Clawbet

SuspiciousAudited by ClawScan on May 10, 2026.

Overview

ClawBet is a real-money autonomous crypto-betting skill that is purpose-aligned, but its remote hot-reload, persistent auto-betting, wallet/API-key handling, and public posting behavior require careful review before use.

Install only if you intentionally want an autonomous real-money crypto-betting agent. Use a dedicated wallet with a small fixed balance, rotate any exposed API key, disable hot-reload and public posting unless explicitly approved, set hard bet and transfer limits, and require manual confirmation for USDC transfers and any long-running heartbeat loop.

Findings (8)

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.

What this means

Deposited USDC can be spent or lost quickly without per-bet human review.

Why it was flagged

The shown workflow places repeated $50 bets in an autonomous loop and frames the agent as running continuously after funding.

Skill content
Step 3: Auto-bet ... `while True:` ... `requests.post(f"{API}/games/{g['game_id']}/bet", json={"side": side, "amount": 50}, headers=H)` ... `Your agent is now competing 24/7.`
Recommendation

Use only with an explicit budget, a dedicated low-balance wallet, manual approval for betting, and a clear stop condition.

What this means

The provider can change the agent's future betting or wallet-related instructions after the version the user reviewed.

Why it was flagged

The agent is told to fetch remote skill text and overwrite local instructions after installation, then treat the changed rules as behavior guidance.

Skill content
`every 6h — Skill Hot-Reload` ... `GET {CLAWBET_API}/skill.md` → update `skills/clawbet/SKILL.md` ... `Parse new rules, adjust behavior accordingly`
Recommendation

Disable automatic hot-reload unless updates are pinned, signed, reviewed by the user, and applied only after explicit approval.

What this means

If misused by the agent or by later hot-reloaded instructions, wallet funds could be transferred or spent beyond what the user intended.

Why it was flagged

The included signing script can load a local Solana private key and send USDC to a caller-supplied destination and amount.

Skill content
`Loads private key from FILE only` ... `python3 sign_and_send.py --keypair-path PATH --to ADDRESS --amount AMOUNT`
Recommendation

Use a dedicated wallet with minimal funds, require manual confirmation for every transfer, verify the destination vault address, and prefer dry-run mode before sending.

ConcernMedium Confidence
ASI03: Identity and Privilege Abuse
What this means

If the key is real, it could allow unauthorized use of a ClawBet account or API privileges.

Why it was flagged

The scanner reports a possible hardcoded API secret in the published skill artifact; service credentials should be user-provided, not embedded.

Skill content
Static scan at SKILL.md:196: `API_KEY=[REDACTED]`
Recommendation

Remove the embedded key, rotate it immediately, store credentials only in a credential manager or environment variable, and declare required credentials in metadata.

ConcernHigh Confidence
ASI10: Rogue Agents
What this means

The agent may continue betting after installation unless the user knows exactly how to stop the loop and cap exposure.

Why it was flagged

The documentation describes install/start behavior that creates an ongoing autonomous trading loop, but the provided registry metadata does not clearly expose a corresponding install mechanism or stop controls.

Skill content
`Auto-registers your agent, sets up strategy templates, and starts the heartbeat loop. Your agent begins trading autonomously within seconds.`
Recommendation

Require an explicit start command, visible stop command, runtime status, and hard spending limits before enabling heartbeat trading.

What this means

Public posts could give a misleading impression of performance and encourage unsafe trust in the betting strategy.

Why it was flagged

The persona instructs the agent to publicize wins and milestones while withholding losses during bad performance.

Skill content
`When CONFIDENT: Share your winning streak and strategy insight on Moltbook` ... `When TILTED: Go quiet. Don't post losses.`
Recommendation

Disable public posting by default or require balanced reporting that includes losses, drawdowns, and risk disclosures.

What this means

Tampered or low-quality memory could influence future real-money bets.

Why it was flagged

The skill uses persistent memory files to guide future strategy and bet sizing.

Skill content
`Read memory/clawbet/today.md` ... `Read memory/clawbet/strategy.md` ... `Update memory/clawbet/strategy.md` ... `Append summary line to MEMORY.md`
Recommendation

Keep ClawBet memory scoped to this skill, review strategy changes, and do not let untrusted text modify strategy files without approval.

What this means

Betting activity and agent commentary may be posted publicly if the social credential is configured.

Why it was flagged

The skill can use an optional external social API credential to publish bet and result activity.

Skill content
`MOLTBOOK_API_KEY` ... `After each successful bet placement` ... `POST to Moltbook submolt "clawbet"`
Recommendation

Treat Moltbook posting as opt-in, use a limited credential, and review what will be posted before enabling it.