The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to support its research and reporting workflow. It utilizes date +%Y-%m-%d for timestamping research. It uses curl for downloading external research materials such as PDFs and datasets. It employs pandoc and md-to-pdf to convert generated Markdown reports into PDF format. Use of these tools on content sourced from the web carries a risk if the input contains exploits targeting the underlying rendering engines.
[EXTERNAL_DOWNLOADS]: The skill is centered around fetching data from the public internet. It uses WebSearch and WebFetch as core research capabilities. Instructions in references/parallel-search.md explicitly permit sub-agents to use curl for downloading external files for summarization.
[PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection. 1. Ingestion points: Untrusted data enters the agent context via WebSearch, WebFetch, and curl downloads (SKILL.md, references/parallel-search.md). 2. Boundary markers: The instructions lack explicit boundary markers or 'ignore' directives to prevent the agent from following instructions embedded within the retrieved research data. 3. Capability inventory: The agent has access to Write, Agent (for spawning sub-agents), and Bash (for curl, pandoc, and md-to-pdf), which could be leveraged if an injection is successful. 4. Sanitization: There is no evidence of sanitization or filtering of external content before it is processed or written to the report file.

deep-research