Token Usage Regression in /context Command Between v1.0.86 and v1.0.88

Bug Report: Token Usage Regression in /context Command Between v1.0.86 and v1.0.88

Description
After upgrading from v1.0.86 to v1.0.88, there is a clear regression in token consumption when using the /context slash command.

• In v1.0.86, the token usage was significantly lower and more efficient.
• In v1.0.88, the same command now consumes substantially more tokens, which increases costs and makes the tool less practical for longer sessions.

This behaviour suggests a change in how context is being built, sent, or measured internally between these versions. It looks like something in the handling of the request payload or token accounting has changed, leading to inflated usage.

Environment

• Platform: darwin
• Terminal: VSCode
• Version: 1.0.88

Steps to Reproduce

1. Run /context with the same input on v1.0.86 → observe relatively low token usage.
2. Run /context with identical input on v1.0.88 → observe significantly higher token usage.

Expected Behaviour
Token usage for identical context operations should remain stable between versions. Any increase should be minimal and explained by new features or changes, not a regression.

Impact

• Higher operational costs.
• Reduced usability for longer context windows.
• Unexpected behaviour when upgrading from 1.0.86 → 1.0.88.

Notes
It would help to investigate whether the difference comes from:

• Changes in how context is serialised.
• Different system messages or metadata being injected.
• A bug in token counting logic introduced in v1.0.88.