The Agent Skills Directory

[PROMPT_INJECTION]: The skill's description uses the 'CRITICAL:' prefix, which is a common pattern for attempting to override agent instructions or bypass safety filters by masquerading as a high-priority system directive.
[METADATA_POISONING]: The YAML frontmatter includes a non-standard 'risk: critical' field. This is an attempt to inject a security classification or status into the skill's metadata, potentially to deceive automated scanners or influence the agent's perception of the skill's authority.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of untrusted data from external LLM providers via Server-Sent Events (SSE) streaming.
Ingestion points: The parse_sse function in SKILL.md processes raw byte streams from external network requests.
Boundary markers: No explicit boundary markers or 'ignore embedded instructions' warnings are present in the provided code snippets to distinguish between instructions and data.
Capability inventory: The skill includes a spawn() utility for background task execution and uses Cx::post_action for UI state updates based on the ingested content.
Sanitization: No logic for sanitizing or escaping the content from the LLM stream is demonstrated before the content is passed to UI components or application state.

molykit