The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the researchclaw package and its dependencies via pip and clones the source code from github.com/aiming-lab/AutoResearchClaw. It also suggests using a well-known academic mirror (tsinghua.edu.cn) for faster downloads in certain regions.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted data from two primary sources: user-provided research topics and academic literature fetched from arXiv and Semantic Scholar. This data is processed by an agentic pipeline with the capability to execute shell commands and write files.
Ingestion points: Research topics provided in the /researchclaw:run command and literature retrieved during the 'Literature Search' and 'Literature Analysis' stages.
Boundary markers: None detected in the skill scripts; there are no explicit delimiters to segregate paper content from agent instructions.
Capability inventory: The skill has access to Bash (executing the ResearchClaw CLI, Docker, and Python), Read, and Write tools.
Sanitization: The skill's scripts do not perform explicit sanitization of the retrieved literature content before it is processed by the pipeline.
[COMMAND_EXECUTION]: The skill utilizes several local shell scripts to manage its lifecycle. check-prereqs.sh verifies system dependencies, pre-config-write.sh performs backups, and notify-completion.sh uses system utilities like osascript (macOS) or notify-send (Linux) to provide desktop notifications.

researchclaw-cn