The explore sub agent is burning too many tokens especially around I suspect the read operations of the files are causing it

[altimate-trace-Explore_dbt_project_structure___explore_subagent_.html](https://github.com/user-attachments/files/26382480/altimate-trace-Explore_dbt_project_structure___explore_subagent_.html)

The explore sub-agent consumes an excessive number of tokens, particularly during file read operations. When tasked with exploring a codebase (e.g., reading multiple staging model SQL/YAML files), it appears to read full file contents into context even when summarized information would suffice. This leads to high token burn for exploratory tasks. The agent should be smarter about how much file content it reads — e.g., reading only config blocks or specific sections rather than entire files.

A trace file for the "Explore dbt project structure" explore subagent task is available showing the token-heavy read operations (user has the trace locally as an HTML export).

---

### Metadata

| Field | Value |
|-------|-------|
| CLI Version | 0.5.16 |
| Platform | darwin |
| Architecture | arm64 |
| OS Release | 25.3.0 |
| Category | improvement |
| Working Directory | altimate-code-test |
| Session ID | ses_2bba96d34ffegc0nuuiGP8mQ0I |