The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The script scripts/register.sh explicitly instructs and checks for the installation of Foundry using curl -L https://foundry.paradigm.xyz | bash. This pattern allows a third-party server to execute arbitrary commands on the user's host machine, and the source is not within the defined list of trusted organizations.
[EXTERNAL_DOWNLOADS] (HIGH): The skill performs unvalidated network requests to the Pinata IPFS API in scripts/register.sh using curl. While intended for agent registration, this mechanism could be abused for data exfiltration or to interact with malicious endpoints if the PINATA_JWT or metadata variables are manipulated.
[COMMAND_EXECUTION] (HIGH): Multiple scripts (scripts/register.sh, scripts/give-feedback.sh, scripts/check-agent.sh) execute shell commands using the cast tool. These commands interpolate user-provided arguments and environment variables without sanitization, creating a high risk of command injection if the AI agent is manipulated into passing malicious payloads.
[PROMPT_INJECTION] (LOW): Detected an Indirect Prompt Injection surface (Category 8).
Ingestion points: scripts/check-agent.sh reads TOKEN_URI and client lists directly from the Avalanche blockchain, which can contain attacker-controlled strings.
Boundary markers: Absent. Blockchain data is printed directly to the shell and returned to the agent context.
Capability inventory: The skill has the capability to sign and send transactions via cast send in scripts/register.sh and scripts/give-feedback.sh.
Sanitization: Absent. No filtering is performed on data retrieved from the Identity or Reputation registries.

erc8004-avalanche