Skip to content

Security: donbagger/dexpaprika-agent-kit

SECURITY.md

Security

Please report a vulnerability privately through GitHub's private reporting rather than in a public issue. I aim to answer within a few days.

Keys

  • DEXPAPRIKA_API_KEY is optional. When set, it is sent in the Authorization header, the key on its own, and only to https://mcp.dexpaprika.com. A different DEXPAPRIKA_MCP_URL with a key set stops the kit before it connects.
  • LLM_API_KEY goes only to LLM_BASE_URL, through the official openai client.
  • TELEGRAM_BOT_TOKEN is part of Telegram's API URL by design. pool-radar never prints that URL or the exception text that could contain it.
  • Keys live in .env, which is in .gitignore. Nothing is written to disk except the list of pool ids pool-radar has already seen.

Untrusted data

Token names and symbols in tool results are set by whoever deployed the token.

  • The model in market_chat.py can only call the tools listed in TOOLS. A tool result that tells it to do something else cannot reach any other tool, and every DexPaprika tool on that list only reads public market data.
  • Text from tool results is stripped of control characters and bidirectional overrides before it reaches a model, a terminal or a chat.
  • In alerts, token symbols and DEX names also lose markup characters and URL schemes, so a token named like a link cannot post as one. Discord alerts are sent with mentions disabled, Telegram alerts as plain text.

Supply chain

Dependencies are installed from a requirements file pinned with hashes. CI actions are pinned to commit SHAs, Dependabot keeps both current, and the workflow token is read-only.

The signals are paper only. Nothing in this repo places orders or holds funds.

There aren't any published security advisories