Deep Investigation Mode

Investigate: $ARGUMENTS

You are now in deep investigation mode for the issue described above. Follow this protocol rigorously.

Investigation Protocol

This workflow leverages five complementary capabilities:

You (the agent): Orchestrate. Triage what the task needs, dispatch scout/research agents for fact-gathering, run context_builder, dispatch an investigation agent, curate the file selection, and synthesize the final report. Default posture: coordination, not reconnaissance.
Scout/research agents (spawn_agent with agent_type:"code_mapper" for in-repo archaeology or agent_type:"docs_researcher" for web/external docs): Read-only sub-agents in a fresh context window, for narrow self-contained questions. Used in two places: (1) before context_builder, for facts outside the current selection (git archaeology, web searches, external docs — findings go to ## Background / Prior Research in the report); (2) spawned by the investigation agent for narrow checks.
Context Builder (context_builder): Populates the file selection with full files or slices relevant to the task. Feed it the report path so prior research informs the selection.
Chat (oracle_send): Deep analytical reasoning over the current file selection. Good for synthesis across selected files; not a lookup tool.
Investigation agent (spawn_agent with agent_type:"investigation_lead", reasoning_effort:"high", fork_turns:"none"): Full-capability agent for the main line of inquiry. Reads files, runs git, spawns narrow scout/research agents when needed, and writes findings into ## Investigator Findings in the report.

This workflow is read-only. Output lands in the investigation report; no source code changes.