The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the hyperresearch Python package from a public registry. This is an external dependency not sourced from a pre-verified trusted organization.
[DYNAMIC_EXECUTION]: The skill uses the hyperresearch CLI tool to generate and install new executable skills, subagent configurations, and hooks into the project directories (.claude/ or .agents/) at runtime. This behavior allows for the dynamic injection of new agent logic based on external tool output.
[PERSISTENCE]: The installation process adds persistent hooks to the agent's environment, such as a PreToolUse hook for Claude Code and a Stop hook for OpenAI Codex. These modifications allow the research pipeline to maintain control across agent sessions and prevent session closure while a run is in progress.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to perform multi-source analysis by fetching data from the network, which creates a vulnerability surface for indirect prompt injection from untrusted external content.
Ingestion points: External network sources and literature fetched during the 16-step research pipeline execution.
Boundary markers: None are specified in the bootstrap instructions; the security of the data processing depends on the implementation of the dynamically installed pipeline components.
Capability inventory: The skill possesses shell execution capabilities (bash), file reading (cat), and the ability to register new skills and lifecycle hooks within the agent environment.
Sanitization: There is no evidence of input validation or content sanitization within the bootstrap logic.

deep-research