The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions require cloning a repository from an untrusted source (https://github.com/aiming-lab/AutoResearchClaw.git) and installing it as an editable package.
[DYNAMIC_EXECUTION]: The research pipeline includes stages for autonomous code generation (Stage 10: CODE_GENERATION) and execution (Stage 12: EXPERIMENT_RUN). While the documentation mentions sandboxed execution, the process of running code generated by an LLM in response to untrusted data carries inherent security risks.
[PERSISTENCE]: The configuration for the 'OpenClaw bridge' includes a use_cron parameter which, when enabled, allows the skill to schedule recurring research tasks on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by ingesting literature from external sources like arXiv and Semantic Scholar to inform its autonomous code generation and experimentation phases.
Ingestion points: Data enters the context during the LITERATURE_COLLECT and KNOWLEDGE_EXTRACT stages through API calls to external services.
Boundary markers: The documentation does not specify the use of delimiters or warnings to prevent the agent from following instructions embedded in the collected literature.
Capability inventory: The skill possesses capabilities to write files to the deliverables/ directory and execute Python code in a sandbox environment.
Sanitization: There is no mention of sanitizing or escaping the external content before it is processed by the multi-agent debate and code generation modules.

autoresearchclaw-autonomous-research