kr2s-was-i-exploited
Warn
Audited by Gen Agent Trust Hub on Jul 31, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to provide several forensic scripts (‘kr2s_scan_active_storage_blobs.rb’, ‘kr2s_recover_variants.rb’, ‘kr2s_decode_variants.rb’, and ‘kr2s_external_file_lists.py’) to be executed by a human operator in a production environment. This process interacts with the production database and object storage to identify exploitation evidence.\n- [DATA_EXFILTRATION]: Facilitates the reconstruction of exfiltrated data by decoding pixel values from image variants back into original byte sequences. While used for forensic auditing, this demonstrates the capability to extract sensitive information from stored application artifacts.\n- [EXTERNAL_DOWNLOADS]: References and requires the use of well-known external libraries, specifically the ‘h5py’ Python package and the ‘ruby-vips’ library, to perform data parsing and image analysis.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes untrusted data uploaded by potential attackers.\n
- Ingestion points: Processes Active Storage blobs (crafted MAT and HDF5 files) and database variant records as specified in ‘SKILL.md’ and ‘references/guide.md’.\n
- Boundary markers: No explicit delimiters or ‘ignore embedded instructions’ warnings are specified for the agent when interpreting the output of the forensic scans.\n
- Capability inventory: The analysis environment includes execution capabilities via ‘bin/rails runner’ and Python scripts, which access both the filesystem and database.\n
- Sanitization: The workflow includes structural verification of file headers to detect ‘crafted’ files, but does not explicitly describe sanitization or escaping of the content recovered from those files before further processing.
Audit Metadata