The Agent Skills Directory

[COMMAND_EXECUTION]: The skill configuration in SKILL.md requests broad permissions for Bash(curl *) and Bash(grep *), allowing the agent to execute shell commands and perform arbitrary network requests.
[EXTERNAL_DOWNLOADS]: The README.md and INSTALLATION.md files provide instructions to download and install multiple external skills from various community repositories to enable core functionality.
[REMOTE_CODE_EXECUTION]: The installation process promotes the use of npx skills add to fetch and execute code from third-party GitHub accounts (e.g., eng0ai, ognjengt, sickn33) which are integrated into the research workflow.
[PROMPT_INJECTION]: The skill processes untrusted content from the internet via WebSearch and WebFetch, creating a surface for indirect prompt injection.
Ingestion points: WebSearch and WebFetch tools are used by all specialist agents to retrieve data from external websites.
Boundary markers: While the skill enforces citation rules and quality gates, it lacks explicit instruction boundary markers to prevent malicious instructions in web content from influencing agent behavior.
Capability inventory: The agents have access to powerful tools including Bash(curl *), Write, and Task for managing sub-processes.
Sanitization: No explicit sanitization or filtering of retrieved web content is documented in the orchestration or agent logic.

deep-dive-research-orchestrator