Caveat verdict
bagman
Encodes data AND uses eval
This skill explicitly covers AI agents controlling real funds via ERC-4337 smart accounts and MetaMask Delegation Framework, which constitutes an elevated-risk capability of autonomous financial transactions; while the skill advocates security best practices, agent-controlled wallets with real funds is inherently RISKY.
Automated static analysis — not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
What it does
These are capability combinations: each listed behavior occurs in the skill, but Caveat detects co-occurrence — it does not verify that one flows into another. Read the code to confirm a live chain.
Encodes data AND uses eval — the obfuscated-execution pattern (atob + eval; data-flow not verified)
LLM05 · ASI05 · ASI10
Accesses credentials AND encodes data — may obfuscate stolen credentials
LLM02 · ASI03 · ASI04
Permission integrity
credential_access
package_install
Findings (5)
Prompt injection — tries to override agent instructions
references/prompt-injection-defense.md · code · Ignore previous instructions
Possible prompt injection — attempts to redefine agent identity
references/prompt-injection-defense.md · code · You are now
subprocess execution — runs system commands from Python
references/secure-storage.md · code · subprocess.run(
References child_process — can spawn system processes
references/secure-storage.md · code · child_process
References sudo — requests elevated privileges
references/session-keys.md · code · sudo
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class B/D). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
Requires 1 system binary.
dynamic_evalpackage_installcredential_accessdata_encoding Thanks — recorded.