Skip to content

Security

Please don’t open a public issue. Report it privately on GitHub: Security → Report a vulnerability on the Shardlens repository. You should get an acknowledgement within 7 days. The full policy is in SECURITY.md.

Threat Controls
Bot token leaks through the SDK REST options and headers are never serialised; only bucketed routes are stored; secret redaction on all error text; fuzz-tested with 10,000 token-shaped strings
Interaction or webhook tokens leak Paths are read in memory only to find the interaction ID; stored routes contain :token, never the token
Forged ingest traffic 32-byte bearer keys stored only as SHA-256 hashes, scoped (ingest or read), checked before the request body is read, HTTPS only
Request floods and scanning Every route is rate limited per client address, unknown paths included; sign-in, key and export routes have tighter limits
Forged Discord webhook events Ed25519 verification, 5-minute timestamp window, 401 on failure
Hostile payloads zod validation on every route, 1 MB compressed and 5 MB decompressed limits, field length caps; NUL characters, IDs beyond 64 bits and anything else the database would refuse are answered with 400, never 500
Slow or stalled requests A client has 30 seconds to send a request (60 for its headers); a client that disconnects early costs nothing
SQL injection Parameterised queries only; queries use enums, never identifiers from input
XSS from server names or usernames Rendered as text only, and a strict Content Security Policy
CSV formula injection Exported cells starting with = + - @, tab or carriage return are neutralised
Cross-site requests SameSite=Lax cookies and a required CSRF header on every change
Session theft HttpOnly and Secure cookies, hashed session IDs, 7-day expiry, rotation on sign-in, server-side sign-out. Revoking a read key ends the sessions opened with it; an admin taken off the team loses access on their next request
Unauthorised dashboard access Full access only for the application owner, team and listed extras; OAuth state and PKCE
SSRF through alert and report webhooks Only Discord webhook URLs are accepted, redirects are refused, and the URLs are stored encrypted
Re-identification of hashes HMAC with secret salts that are destroyed after their period
Audit data exposure AES-256-GCM with a key outside the database, 90-day maximum retention, access log, exact lookups only
Database exposure PostgreSQL is not published by Docker Compose
The SDK harming the bot Bounded queue, priority dropping, no I/O in listeners, timers that don’t keep the process alive, chaos tests
Supply chain Zero-dependency core, lockfile, Renovate, npm provenance, CodeQL, dependency review, OpenSSF Scorecard
Container escape Non-root user that can’t change the collector’s files, read-only root filesystem, no Linux capabilities

Limits are per client address, per minute. Past a limit the collector answers 429 with a Retry-After header, which the SDK honours.

Route Limit
POST /v1/ingest 3,000
POST /v1/query, POST /v1/discord/events 600
GET /v1/config, GET /v1/badge/* 120
POST /v1/privacy/erase 60
/auth/discord, /auth/read-key 20
GET /v1/audit/export.csv, POST /v1/alerts/:id/test, POST /v1/reports/:id/send 10
Everything else, including unknown paths 1,200

Discord’s Developer Policy requires you to notify Discord and affected users of unauthorised access to API data. The incident checklist in SECURITY.md lists what to rotate and who to tell.