hono-cloudflare
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes instructions for ingesting untrusted data from HTTP requests and storing it in edge databases or object storage. This represents a potential vulnerability surface if the processed data is eventually consumed by an LLM without proper sanitization.
- Ingestion points: Data enters the skill's logic via
c.req.json(),c.req.param(), andc.req.query()as shown in the code snippets withinSKILL.md. - Boundary markers: The skill does not explicitly define prompt boundary markers, as it describes general-purpose API patterns rather than LLM-specific data handling.
- Capability inventory: The skill provides code for writing to persistent storage (
c.env.DB.run,c.env.BUCKET.put,c.env.CACHE.put) and dispatching tasks to background processes (c.env.MY_QUEUE.send). - Sanitization: The code examples demonstrate secure practices by using parameterized SQL statements (e.g.,
prepare(...).bind(id).first()) which effectively mitigates direct injection into backend commands.
Audit Metadata