Coordinate Parallel Subagents
Use subagents when independent investigations need more context than a few tool calls.
#Scenario: audit every provider scraper
The tracker has separate implementations for OpenAI, Anthropic, Google, and AWS. They can be reviewed independently, then compared against the shared DeprecationEntry contract.
Audit the provider scrapers with three parallel subagents:
- Review scraper/openai_scraper.py and scraper/anthropic_scraper.py.
- Review scraper/vertex_scraper.py and scraper/gemini_scraper.py.
- Review scraper/bedrock_scraper.py.
Each subagent must:
- compare its scrapers with scraper/base.py
- report date parsing, deduplication, and missing-data risks
- cite file paths
- make no edits
After they finish, combine the findings into one prioritized report.
Each assignment includes its files, checks, output, and no-edit constraint because subagents work independently and cannot ask follow-up questions.
#Review the combined result
Use Tab to inspect active subagents. Sensitive confirmations still appear in the main terminal.
When all workers finish, ask the main agent to resolve disagreements and remove duplicate findings. Delegate edits only when workers own different files; worktrees reduce interference but do not make overlapping changes safe.
#Expected result
You receive three bounded audits and one reconciled list of scraper risks, while the main conversation retains final ownership.