RFC: Request context:fork for subagents - Independent token budgets for pr-review-toolkit

[evansenter]

Summary

Request/track the context: fork feature for Claude Code subagents. This would enable pr-review-toolkit agents to run with independent token budgets and isolated execution environments.

Problem / Motivation

When running multiple pr-review-toolkit agents in parallel (code-reviewer, silent-failure-hunter, type-design-analyzer, etc.):

• All agents share the parent session's 200K token budget
• Large PRs cause 8192 token output limits to be exceeded
• Each failed agent wastes 10-15K tokens
• Must run agents sequentially as a workaround

Context

• Discovered during: Release notes review for Claude Code updates
• Relevant issues: [FEATURE] Independent Context Windows for Sub-Agents anthropics/claude-code#10212, #9284, #5812, #12790
• Related files: home/.claude/agents/, pr-review-toolkit plugin

Proposed Solution

When Claude Code implements context: fork, update pr-review-toolkit agent definitions to use it:

# In agent definition
context: fork  # Each agent gets independent 200K token budget

Benefits

Agent	Current Constraint	With context:fork
code-reviewer	Shares token budget	Independent analysis context
silent-failure-hunter	Output truncated	Full error handling analysis
type-design-analyzer	Limited by parent	Complete type analysis
pr-test-analyzer	Sequential execution	Parallel with isolated env

Assumptions

Assumption	Confidence	Impact if Wrong
context:fork will be implemented	Medium	Wait for alternative
Will work with Task tool batching	High	Minor refactoring

Open Questions

1. When will context:fork be available in Claude Code?
2. Will it require plugin updates or just agent config changes?

Actionable Requirements

#	Requirement	Owner	Blocked By
1	Track anthropics/claude-code for context:fork implementation	Claude	Anthropic
2	Update pr-review-toolkit agents when available	Claude	#1
3	Test parallel execution with independent budgets	Claude	#2

Implementation Checklist

• Monitor Claude Code releases for context:fork
• Update agent definitions when feature ships
• Validate parallel agent execution
• Measure token efficiency improvements