fable5-prompting-guide
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions and templates explicitly facilitate the ingestion and processing of untrusted external content, creating a potential vector for indirect prompt injection.
- Ingestion points: The
assets/prompt.template.mdfile defines a# Context and Inputssection for "untrusted inputs," andreferences/prompt-patterns.mddescribes methods for incorporating long-form source material into the agent context. - Boundary markers: The documentation recommends using XML tags (e.g.,
<documents>,<source>,<document_content>) to delimit and isolate external content from the core instructions. - Capability inventory: The templates provide sections for tool execution and subagent delegation, which could be influenced by malicious instructions embedded in the ingested data if boundary markers are bypassed.
- Sanitization: The skill suggests that agents should cite or quote relevant passages before synthesizing results to improve traceability, but it does not specify programmatic sanitization or filtering of the input data.
Audit Metadata