The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions direct users to set the FITTINGS_PRIVATE_KEY environment variable with a plaintext blockchain private key. The call.mjs script accesses this key to sign financial transaction authorizations. Handling raw private keys in environment variables is a high-risk practice as they can be easily exposed through environment logs or memory dumps.
[EXTERNAL_DOWNLOADS]: The documentation instructs the agent to download a helper script from https://fittings.sh/skill/call.mjs using curl before execution.
[COMMAND_EXECUTION]: The skill executes the downloaded call.mjs script using the node runtime to perform network operations and cryptographic signing.
[REMOTE_CODE_EXECUTION]: The skill downloads and runs code from a remote source. The call.mjs script further performs dynamic imports of external packages (@x402/fetch, @x402/evm, and viem) at runtime, increasing the execution risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests authoritative regulation text from ecfr.gov via fittings.sh into the agent's context. This represents an external data ingestion point where no sanitization or boundary markers are applied to the returned text. While the source is a government service, the lack of isolation between untrusted data and agent instructions is a recognized vulnerability surface.

cfr-section-text