The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute the opencode run "test" bash command to validate configuration changes, granting the agent the ability to execute arbitrary local commands.
[DATA_EXFILTRATION]: The configuration schema supports variable substitution from local files (e.g., {file:~/.secrets/anthropic-key}), providing a mechanism for the agent to read sensitive data into its execution context.
[COMMAND_EXECUTION]: The configuration facilitates the setup of MCP (Model Context Protocol) servers, which can be configured to execute local commands (e.g., npx -y @org/package) or connect to remote URLs.
[EXTERNAL_DOWNLOADS]: The schema allows defining remote MCP servers via URLs (e.g., https://api.example.com/mcp), which triggers the agent to fetch data or instructions from external network sources.
[PROMPT_INJECTION]: The skill contains a 'STRICTLY PROHIBITED MODELS' section that uses authoritative language to override standard model selection, potentially leading to unexpected agent behavior or failures.
[PROMPT_INJECTION]: The skill manages AGENTS.md and the instructions configuration field, which serve as surfaces for indirect prompt injection. Ingestion points: opencode.json, AGENTS.md. Boundary markers: Absent. Capability inventory: Bash tool, file editing, MCP server configuration. Sanitization: Absent.

opencode-config