The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing and running the @steipete/oracle package from the NPM registry using standard package managers or npx.
[COMMAND_EXECUTION]: The skill provides numerous command-line examples for the oracle tool, which accesses the local file system and performs network requests to transmit data to remote LLM engines or to host a remote browser server (oracle serve).
[PROMPT_INJECTION]: The tool's primary workflow is vulnerable to indirect prompt injection, as it ingests local repository content and provides it as context to an AI model.
Ingestion points: Local source files and directories specified via the --file flag (e.g., oracle ... --file "src/**" in SKILL.md).
Boundary markers: The skill does not specify any delimiters or protective prompts to prevent an LLM from obeying instructions found within the bundled source code.
Capability inventory: The agent is directed to use the oracle CLI to read file contents and send them to external API providers or browser-based models.
Sanitization: No sanitization or filtering of file content is described; the tool is designed to relay the raw content of the selected files.

oracle