Torch Domain Auction Bot
WarnAudited by ClawScan on May 10, 2026.
Overview
This is an openly described Solana DeFi keeper, but it can autonomously perform high-impact mainnet actions through a vault and has unclear install/SDK provenance.
Only install or run this if you understand Solana DeFi keeper risk. Use a fresh controller key, a separate low-funded vault, exact pinned package versions, and explicit operating limits; monitor all transactions and revoke the vault link when done.
Findings (6)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If run with a linked vault, the bot can submit irreversible financial transactions and change who controls a domain token lease.
These are high-impact Solana actions that create tokens, spend through a vault, seize collateral, and change domain control. The artifacts do not show per-transaction human confirmation or hard spend/market limits.
"TORCH_LAUNCH_DOMAIN_TOKEN" ... "Launch a domain as a Torch Market token"; "TORCH_LIQUIDATE" ... "Liquidate an underwater domain loan (LTV > 65%) via vault -- collateral tokens change hands, domain lease rotates"
Run only with a purpose-built low-funded vault, use dry-run/manual confirmation where possible, and configure explicit allowlists, liquidation limits, and spend caps before enabling automation.
A misconfigured or overfunded vault could let the bot spend more SOL than intended or act across more markets than the user expected.
The skill depends on delegated vault/controller authority. Even if the controller key is disposable, a linked wallet can act through the vault; the provided artifacts do not clearly bound that authority by value, token, or transaction count.
"VAULT_CREATOR" ... "identifies which Torch Vault the bot operates through"; "SOLANA_PRIVATE_KEY" ... "Disposable controller keypair ... Optional"; "capabilities": ["vault-full-custody", "vault-escrow", "authority-separation"]
Never use a main wallet key. Use a new controller key, fund a separate vault with only the amount you are willing to risk, and revoke the vault-wallet link when finished.
The installed/runtime code could differ from the reviewed bundled artifacts if npm dependency resolution pulls another package version.
The install reference is not pinned to an exact version, and the included runtime code imports an external package name rather than a local bundled SDK path. For a financial signing bot, this creates ambiguity about which code will actually execute.
SKILL.md: "package: torch-domain-auction-bot@^2.0.1" and "The Torch SDK is bundled in lib/torchsdk/ -- all source included for full auditability"; lib/kit/scanner.js: "const torchsdk_1 = require(\"torchsdk\");"
Pin exact package and dependency versions, publish integrity hashes or a lockfile, and make the code import the reviewed bundled SDK or clearly declare the external dependency.
A bad configuration, bad market signal, or software bug could cause repeated transactions across many domain-token markets before the operator notices.
The design applies liquidation logic across scanned tokens and borrowers. The provided artifacts do not show a containment boundary such as maximum liquidations per cycle, maximum SOL per cycle, or token allowlists.
"for each token with active loans:" ... "Liquidator.tryLiquidate(connection, scored)"
Add and document cycle-level limits, per-token allowlists, maximum SOL spend, maximum liquidation count, and emergency stop procedures.
Users may over-trust the bot because it presents audit and verification language.
The package includes strong audit/safety claims. They may be useful context, but they are self-contained artifact claims and should not replace independent review for a mainnet financial bot.
"Auditor: Claude Opus 4.6 (Anthropic)" and "Overall Assessment" ... "PASS"
Treat the audit documents as supporting documentation only; verify the code, dependencies, program ID, and transaction behavior independently.
The bot may keep scanning and acting on markets while it is running, not just perform one requested action.
A long-running autonomous loop is disclosed and purpose-aligned for a keeper bot, but users should understand that it can continue acting after startup until stopped.
"autonomous-scan-loop" and "default_scan_interval_ms": 60000, "min_scan_interval_ms": 5000
Run it in a controlled session with monitoring, logs, rate limits, and a clear stop/revocation process.