The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to automatically update its own documentation and tools based on external observations (Notion, API responses, web search) and user confirmation. This creates a high risk of persisting malicious instructions if the external data sources are poisoned.
Ingestion points: Data enters the context via Notion tools (notion_*), web research tools (exa_*), and API interactions.
Boundary markers: Absent. The provided templates and instructions do not define delimiters or specific 'ignore' instructions for the data being processed.
Capability inventory: The skill possesses extensive capabilities across all files, including file system writes (.opencode/), bash execution (pnpm install, npx), and configuration of remote services (opencode.json).
Sanitization: Absent. There is no guidance on sanitizing or validating information before incorporating it into the agent's internal skill set.
[DYNAMIC_EXECUTION]: The skill provides templates and instructions for the agent to generate and execute its own TypeScript code (plugins and tools) at runtime in .opencode/plugin/ and .opencode/tool/.
[COMMAND_EXECUTION]: The bootstrapping and setup workflows involve executing shell commands including pnpm install and npx for environment preparation and MCP server configuration.
[EXTERNAL_DOWNLOADS]: The skill configures remote Model Context Protocol (MCP) servers and downloads packages using npx and pnpm as part of its operational flow.

self-improve