The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for using the opencode CLI and its subcommands (such as opencode run, opencode serve, and opencode acp), which are designed to execute shell commands and manage background processes as part of the agent's primary coding functionality.
[INDIRECT_PROMPT_INJECTION]: The skill describes an agent runtime that ingests untrusted data from multiple ingestion points, including local project files, GitHub pull requests, and Model Context Protocol (MCP) servers. The capability inventory for the skill includes broad shell access and file system modification. Boundary markers are documented in the form of a native permission system (allow/ask/deny rules) designed to gate tool actions, and the skill provides specific warnings against using the --dangerously-skip-permissions flag in sensitive environments.

opencode