Explore

Gives the agent a single explore tool: a read-only sub-agent that navigates and searches the codebase, then returns a synthesised answer instead of dumping raw files into the main conversation.

Typical uses: "where is X defined and who calls it?", "how does subsystem Y fit together?"

Configuration

plugins:
  - module: codumentor.plugins.explore
    class: ExplorePlugin
    args:
      enabled: true
      target_agents: "main"
      # max_iterations: 40   # optional cap on explorer steps

• enabled (optional, default true).
• target_agents (optional, default "main"): "main", "subagent", or "all".
• max_iterations (optional): Maximum reasoning steps before the explorer must answer. If omitted, the global agent iteration limit applies. max_turns is a deprecated alias.

Tool: explore

Argument	Required	Description
query	yes	Self-contained briefing: framing, hypotheses, specific questions, and hints about relevant files. The explorer does not see the parent conversation.
subagent_id	no	Continue a previous exploration session.

The explorer can search with ripgrep (when rg is on PATH), semantically search the knowledge base (when vector search is configured and not disabled), read files, and run read-only shell commands (listings, find, head/tail, wc, …). It has no write, edit, or run-to-modify tools.

Notes

• Write everything the explorer needs into query; reuse subagent_id for follow-ups.
• Ripgrep is recommended but not required. Semantic search is skipped when tools.disable_vector_search is set.
• To delegate to an external CLI agent instead, see External Agent.