Redis Security

Redis Security Audit

Audit Redis deployment and application usage. Redis security has historically been weak in defaults (no auth, plain TCP); current versions are better but require explicit configuration.

When this skill applies

• Reviewing Redis server configuration (redis.conf)
• Auditing client library usage and connection handling
• Reviewing key naming/namespacing for tenant isolation
• Checking Lua scripts (EVAL) for safety
• Reviewing pub/sub channel scoping

Workflow

Follow ../_shared/audit-workflow.md.

Phase 1: Stack detection

# Client libraries
grep -E '"(ioredis|redis|redis-py|@upstash/redis)":' package.json
grep -E '^redis|^ioredis' requirements.txt 2>/dev/null

# Connection strings
grep -rn 'redis://\|rediss://' src/ . --include='*.js' --include='*.ts' --include='*.py' 2>/dev/null

Phase 2: Inventory

# Key patterns
grep -rnE '"[a-zA-Z]+:[^"]*"' src/ | grep -iE 'set\(|get\(|hset|hget|lpush|rpush' | head -20

# Lua scripts
grep -rn 'EVAL\|eval(\|evalsha' src/ | head

# Pub/sub
grep -rn 'subscribe\|publish\|psubscribe' src/

# redis.conf if present
ls **/redis.conf 2>/dev/null

Phase 3: Detection — the checks

Network exposure

• RDS-NET-1 Redis bound to localhost or VPN-only interface; bind 0.0.0.0 only with firewall enforcing isolation.
• RDS-NET-2 protected-mode yes (default in modern versions). Refuses connections without auth from external interfaces.
• RDS-NET-3 Port not exposed to public internet. Verify with nmap -p 6379 from outside.

Authentication

• RDS-AUTH-1 requirepass set (long random string) OR ACL users configured. No anonymous access.
• RDS-AUTH-2 Application uses ACL user with minimum privileges (read-only for cache reads, write for cache writes; no FLUSHDB/FLUSHALL/CONFIG).
• RDS-AUTH-3 Default user (default) either disabled or password-protected.
• RDS-AUTH-4 Connection string with password in env var, not committed.

# redis.conf example
requirepass 
user default off
user app on >appPassword ~app:* +@read +@write +@list -@dangerous
user readonly on >roPassword ~app:* +@read -@write -@dangerous

TLS

• RDS-TLS-1 Production Redis uses TLS (rediss://, port 6380, or TLS port). Client validates server cert.
• RDS-TLS-2 Self-managed Redis: tls-port, tls-cert-file, tls-key-file, tls-ca-cert-file configured.
• RDS-TLS-3 Managed Redis (Upstash, AWS ElastiCache with encryption in transit, Redis Cloud) — verify TLS option enabled.

Command restrictions

Dangerous commands disabled or restricted:

• RDS-CMD-1 FLUSHALL, FLUSHDB, CONFIG, DEBUG, KEYS, SHUTDOWN restricted to admin user only. App user denied via ACL -@dangerous or rename-command in older configs.
• RDS-CMD-2 KEYS * not used in app code (blocks Redis on large datasets). Use SCAN.
• RDS-CMD-3 EVAL requires @scripting category in ACL; consider restricting if not used.

Key namespacing

• RDS-KEY-1 Multi-tenant apps namespace keys with tenant: tenant:{tid}:session:{sid}. Without namespace, one tenant can guess/scan another's keys.
• RDS-KEY-2 ACL key patterns enforce namespace: ~tenant:abc123:* for tenant-scoped users.
• RDS-KEY-3 Random keys (session IDs, cache busters) use cryptographic random, not sequential IDs.

Lua scripts (EVAL)

• RDS-LUA-1 Scripts don't accept arbitrary code from request bodies. Hardcoded scripts loaded with SCRIPT LOAD; EVALSHA used.
• RDS-LUA-2 Scripts validate input within the script (Lua-side type checks).
• RDS-LUA-3 Scripts don't run for too long (default timeout 5s); long scripts can cause cluster-wide pauses.

Pub/sub

• RDS-PS-1 Channel names namespaced by tenant. psubscribe('*') reveals all activity.
• RDS-PS-2 Subscribers don't receive other tenants' messages — enforced via channel naming convention + ACL channel pattern.
• RDS-PS-3 ACL channel patterns: &tenant:abc123:* (Redis 6.2+).

Cache poisoning

• RDS-CP-1 Cache keys derived from request include user context (user ID, tenant ID) where the cached value is per-user.
• RDS-CP-2 Cache values validated on read; corrupted/unexpected types treated as cache miss, not bubbled to app.

Connection handling

• RDS-CONN-1 Connection pool capped; no unbounded new Redis() per request.
• RDS-CONN-2 Connection retries with exponential backoff; not tight loop.
• RDS-CONN-3 Sentinel / Cluster client config has fail-over without leaking data across nodes.

Persistence files

If Redis is configured with RDB or AOF:

• RDS-PER-1 Persistence files (dump.rdb, appendonly.aof) not in web-accessible directories.
• RDS-PER-2 File permissions restrict to redis user.
• RDS-PER-3 Backups encrypted at rest, transferred over TLS.

Managed Redis specifics

Upstash:

• RDS-UPS-1 REST token (used for HTTP API) treated as secret. Separate read/write tokens if available.
• RDS-UPS-2 Database scope (single-region vs global) matches data sensitivity.

AWS ElastiCache:

• RDS-AWS-1 Encryption at rest enabled.
• RDS-AWS-2 Encryption in transit enabled.
• RDS-AWS-3 Authentication via Redis AUTH token OR IAM (newer).
• RDS-AWS-4 Subnet group in private subnets; SG restricts ingress.

Redis Cloud / Redis Enterprise:

• RDS-RC-1 Default user disabled; per-app ACL user.
• RDS-RC-2 Source IP allowlist or VPC peering.

Memory limits and eviction

• RDS-MEM-1 maxmemory set; without it, OOM can crash the host.
• RDS-MEM-2 maxmemory-policy set appropriately (allkeys-lru for cache, noeviction for source-of-truth).
• RDS-MEM-3 Eviction of session keys planned for; lost sessions handled gracefully (log out, not crash).

Slow log

• RDS-LOG-1 slowlog-log-slower-than configured; reviewed periodically. Slow log can contain sensitive data — restrict access.

Application patterns

• RDS-APP-1 Sensitive data (passwords, full PII) NOT cached. Use Redis for tokens, session IDs, derived caches — not the primary store of secrets.
• RDS-APP-2 Set TTLs on session keys; idle sessions expire.
• RDS-APP-3 Rate limit counters use atomic operations (INCR with EXPIRE) to avoid race conditions.

Dependencies

• RDS-DEP-1 Redis server version current (7.x or later). Older versions had Lua sandbox escapes.
• RDS-DEP-2 Client library current; ioredis 5+, node-redis 4+.

Phase 4: Triage

Critical: Redis exposed to internet without auth; default password / no requirepass; EVAL of user-supplied scripts; app user has FLUSHALL.

Phase 5: Report

Use ../_shared/findings-schema.md. Prefix IDs with RDS-.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: hlsitechio
• Source: hlsitechio/claude-skills-security
• License: MIT

Install and usage instructions live in the source repository linked above.