Bug: resume command ignores default model setting in settings.json

Bug Report

Summary: The resume command does not respect the default model setting in ~/.claude/settings.json, instead defaulting to base sonnet (200k context) even when configured for sonnet[1m] (1M context).

Environment:

• Claude Code version: Latest (as of 2025-11-04)
• OS: macOS (Darwin 25.0.0)
• Model configured: sonnet[1m] (1M context)
• Model actually used: sonnet (200k context)

---

Steps to Reproduce

1. Configure default model in ~/.claude/settings.json:

``json
{
"model": "sonnet[1m]"
}
``

1. Resume a project using the resume command:

``bash
resume @projects/active/feature-microsoftofficeo365mcpintegration/
``

1. Check context usage with /context:

• Expected: sonnet[1m] with 1M context window
• Actual: sonnet with 200k context window

---

Expected Behavior

When resuming a project, Claude Code should use the default model specified in settings.json, which is sonnet[1m] (1M context).

---

Actual Behavior

Claude Code starts with base sonnet (200k context) instead, immediately overflowing the context window:

Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   claude-sonnet-4-5-20250929 · 239k/200k tokens (120%)

After manually running /model sonnet[1m], context usage drops to 25%:

Context Usage
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   claude-sonnet-4-5-20250929[1m] · 246k/1000k tokens (25%)

---

Configuration Evidence

User's ~/.claude/settings.json:

{
  "model": "sonnet[1m]",
  "alwaysThinkingEnabled": true,
  ...
}

Startup behavior:

• /context at startup: 239k/200k tokens (120%) ← Wrong model
• /model sonnet[1m] manually applied: 246k/1000k tokens (25%) ← Correct model

---

Impact

This bug is particularly problematic for projects with heavy MCP tool usage:

• MCP tools alone: 154.2k tokens (77.1% of 200k window)
• Total at startup: 239k tokens (120% over capacity)
• Result: Immediate context overflow on resume

For projects requiring large context (like multi-application MCP integrations), this bug makes resume unusable without manual intervention.

---

Workaround

After running resume, immediately run:

/model sonnet[1m]

---

Possible Causes

1. Resume command override: resume might hardcode base sonnet regardless of config
2. Session state persistence: Previous session model might be restored instead of config default
3. Config read order: resume might read config before settings.json is parsed

---

Additional Context

• This was discovered while resuming a Microsoft 365 MCP integration project
• Project has 154k tokens of MCP tools (automem, dbt-mcp, orchestra-mcp, snowflake-mcp, aws-api, github, slack, clickup, etc.)
• Without 1M context, the project is unusable due to immediate overflow

---

Request: Please ensure resume command respects the default model setting in settings.json, or provide a way to specify model when calling resume.

Thank you! 🙏