The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill makes extensive use of npx starkfi@latest, which downloads and executes the latest version of the starkfi package from the npm registry at runtime. This introduces a risk of executing unverified code from an external source.
[EXTERNAL_DOWNLOADS]: The use of npx without pinned versions ensures that external code is fetched and executed during every session, which is a supply chain risk for an unverified package.
[DATA_EXFILTRATION]: The skill manages highly sensitive data, specifically Tongo private keys used for confidential transfers.
The skill instructs users to pass the private key as a command-line argument: npx starkfi@latest conf-setup --key <TONGO_PRIVATE_KEY>. This exposes the secret key to system process listings and shell history.
The skill references a sensitive local file path for credential storage: ~/.local/share/starkfi/confidential.json.
[COMMAND_EXECUTION]: The skill is configured with multiple allowed-tools that grant the agent the ability to execute shell commands using the Bash tool, including dynamic commands with wildcards (*).
[INDIRECT_PROMPT_INJECTION]: The skill processes output from external commands (e.g., conf-balance, tx-status) to inform agent decisions.
Ingestion points: Output from npx starkfi@latest commands (SKILL.md).
Boundary markers: Absent.
Capability inventory: Full subprocess execution via Bash (SKILL.md).
Sanitization: No evidence of output sanitization or validation before processing.

confidential