The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is vulnerable to instructions embedded within the academic papers it retrieves and processes.
Ingestion points: The agent fetches the full HTML content of scientific papers from a remote source (arxiv.org) using web.fetch in Step 1.
Boundary markers: The instructions do not define clear boundaries or provide system-level warnings to the agent to ignore any natural language instructions found within the downloaded paper text.
Capability inventory: The agent possesses web.fetch for network requests, file-system write access to the ~/.openclaw/workspace/ directory, and the ability to spawn sub-agents with verbatim instructions derived from the processed content.
Sanitization: There is no evidence of filtering, escaping, or sanitizing the paper content before it is used to generate summaries or define tasks for sub-agents.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and research papers from arxiv.org. This is a well-known service and the activity is central to the skill's stated purpose of academic summarization.

paper-digest