge providers.
- Do not hard-block accessibility users, privacy browsers, or whole regions
without documented product risk acceptance.
When To Use
- Defining signal inventory (network, TLS, browser, behavioral, reputation)
- Building/reviewing a risk score and action ladder (allow → challenge →
soft throttle → hard block)
- Placing collection (edge CDN, app, mobile SDK) with server-side enforce
- Tuning false positives, allowlists, step-up on login/signup/checkout abuse
- Keywords: bot management, bot score, device fingerprint, JA3/JA4, TLS
fingerprint, headless signals, challenge step-up, anti-automation policy,
人机验证策略, risk-based bot rules
Do not use as primary for:
| Need | Skill instead |
|---|
Soft API quotas / 429 budgets alone | api-rate-limit-design |
| Authorized CAPTCHA/token enforcement gaps | captcha-bypass-research |
| Authorized rate-limit keying bypass | rate-limit-bypass-testing |
| Auth lockout after password/OTP fails | account-lockout-design |
| Edge WAF signature rule syntax only | waf-rule-tuning-basics |
| Implementation hygiene baseline | code-quality-standards |
Workflow
1. Inventory surfaces and abuse goals
| Surface | Typical abuse | Signal priority |
|---|
| Login / OTP / reset | Stuffing, spray, SMS flood | Strict; dual rate + challenge |
| Signup / invite | Account factory | Device + reputation + challenge |
| Search / scrape-prone read | Bulk extraction | Rate + behavior + IP rep |
| Checkout / vote / promo | Integrity fraud | Session integrity + velocity |
| Public API / partner | Quota theft, scripts | API key + plan + edge score |
Document decisions that must be server-side (never trust client botScore).
2. Signal taxonomy (collect only what you use)
| Class | Examples | Notes |
|---|
| Network | Trusted IP, ASN, geo, VPN/DC | IP from edge overwrite only |
| Transport | JA3/JA4, H2 settings, ALPN | High cardinality; version carefully |
| Client | UA consistency, automation flags | Spoofable alone; low solo weight |
| Behavioral | Velocity, path entropy, session age | Prefer server-observable events |
| Reputation | IP lists, prior abuse, good bots | Separate verified-bot allow path |
| Challenge | CAPTCHA/Turnstile, PoW, MFA step-up | Server bind; single-use TTL |
| Identity | User/org, API key tier | Post-auth vs anonymous differ |
Every signal needs source of truth, TTL, and missing-data fail-open/closed.
3. Score and action ladder
- Normalize contributions; one spoofable bit must not equal hard block alone.
- Internal risk bands (low/med/high/critical) — do not leak exact weights in errors.
- Map bands: low allow+sample log; med soft rate / passive challenge; high
interactive challenge or MFA on sensitive acts; critical block/quarantine + alert.
- Login, pay, reset may force challenge even at medium.
- Verified automation (crawlers, monitors, partners): explicit allowlist/API-key path.
4. Placement, feedback, and ops
- Edge: coarse IP/TLS/reputation. App: identity velocity, business events, challenge verify.
- One authoritative decision per class; shared support reason codes; normalize route aliases
into shared velocity buckets (
api-rate-limit-design for keying).
- Log reason codes + hashed IDs, not full fingerprints; short retention for high-cardinality
signals; track challenge rate, FP tickets, good-bot miss.
- Document vendor/store outage fail-open vs closed (auth often stricter).
- Accessibility: alternate challenge paths; never pointer-only sole UX.
- Validate: missing signal, spoofed headers, omit/replay challenge. Authorized gaps →
CAPTCHA/rate-limit skills. Code with
code-quality-standards.
Routing
| Situation | Primary | Helper |
|---|
| Signal taxonomy, score bands, action ladder, FP controls | This skill | — |
| Soft QPS / 429 / budget windows | api-rate-limit-design | this for bot-triggered tighten |
| CAPTCHA/token binding research (authorized) | captcha-bypass-research | this for when to challenge |
| Rate key split / XFF bypass testing | rate-limit-bypass-testing | this for intended policy |
| Auth hard lockout after secret fails | account-lockout-design | this for pre-auth bot step-up |
| WAF rule/signature tuning | waf-rule-tuning-basics | this for score→action design |
| Code, tests, logging, config | code-quality-standards | always on implementation |
Keep this skill primary for multi-signal bot policy design. Switch for pure
quota math, pure CAPTCHA verify gaps, or pure WAF rule syntax.
Output Checklist