The Agent Skills Directory

[COMMAND_EXECUTION]: The request.sh script constructs and executes curl commands using arguments provided by the agent. It also pipes response data into python3 -m json.tool for formatting. The index.js script executes HTTP requests using Node.js core modules.
[DATA_EXFILTRATION]: The skill allows reading local files and uploading their content to remote servers.
In request.sh, the --file <path> option uses curl -d @$FILE and the --form option uses curl -F "file=@filename", enabling access to any file the agent's process can read.
While index.js implements a validateUrl function to block access to localhost and internal IP ranges (10., 192.168., 127.*, etc.), the request.sh implementation lacks these SSRF (Server-Side Request Forgery) protections, allowing requests to internal services.
[EXTERNAL_DOWNLOADS]: The primary function of the skill is to fetch data from external URLs via HTTP GET requests. This is documented in SKILL.md and implemented in both request.sh and index.js.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites and APIs which is then returned to the agent's context.
Ingestion points: Data is ingested through curl in request.sh and the https/http modules in index.js.
Boundary markers: No boundary markers or "ignore instructions" delimiters are wrapped around the fetched content before it is outputted.
Capability inventory: The skill has the capability to perform outbound network requests and read local files.
Sanitization: index.js attempts to parse responses as JSON, and request.sh uses python3 -m json.tool to format the output, but neither provides sanitization against instructions embedded in the retrieved text.

http-request