Rule Reference

All 159 review rules organized by layer. Generated from the KERN registry.

Base Rules

14 rules that apply to every TypeScript and Python file. These catch universal code-quality problems regardless of framework or target.

• floating-promise[error] — Unresolved async operation — missing await/void/return
• state-mutation[error] — Illegal state mutation outside designated setter
• empty-catch[warning] — Catch block swallows exception silently
• machine-gap[warning] — Unreachable state or missing transition in state machine
• config-default-mismatch[warning] — Config schema default does not match type
• event-map-mismatch[warning] — Event handler type mismatch with event map
• non-exhaustive-switch[warning] — Switch/map missing cases for known variants
• cognitive-complexity[warning] — Function exceeds cognitive complexity threshold
• template-available[info] — Pattern matches a registered KERN template
• handler-extraction[info] — Handler-like pattern could be extracted to KERN
• memory-leak[error] — Event listener added without cleanup
• unhandled-async[warning] — Async function without error handling
• sync-in-async[warning] — Synchronous blocking call inside async function
• bare-rethrow[warning] — Catch rethrows error without adding context

Security Rules — v1

8 rules covering injection, XSS, secrets, and common web security misconfigurations.

• xss-unsafe-html[error] — innerHTML/dangerouslySetInnerHTML with untrusted data
• hardcoded-secret[error] — API key, password, or secret in source code
• command-injection[error] — exec/spawn with user-controlled input
• no-eval[error] — eval() or Function() constructor usage
• insecure-random[warning] — Math.random() used for security-sensitive operations
• cors-wildcard[warning] — CORS wildcard (*) origin allowed
• helmet-missing[warning] — Express app without helmet security headers
• open-redirect[error] — Unvalidated redirect target from user input

Security Rules — v2

6 rules for authentication hardening, CSRF, and cryptographic misuse.

• jwt-weak-verification[warning] — JWT verified without algorithm restriction
• cookie-hardening[error] — Cookie missing secure/httpOnly/sameSite flags
• csrf-detection[error] — State-changing endpoint without CSRF protection
• csp-strength[warning] — Weak Content-Security-Policy headers
• path-traversal[error] — File path from user input without sanitization
• weak-password-hashing[error] — MD5/SHA1 for password hashing instead of bcrypt/argon2

Security Rules — v3

5 rules targeting modern attack vectors including prototype pollution, ReDoS, and LLM prompt injection.

• regex-dos[warning] — Regex vulnerable to catastrophic backtracking (ReDoS)
• missing-input-validation[warning] — User input used without validation
• prototype-pollution[error] — Object.prototype mutation via user-controlled keys
• information-exposure[error] — Stack traces or internal details in error responses
• prompt-injection[warning] — User input concatenated into LLM prompts

Security Rules — v4 (LLM Attack Surface)

10 rules covering the full OWASP LLM01 prompt injection surface — indirect injection, output execution, RAG poisoning, encoding bypasses, and more. No other SAST tool covers this.

• indirect-prompt-injection[warning] — LLM prompt includes data from external/DB sources
• llm-output-execution[error] — LLM-generated code executed without sandboxing
• system-prompt-leakage[warning] — System prompt exposed in error paths or responses
• rag-poisoning[warning] — RAG documents injected without provenance check
• tool-calling-manipulation[error] — Tool/function call parameters from untrusted LLM output
• encoding-bypass[warning] — Base64/unicode encoding used to bypass prompt filters
• delimiter-injection[warning] — Prompt delimiter breakout via user input
• unsanitized-history[warning] — Chat history concatenated without sanitization
• json-output-manipulation[warning] — LLM JSON output used without schema validation
• missing-output-validation[warning] — LLM output consumed without validation

Dead Logic

8 rules that detect code which compiles but can never affect runtime behavior. Often symptoms of refactoring mistakes.

• identical-conditions[error] — Duplicate conditions in if/else chain
• identical-expressions[error] — Same expression on both sides of operator
• all-identical-branches[error] — All branches produce identical code
• constant-condition[warning] — Condition is always true or always false
• one-iteration-loop[warning] — Loop body always exits on first iteration
• unused-collection[warning] — Collection created but never read
• empty-collection-access[warning] — Accessing elements of provably empty collection
• redundant-jump[info] — Unreachable code after return/break/continue

Null Safety

3 rules that catch null/undefined errors before they become runtime crashes.

• unchecked-find[warning] — array.find() result used without null check
• optional-chain-bang[warning] — Optional chain (?) immediately negated by non-null assertion (!)
• unchecked-cast[warning] — Unsafe type assertion without runtime guard

React Rules

7 rules active when target is nextjs, tailwind, web, native, or ink. Catch hook misuse and rendering pitfalls.

