The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The INSTALL.md file contains setup instructions that guide users or agents to use sudo for copying skill files into protected system directories and modifying file ownership/permissions (chmod, chown).
[COMMAND_EXECUTION]: The skill facilitates interaction with the opencode CLI tool, allowing the AI to run tasks, refactor code, and manage GitHub PRs through terminal commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data from local codebases and GitHub Pull Requests, creating a surface for indirect prompt injection.
Ingestion points: Source code files and GitHub Pull Request data.
Boundary markers: No explicit boundary markers or delimiters defined to separate code data from instructions.
Capability inventory: File system writes, command execution, and network sharing via the opencode tool.
Sanitization: None identified within the skill's instruction set.
[EXTERNAL_DOWNLOADS]: The documentation directs the installation of the core binary via Homebrew (brew install opencode), which is a well-known and trusted package management service.
[PERSISTENCE]: Instructions in the README.md and INSTALL.md files guide the user to modify their ~/.zshrc file to ensure necessary system paths are preserved across shell sessions.
[DYNAMIC_EXECUTION]: The tool is designed to generate and execute code at runtime based on natural language instructions, which is the core intended functionality of the AI editor.
[DATA_EXFILTRATION]: The skill includes functionality to share sessions externally (--share), which generates a public link and could lead to accidental exposure of source code context if not managed carefully by the user.

opencode