The Agent Skills Directory

[COMMAND_EXECUTION]: The skill parses and writes command sequences (e.g., npx commands) from local catalog files into the project-level configuration file (.opencode/opencode.jsonc). This defines the specific executable behavior for local MCP servers when they are later invoked by the agent.
[PERSISTENCE_MECHANISMS]: In Step 7, the skill instructs the user to manually modify their shell configuration file (~/.zshrc) to include a source command. While this is a user-facing instruction rather than a direct automated modification, it involves system-level persistence mechanisms.
[PROMPT_INJECTION]: The skill exhibits an indirect injection surface by processing external configuration data to define agent capabilities.
Ingestion points: Catalog markdown files located at ~/.config/opencode/mcp/<name>.md are read and parsed for YAML frontmatter.
Boundary markers: No explicit boundary markers or delimiters are used when interpreting the catalog data.
Capability inventory: The skill has the capability to modify .opencode/opencode.jsonc, which directly influences the tools and commands the agent can execute within the workspace.
Sanitization: The skill lacks validation or sanitization logic for the command, url, or env fields extracted from the catalog before writing them to the active configuration.

install-mcp