• async-effect[error] — Async function passed directly to useEffect
• render-side-effect[error] — Side effect (fetch, mutation) during render
• unstable-key[warning] — Non-stable key prop (index, random, Date.now)
• stale-closure[warning] — Stale variable captured in hook closure
• state-explosion[warning] — Excessive useState calls — consider useReducer
• hook-order[error] — React hook called inside condition or loop
• effect-self-update-loop[error] — useEffect updates its own dependency — infinite loop

Next.js Rules

3 rules specific to the Next.js App Router model.

• server-hook[error] — React hook used in Server Component
• hydration-mismatch[warning] — Nondeterministic expression causes SSR/client mismatch
• missing-use-client[warning] — Event handler in Server Component — needs use client

Vue Rules

4 rules active when target is vue or nuxt.

• missing-ref-value[warning] — ref() used without .value in script setup
• missing-onUnmounted[error] — watch/addEventListener without onUnmounted cleanup
• setup-side-effect[warning] — Top-level await in setup without onMounted
• reactive-destructure[warning] — Destructuring reactive() loses reactivity

Nuxt Rules

3 rules specific to Nuxt 3.

• missing-ssr-guard[error] — Browser global accessed without SSR guard
• nuxt-direct-fetch[warning] — Raw fetch() instead of $fetch/useFetch in Nuxt component
• server-route-leak[error] — Server API route may expose sensitive fields

Express Rules

5 rules for Express.js server applications.

• unvalidated-input[error] — req.body/params/query used without validation
• missing-error-middleware[warning] — Express app without error-handling middleware
• sync-in-handler[warning] — Blocking I/O in request handler
• double-response[error] — Response sent twice without early return
• express-missing-next[error] — Middleware accepts next but never calls it — request hangs

FastAPI Rules

5 rules for Python FastAPI applications.

• fastapi-missing-response-model[warning] — Endpoint without response_model — undocumented response
• fastapi-blocking-sync-route[warning] — Blocking call in async route stalls event loop
• fastapi-shared-state[error] — Route mutates global/module state — race condition
• fastapi-broad-except[warning] — Broad except without re-raising HTTPException
• fastapi-broad-cors[warning] — CORSMiddleware with allow_origins=["*"] — overly permissive

CLI Rules

4 rules for Commander.js CLI applications.

• cli-missing-shebang[warning] — Commander CLI entrypoint missing #!/usr/bin/env node
• cli-missing-parse[error] — Command instance created without parse()/parseAsync()
• cli-async-parse-sync[error] — Async Commander action paired with parse() instead of parseAsync()
• cli-process-exit-in-action[warning] — Commander action handler calls process.exit() directly

Terminal Rules

7 rules for terminal UI applications.

• terminal-missing-tty-guard[warning] — Interactive terminal code runs without TTY guard
• terminal-raw-mode-no-restore[error] — stdin raw mode enabled without restore on exit
• terminal-readline-no-close[warning] — Readline interface never closed — process can hang
• terminal-alt-screen-no-restore[warning] — Alternate screen entered without restore on exit
• terminal-missing-signal-handler[warning] — No SIGINT/SIGTERM handler for cleanup
• terminal-cursor-not-restored[warning] — Cursor hidden without restore on exit
• terminal-unthrottled-render[warning] — Render loop with excessive refresh rate

Ink Rules

6 rules for Ink (React terminal UI) applications. Active on top of React rules.

• ink-console-output[warning] — console.* output corrupts Ink terminal rendering
• ink-direct-stdout[error] — Direct stdout/stderr writes bypass Ink renderer
• ink-process-exit[warning] — process.exit() used instead of useApp().exit()
• ink-stdin-bypass[warning] — Raw stdin/readline listeners bypass Ink useInput()
• ink-uncleared-interval[warning] — setInterval without cleanup in Ink component
• ink-missing-error-boundary[warning] — Ink render() without error handling

Concept Rules

5 language-agnostic rules that operate on semantic concepts extracted from the KERN IR. Always active.

• boundary-mutation[warning] — Global/shared state mutation across boundaries
• ignored-error[warning] — Caught exception silently ignored
• illegal-dependency[warning] — Module imports from a disallowed layer or boundary
• unguarded-effect[warning] — Network/DB effect without auth/validation guard
• unrecovered-effect[warning] — Network/DB effect without error recovery

Severity levels

Every finding is assigned one of three severity levels. CI enforcement via --enforce fails on any error-level finding by default.

Confidence scores

Each finding includes a confidence score from 0 to 1 indicating how certain the analyzer is that the finding is a true positive.

• 0.9 – 1.0 — High confidence. Pattern matched exactly with full taint-tracking or type evidence.
• 0.7 – 0.89 — Medium confidence. Pattern matched but some context is ambiguous (e.g., dynamic import, indirect call).
• 0.5 – 0.69 — Low confidence. Heuristic match that may be a false positive. Review manually.
• Below 0.5 — Suppressed by default. Enable with --min-confidence=0 to see all findings.

Filter findings by confidence:

# Only show findings with >= 80% confidence
kern review src/ --recursive --min-confidence=0.8