bun
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions and examples facilitate the creation of HTTP servers that ingest untrusted data, creating a potential surface for indirect prompt injection attacks.
- Ingestion points: Example 1 demonstrates the use of
request.json()to parse incoming data from HTTP requests in a REST API context. - Boundary markers: The instructions lack guidance for the agent to use delimiters or protective instructions when generating code that handles external input.
- Capability inventory: The skill encourages the use of file system operations (
Bun.write()), package management (bun install), and code compilation/bundling (Bun.build()), which could be targeted by an injection attack. - Sanitization: The provided examples do not include input validation or sanitization logic for data retrieved from network requests.
- [INDIRECT_PROMPT_INJECTION]: The skill includes external metadata files that make self-referential claims about its own security verdict.
- Evidence: The
_scores.jsonfile contains a pre-defined"verdict": "SAFE"field, which matches patterns of content designed to influence automated analysis results.
Audit Metadata