The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates a Python script (generate_report.py) at runtime to aggregate research results into a Markdown report. It then executes this generated script using the system's Python interpreter to perform data processing.
[COMMAND_EXECUTION]: The skill uses shell commands to execute a validation script (validate_json.py) intended to ensure that JSON output from sub-agents matches the defined schema.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content retrieved from the web via WebSearch and WebFetch tools. This data is then processed and formatted into final reports, creating a potential surface for indirect prompt injection attacks where malicious content from the internet could influence the agent's output.
Ingestion points: Research results collected from external websites via sub-agents in research-deep/SKILL.md and agents/web-search-agent.md.
Boundary markers: The skill does not explicitly define robust delimiters or "ignore instructions" warnings for the external data it processes during report generation.
Capability inventory: The skill has Write access to the filesystem to create reports and scripts, and Bash access to execute code.
Sanitization: The skill implements basic formatting such as line breaks for long strings and uses yaml.safe_load for parsing, but lacks comprehensive sanitization against malicious instructions embedded in the research results themselves.

deep-research