The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes data from the internet, which introduces a surface for indirect prompt injection.
Ingestion points: Data is ingested through the WebSearch, WebFetch, and mcp__web_reader__webReader tools during the iterative querying phase in instructions.md.
Boundary markers: The instructions provide clear structure for agent responses and require inline citations. Phase 6 (Quality Assurance) explicitly defines a process to verify claims against original evidence.
Capability inventory: The skill can deploy sub-agents (Task), perform network operations (WebSearch), and write to the local file system (Read/Write).
Sanitization: Security is addressed through a 'Chain-of-Verification' and 'Source Triangulation' (Phase 4), which ensures that claims are corroborated by multiple authoritative sources before inclusion in the final report.
[COMMAND_EXECUTION]: The skill orchestrates multiple sub-agents to perform specialized research tasks using the Task tool.
Evidence: Phase 3 of instructions.md provides templates for 'Web Research Agents' and 'Academic/Technical Agents' that are launched in parallel using Task calls with run_in_background: true to perform background processing.

research-executor