url-summarization
Warn
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically constructs a file path to load output templates using a variable identifier:
../summarizer/templates/{format_id}.md. This pattern represents dynamic loading from a computed path, which could potentially be exploited for path traversal or local file inclusion if the{format_id}is influenced by untrusted input or malicious data processed by the agent. - [INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves fetching and processing data from arbitrary URLs, creating a surface for instructions embedded in external content to influence agent behavior.
- Ingestion points: External URLs fetched via the
WebFetchtool ormcp__Ref__ref_read_urlas defined in the fetching methodology section ofSKILL.md. - Boundary markers: The skill implements a logical boundary using the 'Quote-Grounding Technique', requiring the agent to extract verbatim passages before summarizing.
- Capability inventory: The skill utilizes network-capable tools (
WebFetch,mcp__Ref__ref_read_url) to retrieve content. - Sanitization: Fidelity rules require 'Read before summarizing' and 'Extract before abstracting' to prevent fabrication, though these do not strictly sanitize for malicious prompt instructions.
- [EXTERNAL_DOWNLOADS]: The skill fetches documentation and configuration guidelines from Anthropic's official documentation repository.
Audit Metadata