The Agent Skills Directory

[SAFE]: The skill consists of documentation and instructions for delegating tasks to subagents. It does not contain any scripts, executable code, or suspicious commands. The installation instructions in the README point to the author's official GitHub repository.
[SAFE]: All external URL references in references/claude-explore-agent-research.md target well-known and trusted official domains (claude.com and code.claude.com) for technical documentation purposes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process codebase content through explorer subagents, which naturally introduces a surface for indirect prompt injection if a target repository contains malicious instructions. However, the skill implements structural guardrails to mitigate this risk.
Ingestion points: Explorer subagents are instructed to read files and search across the target codebase to answer research questions (SKILL.md).
Boundary markers: The workflow requires explorers to use a specific "Explorer Prompt Template" and produce a "Key files table," which helps isolate raw data from the main reasoning loop.
Capability inventory: The skill itself is instructional and does not perform file system writes, network operations, or subprocess executions (SKILL.md).
Sanitization: There are no explicit sanitization steps mentioned for the code content being processed, but the reconnaissance is explicitly restricted to be "read-only."

explore