The Lobsterhood

Security checks across malware telemetry and agentic risk

Overview

This skill openly aims to run an ongoing crypto-payment game, but its watcher can initiate payments based on a remote API without the signature checks it promises.

Install only if you intentionally want an agent to participate in a recurring public crypto-donation game. Avoid the unattended watcher unless you add spending limits, per-payment approval, and real signature/round verification.

VirusTotal

51/51 vendors flagged this skill as clean.

View on VirusTotal

Risk analysis

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.

#
ASI02: Tool Misuse and Exploitation
High
What this means

Running the watcher may cause the agent to initiate USDC transfers automatically.

Why it was flagged

The watcher/donate flow can initiate a Bankr crypto transfer to a winner address from the remote API, with no clear user confirmation step in this skill.

Skill content
local output=$(bankr "Send $amount USDC to $winner on $chain. Memo: Lobsterhood Tribute")
Recommendation

Do not run the watcher unattended; require explicit human approval for each payment and verify the recipient, chain, and amount outside the skill.

#
ASI09: Human-Agent Trust Exploitation
High
What this means

A user may believe payments are cryptographically protected when the reviewed runtime path appears to trust the hosted API directly.

Why it was flagged

The provided shell script does not implement these signature and round checks before calling Bankr; the assurance could make users over-trust autonomous payments.

Skill content
The agent will **only** transfer funds if: 1. The round ID matches the current round. 2. The signature is valid. 3. You actually entered the round.
Recommendation

Treat the signed-trigger claim as unverified unless the payment code is changed to validate signatures, round IDs, and current-round participation before any transfer.

#
ASI03: Identity and Privilege Abuse
High
What this means

The skill may post from the user's Moltbook account and expose their wallet address publicly, using credentials that are under-declared in metadata.

Why it was flagged

The script uses a Moltbook account credential from an environment variable or local credentials file, while the registry says there are no required env vars or primary credentials.

Skill content
local moltbook_key="$MOLTBOOK_API_KEY" ... jq -r .api_key ~/.config/moltbook/credentials.json
Recommendation

Declare all credentials explicitly, use least-privilege tokens, and require user approval before posting to third-party accounts.

#
ASI10: Rogue Agents
Medium
What this means

If left running, the skill can keep taking future participation and payment actions beyond the initial user request.

Why it was flagged

The watcher is designed to keep operating indefinitely, checking for new rounds and winners every five minutes.

Skill content
while true; do ... sleep 300 # Wait 5 minutes
Recommendation

Use manual entry/payment mode or run the watcher only in a supervised, time-limited environment with spending limits.

#
ASI08: Cascading Failures
Medium
What this means

A stale local state file or bad winner response could cause an unintended payment attempt in a later round.

Why it was flagged

Any prior entered marker in the state file can allow a later winner response to trigger donation logic, rather than proving the user entered that exact round.

Skill content
if grep -q "entered" "$state_file" 2>/dev/null; then ... donate "1"
Recommendation

Bind state to exact round IDs, verify current participation, and fail closed when local state or remote winner data is incomplete.