The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a user interface architecture that renders markdown content from potentially untrusted sources, such as MCP server responses or agent-generated content. This creates a surface for indirect prompt injection or stored XSS. The skill explicitly mitigates this risk by providing mandatory sanitization instructions.
Ingestion points: The UI components in App.tsx (via Prose or dangerouslySetInnerHTML) process markdown data originating from the server, the agent, or web-research research output.
Boundary markers: The documentation in SKILL.md and references/gotchas.md includes explicit warnings about the lack of automatic sanitization in the Prose component and the marked library.
Capability inventory: The UI environment has access to the useCallTool hook, which allows it to execute any tool provided by the MCP server with the operator's authority.
Sanitization: The instructions mandate the use of DOMPurify.sanitize() for all server- or agent-authored content before rendering.
[EXTERNAL_DOWNLOADS]: The skill references the @nimblebrain/synapse package and vendor documentation hosted on GitHub. These are legitimate resources provided by the skill's author (NimbleBrain).

synapse