The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (such as GitHub repositories and web documentation) to answer user queries. This creates a surface for indirect prompt injection, where instructions embedded in external content could potentially influence the agent. However, the risk is mitigated by the skill's restricted toolset.
Ingestion points: The skill uses WebFetch, WebSearch, and mcp__deepwiki tools to retrieve technical content from external repositories and the internet.
Boundary markers: There are no explicit instructions in the prompt to separate ingested data from the system instructions using delimiters.
Capability inventory: The skill is restricted to research tools (Read, Glob, Grep, Task, WebFetch, WebSearch, and specific MCP tools). It lacks the capability to write files, execute shell commands, or access network resources outside of its defined scope.
Sanitization: No explicit sanitization or filtering of external content is specified before processing.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch data from the vendor's official domain (zkcompression.com) and GitHub organization (github.com/Lightprotocol). It also retrieves technical context from official repositories of well-known organizations. These network operations are limited to documentation retrieval and research tasks consistent with the skill's primary function.

ask-mcp