The Agent Skills Directory

[PROMPT_INJECTION]: The skill workflow includes a directive stating that 'a requested scoped integration already authorizes routine source/dependency edits.' This instruction is designed to override the agent's standard behavior of requiring user confirmation before modifying project files, effectively bypassing a safety checkpoint.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle and render markdown content and complex diagrams (Mermaid, KaTeX) which may originate from untrusted external sources.
Ingestion points: The {content} property in the Svelte component and the AST nodes processed by background workers.
Boundary markers: The skill documentation suggests using htmlPolicy="safe" to constrain the rendering environment.
Capability inventory: The skill operates within Svelte/SvelteKit applications which have access to browser workers and potential network-driven asset loading.
Sanitization: The instructions explicitly recommend using safe HTML policies and strict diagram defaults to mitigate risks from malicious or malformed input data.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the markstream-svelte package and its peer dependencies (katex, mermaid) from public registries. These are third-party dependencies from an unverified author and should be reviewed for supply-chain risks before installation.

markstream-svelte