ZeroClaw Solana Risk Gate

A self-hosted Telegram agent that screens a Solana SPL or Token-2022 mint before a human decides whether to interact with it.

The agent runs locally through ZeroClaw, selects a read-only WASM tool, gathers public Solana RPC and market evidence, and returns a deterministic red/amber/green report. It cannot connect a wallet, sign a transaction, or move funds.

Submission status

• Working ZeroClaw use case: verified
• Real Telegram channel: verified
• Model-selected WASM execution: verified
• Live Solana RPC, DexScreener and GoPlus evidence: verified
• Host tests: 18 passing
• Continuous public video, three minutes or less: pending
• ZeroClaw Discord showcase post: pending

The generated player in this repository is an architecture explainer, not the required real-channel evidence. The final submission will use one continuous capture of Telegram and the live ZeroClaw runtime.

Who it is for

This is for an operator, researcher, community moderator, or treasury team that wants a fast first-pass risk gate inside the chat channel they already use. It does not replace an audit. Its job is to stop the agent from inventing a safety conclusion when evidence is incomplete.

What happens in Telegram

The operator sends a mint:

Check this Solana mint: EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v

ZeroClaw then:

1. validates that the input is exactly one canonical 32-byte Solana public key;
2. selects token-risk-check;
3. runs the WASM component with only http_client and config_read;
4. reads mint authorities, owner-level concentration, Token-2022 extensions, public DEX liquidity, and LP-control evidence;
5. returns typed facts and deterministic findings to the model;
6. posts a human-readable second-turn answer back to Telegram.

Missing or contradictory required evidence produces complete: false and a red result. The tool never silently converts an outage into a green rating.

Verified live run

On 22 July 2026, an authenticated Telegram user sent the canonical USDC mint to the locally running agent. The runtime trace recorded:

UTC time	Event
13:26:44.764	Model response parsed one tool call
13:26:44.765	token-risk-check started with the exact mint
13:26:57.991	Live risk check completed
13:26:57.993	Structured evidence returned to the agent
13:27:16.718	Second turn completed and Telegram received the answer

The release WASM was 442,725 bytes with SHA-256 537A0D594A4308DE66B5C3F1C7D336671D8E75EA9D1386441BD29A3DF09F64A5. Live concentration and liquidity values naturally change between runs.

Custody tier

T0 — read-only.

Capability	Purpose	Explicit boundary
http_client	Read public RPC and market evidence	No wallet or signing access
config_read	Read jailed operator endpoints and thresholds	The model cannot replace them

There is no recovery phrase, private key, transaction builder, signer, simulation path, or submission path in this use case.

Reproduce it

1. Build ZeroClaw with the WASM plugin runtime and a compiler backend.
2. Build the plugin from the public source branch: https://github.com/AntonsBB/zeroclaw-plugins/tree/codex/token-risk-check/plugins/token-risk-check
3. Copy manifest.toml, token_risk_check.wasm, and LICENSE into the configured plugin directory.
4. Start from config.example.toml. Add the Telegram credential with ZeroClaw's masked secret flow; never commit it.
5. Copy agent-workspace/AGENTS.md into the agent workspace.
6. Bind only the intended Telegram user, run channel doctor, start the channel, send /new, and then send the canonical request above.

Build and test the plugin:

rustup target add wasm32-wasip2
cd plugins/token-risk-check
cargo fmt --check
cargo test --locked
cargo clippy --all-targets -- -D warnings
cargo clippy --target wasm32-wasip2 -- -D warnings
cargo build --locked --target wasm32-wasip2 --release

Public implementation and evidence

• Use-case package: this repository
• Plugin source: https://github.com/AntonsBB/zeroclaw-plugins/tree/codex/token-risk-check/plugins/token-risk-check
• Archived technical-review context: zeroclaw-labs/zeroclaw-plugins#93 (closed without deleting the code so the showcase follows the sponsor's instruction to defer registry PRs until maintainers invite an implementation after judging)
• Threat model: THREAT_MODEL.md
• Reproduction config: config.example.toml
• Discord showcase draft: SHOWCASE_DRAFT.md
• Architecture player: https://antonsbb.github.io/token-risk-check-demo/

Limitations

• It is a deterministic screen, not a contract audit or price prediction.
• Largest-account RPC data is a bounded sample.
• Public data sources can rate-limit or change.
• Concentrated-liquidity control cannot be proven from fungible LP-token fields, so that condition remains explicitly unverified.
• Token metadata, off-chain identity, social reputation, and price momentum are outside this T0 version.

License

The showcase material is MIT-licensed. The plugin source carries its own MIT license in the linked implementation branch.