huanzhi-character-prompt-generator
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves using
WebSearchandWebFetchto collect character data from external sources such as Moegirl Wiki, Fandom, and fan forums. This introduces a vulnerability to indirect prompt injection, where an attacker could poison these external pages with instructions intended to hijack the agent's behavior or manipulate the generated output. - Ingestion points:
WebSearchandWebFetchtools are used to ingest untrusted data from the internet (specified inSKILL.mdunder Step 1). - Boundary markers: While the skill instructs the agent to "cross-validate" and "refine" information, it lacks explicit technical sanitization to strip potential prompt injection payloads from the fetched content.
- Capability inventory: The agent has the capability to search the web and write files to the local file system using the
Writetool. - Sanitization: No specific sanitization logic is provided to filter out instructions embedded in the external data.
- [COMMAND_EXECUTION]: The skill directs the agent to save generated prompts as
.mdfiles to a specific local directory (/Users/lishibo/code/skill/幻织/output/) using a file writing tool. While this is the intended functionality for a file-generating tool, it involves interacting with the local file system. - [EXTERNAL_DOWNLOADS]: The skill references and fetches content from several well-known and community-driven platforms, including GitHub, Fandom, and Moegirl Wiki. These resources are utilized for gathering character lore and documentation.
Audit Metadata