About this project
Warden Desk is an open-source Python project that acts as a decision layer, not a trading bot, for evaluating on-chain risk on Solana launchpad tokens. It is built around eight agents that share a common interface and read all thresholds from a single rules.json file. Seven agents correspond to explicit rules: GRAVEYARD looks for coins that died on an empty bonding curve, MIDWIFE detects manufactured ticker waves, COLLECTOR avoids clones and already-moved originals, CLOCK enforces a minimum holding lock, TOLL refuses entries when fees overwhelm profit, LADDER bans averaging down and only allows adds to winning positions, and UNDERTAKER expires trades after a thesis timer. The eighth agent, WARDEN, runs all seven checks and then asks a final liquidity question: if I buy this, who buys it from me?
The project ships with an offline SQLite snapshot of a public Solana launch feed and a set of collected candidates, so a fresh clone can run the full pipeline without network access, API keys, wallets, or transaction signing. Commands can display rules, inspect the bundled index, run the full desk, or invoke individual agents with command-line parameters. Verdicts are written to refusals.jsonl and can be exported as CSV, with contract addresses attached so later observers can compare refusals against actual token performance. Exit codes allow the desk to be composed as a veto layer in front of separate execution systems.
The README is explicit about limitations: it does not execute trades, handle keys, or make performance claims; holder data requires a user-supplied Solana RPC endpoint and unknown data fails closed as a refusal; the bundled index is historical rather than live; and the high refusal rate is presented as a measured property of the filters rather than a tuned marketing figure. Development includes offline tests, linting, schema validation, and CI checks across multiple Python versions.
Comments
0 Rating appears after 10 ratings
Sign in to join the discussion.