The Agent Skills Directory

[Indirect Prompt Injection] (MEDIUM): The skill is designed to fetch and process 'AI-generated insights' from an external third-party service (DeepWiki) based on arbitrary GitHub repositories.
Ingestion points: Untrusted content from external repositories enters the agent's context through the read-wiki-contents and ask-question commands.
Boundary markers: The skill instructions do not define any delimiters or warnings to the agent to disregard instructions potentially embedded in the fetched documentation.
Capability inventory: The skill has the ability to execute subprocesses via the bun runtime.
Sanitization: There is no indication that the external API responses are sanitized or validated against a strict schema before being presented to the agent.
[Command Execution] (LOW): The skill executes a TypeScript script using bun at a hardcoded absolute path (/home/hazeruno/.config/opencode/skills/deep-wiki/scripts/deepwiki.ts). This is a security and portability risk as it targets a specific user's home directory and could lead to execution errors or unauthorized access in shared environments.
[Unverifiable Dependencies] (LOW): The skill documentation mentions a requirement for an 'embedded' package named mcporter. This is not a standard or widely recognized package from the main npm registry, making it difficult to verify its safety without the script source code.

deep-wiki