research-pipeline
Warn
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically resolves paths for internal management tools (such as run_state.py, iteration_log.py, and watchdog.py) from an environment configuration file (.aris/installed-skills-codex.txt) and executes them using python3. This involves dynamic execution from computed paths, which is a potential risk if the underlying configuration file is modified to point to malicious scripts.
- [COMMAND_EXECUTION]: The skill utilizes shell commands and Python scripts for workflow orchestration, state management, and large file handling. For example, it uses bash here-docs (cat << 'EOF') to write files in chunks when standard tool limits are exceeded.
- [INDIRECT_PROMPT_INJECTION]: There is a significant attack surface for indirect prompt injection as the pipeline ingests untrusted data from multiple external sources. This includes user-provided research briefs (RESEARCH_BRIEF.md), external repositories cloned via the BASE_REPO parameter, and literature metadata or content fetched from arXiv. This data influences autonomous decision-making in the idea discovery and review stages.
- [EXTERNAL_DOWNLOADS]: The pipeline is designed to fetch content from external services. It clones codebases from GitHub repositories and retrieves scientific metadata from arXiv. These operations are core to the skill's functionality but introduce external data into the agent's context.
Audit Metadata