corpus-discovery-dialogue
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text samples from the user in Phase 0 of the dialogue. These samples are used to generate observations and a research framework that is subsequently written to a local file in Phase 5.
- Ingestion points: User-provided corpus samples (SKILL.md, Phase 0).
- Boundary markers: None specified; the agent is not instructed to use delimiters for the untrusted input.
- Capability inventory: Writing a markdown framework to a local file path (SKILL.md, Phase 5).
- Sanitization: No explicit sanitization or filtering of the ingested content is performed before inclusion in the output file.
- [EXTERNAL_DOWNLOADS]: The skill includes instructions for installing several well-known natural language processing libraries, such as vaderSentiment, transformers, spacy, bertopic, and scattertext. These are standard packages from official registries.
- [COMMAND_EXECUTION]: The skill guides the user to execute shell commands for environment setup, including pip install for dependencies and python -m spacy download for language models.
Audit Metadata