heyM Multiple Secrets Stored or Replayed in Plaintext Vulnerability Advisory

Vulnerability Report Summary: heyMrn/heyM Vulnerability Overview Title: Multiple secrets stored or replayed in plaintext: webhook header-auth secrets, MCP API keys, portal sessions, execution JWTs, Discord interaction tokens, global variables Severity: Moderate (CVSS 6.9 / 10) Core Issue: Flaws in key handling within the cluster result in capability secrets being stored, returned, or persisted in plaintext, allowing database readers or low-privilege workspace members to recover and replay these secrets. Scope Affected Versions: <= 0.0.90 Fixed Version: 0.0.91 Affected Components: Webhook Authentication Keys MCP API Keys Portal Session Tokens Execution JWTs Discord Interaction Tokens Global Variables Detailed Vulnerability Description 1. Webhook Authentication Key Leakage (High) Description: returns to any user with workspace access (including shared and team members). Risk: Attackers can obtain the and replay it in the , bypassing the workspace's own check. Remediation: Write the value only once (as a secret), and set to in . 2. MCP API Key Stored and Replayed in Plaintext (High) Description: is a plaintext column. Authentication comparison uses the raw value and accepts keys passed via query string ( ). Risk: DB readers or backups can obtain keys used to execute workflows; query string keys leak into access logs, proxies, and browser history. Remediation: Store SHA-256 hashes, compare hashes, return the key only once upon creation, and remove query string authentication. 3. Portal Session Token Stored in Plaintext (Medium) Description: stores the raw value of and verification uses equality comparison. Risk: DB readers can replay portal sessions without needing the portal password. Remediation: Store SHA-256 hashes and compare hashes. 4. Workflow Execution JWT Storage and Replay (Medium) Description: stores signed JWTs and returns every token string again in . Risk: Anyone with DB access can replay execution access until expiration, and the schema allows for long TTLs. Remediation: Store only (JWT ID) and never return the token string. 5. Discord Interaction Token Persistence (Medium) Description: The complete interaction body (including and ) is stored in . Risk: Anyone with DB access can follow up on messages or exploit Discord's 15-minute TTL window. Remediation: Strip (and ) from the persisted copy, retaining them only in the temporary execution context. 6. Global Variables Readable/Editable in Plaintext (Medium) Description: is plain JSON, and returns the unwrapped value to owners and sharers. Risk: Anyone with shared or edit permissions can view values (e.g., holds passwords). Remediation: Use the same Fernet AEAD encryption values as credentials, mask read values (visible only to the owner, or never written), and add a read-only sharing level. Remediation Summary Unified Pattern: Use SHA-256 hashing for capability tokens (such as refresh tokens), mark shared secret fields as write-only in the response schema, and pass request-derived secrets through an explicit denylist (or preferably an allowlist) before persisting to execution history. Status**: The HeyM team has confirmed and patched most of the above issues; the patch is scheduled for version 0.0.91.

Goal Reached Thanks to every supporter — we hit 100%!

heyM Multiple Secrets Stored or Replayed in Plaintext Vulnerability Advisory

More from this source