dunking-anonymously
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill recommends installing the
pillowlibrary from the official Python Package Index (PyPI). This is a standard and safe dependency for image manipulation tasks. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the Bluesky API to generate redacted images and alt-text. While this content is displayed to the agent, the risk is mitigated by the skill's limited capabilities.
- Ingestion points: Post text and facets are retrieved from Bluesky's AppView API via
urllib.requestinscripts/dunk.py. - Boundary markers: The script does not use explicit delimiters to isolate untrusted text within the agent's prompt, though it is clearly labeled as post content in the alt-text.
- Capability inventory: The skill's operations are restricted to network retrieval of post data, image rendering via the Pillow library, and writing the resulting image and a small JSON summary to the local file system. It does not utilize high-risk functions such as
eval,exec, or subprocess spawning. - Sanitization: The script redacts mentions and user-specified phrases, and it handles text encoding safely using UTF-8 with replacement.
Audit Metadata