jarrodwatts/claude-hud v0.0.10

on GitHub

Added

• Configurable HUD color overrides, including named presets, 256-color indices, and hex values.
• display.customLine support for a short custom phrase in the HUD.
• New opt-in display toggles for session name, combined context mode (display.contextValue: "both"), Claude Code version, and approximate system RAM usage in expanded layout.

Changed

• Setup and plugin detection now better handle CLAUDE_CONFIG_DIR, Windows shell quoting, and Bun --env-file installs without inheriting project environment files.
• Usage display now prefers Claude Code stdin rate_limits data when available, still falls back to the existing OAuth/cache path, and presents weekly-only/free-user usage more cleanly.
• Context percentages and token displays now follow Claude Code's reported context window size, including newer 1M-context sessions, with a lower fallback autocompact estimate that better matches /context.
• Usage text output now keeps the last successful values visible while syncing, shows the 7-day reset countdown when applicable, and clarifies that standard proxy environment variables are the supported way to route Anthropic traffic.
• Progress bars and expanded-layout output now adapt more cleanly to narrow terminal widths.

Fixed

• Setup is more reliable in sessions that previously failed to surface the HUD until Claude Code restarted, and plugin command discovery no longer fails with unknown-skill errors after install.
• Usage handling is more resilient under OAuth token refreshes, proxy tunnels, explicit TLS overrides, zero-byte lock files, stale-cache recovery, and rate-limit edge cases that previously caused repeated 429 or syncing failures.
• Account-scoped credential lookup and plugin selection are more reliable for multi-account setups and multiple installed plugin versions.
• Expanded-layout rendering now preserves speed, duration, extra labels, and weekly-only usage output correctly.
• Tool execution no longer scrolls the terminal to the top, and transcript reparsing now avoids repeatedly caching partial parse results on large histories.