research-writing-assistant
Pass
Audited by Gen Agent Trust Hub on Mar 4, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: Downloads Miniconda installers from official repositories (repo.anaconda.com) to automate the setup of a Python-based research environment.
- [COMMAND_EXECUTION]: Executes local initialization scripts (init_plan.sh and init_plan.ps1) to create project structures and uses 'conda init' to manage environment persistence across shell sessions.
- [COMMAND_EXECUTION]: Runs PowerShell scripts using 'ExecutionPolicy Bypass' to facilitate the automated creation of planning templates.
- [PROMPT_INJECTION]: Contains instructions regarding project boundaries that prioritize final output generation, which may influence the model's focus during content generation.
- [PROMPT_INJECTION]: The skill has the surface for indirect prompt injection as it processes external research data like literature abstracts and user drafts while possessing 'Bash' and file-writing capabilities.
- Ingestion points: modules/literature-review.md (abstract ingestion) and modules/writing-core.md (draft processing).
- Boundary markers: Employs Markdown headers and specific output formats (e.g., 'Part 1 [LaTeX]') to separate instructions from content.
- Capability inventory: SKILL.md authorizes the use of 'Read', 'Write', 'Edit', 'Bash', and 'WebSearch' tools.
- Sanitization: Relies on instructional constraints and structured workflows rather than automated input sanitization.
Audit Metadata