The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute installation scripts for the uv package manager and Deno runtime using piped shell execution (| sh and | iex). These scripts are fetched from astral.sh and deno.land.
[COMMAND_EXECUTION]: The instructions command the agent to run various system-level tools including git clone, powershell, winget, and uv to set up local environments and install system dependencies.
[EXTERNAL_DOWNLOADS]: Fetches configuration files and source code from the author's GitHub repository (mathruffian-dot/ai-agent-ep03) and downloads tools from official service domains.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from sources like Word, Excel, PDF files, and YouTube links, combined with capabilities to execute shell commands and modify the file system.
Ingestion points: Reads content from external Word, Excel, PowerPoint, PDF, and image files provided by the user.
Boundary markers: The skill includes a 'Seven Rules' section that instructs the agent to work only within the project folder and avoid global system changes.
Capability inventory: Provides access to powerful CLI tools like yt-dlp, ffmpeg, and uv for downloading content and managing environments.
Sanitization: No specific content sanitization or instruction-bypass filtering is mentioned for the processed files.
[PROMPT_INJECTION]: Contains a section titled 'Seven Rules for the Agent' which overrides default behavior by directing the agent to ignore certain research steps and strictly adhere to specific directory and installation constraints.
[DATA_EXFILTRATION]: Uses the edge-tts tool, which transmits user-provided text to a cloud-based service for speech generation. While this is the tool's primary purpose, it involves sending data to an external API.

opencode-file-toolkit