The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to download a JavaScript file from a remote server and execute it using Node.js via curl -sO https://fittings.sh/skill/call.mjs followed by node call.mjs. This allows for arbitrary code execution where the logic is controlled by the remote server at runtime.
[CREDENTIALS_UNSAFE]: The documentation explicitly tells users to export a private key to their environment variables using export FITTINGS_PRIVATE_KEY=0x.... Storing private keys in plaintext environment variables or shell history is a high-risk security practice as it exposes sensitive credentials to other processes and users on the system.
[EXTERNAL_DOWNLOADS]: The skill performs multiple downloads from https://fittings.sh, including catalog data and the execution script. While this is the vendor's own domain, the pattern of fetching and running external scripts introduces supply chain risk.
[COMMAND_EXECUTION]: The skill provides bash commands for the agent or user to execute, such as curl -s ... | jq, which involves piping network output directly into a command-line tool.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it ingests untrusted data from the external USAspending.gov API (proxied via fittings.sh) based on user input. Evidence includes: 1. Ingestion point: recipient parameter and the resulting API data; 2. Boundary markers: Absent; 3. Capability inventory: Network requests and local script execution; 4. Sanitization: No sanitization or validation of the API response is mentioned before processing.

fittings-gov-federal-spending