v2.5.0

Added

• Profile effort: ModeConfig.effort on each mode entry in model-profiles.json. Switching modes applies model and thinking level (off|minimal|low|medium|high|xhigh). Explicit effort wins over a :suffix on the model string; bare "provider/model:high" still works. If neither is set, defaults to medium.

Install: pi install npm:@georgedong32/permission-modes@2.5.0

v2.4.2 — near-fullscreen plan ready

Changed

• Near-fullscreen plan ready dialog (92% height) with sticky bottom options

Fixed

• Plan mode no longer blocks read-only commands that append 2>&1

Install

pi install npm:@georgedong32/permission-modes@2.4.2

v2.4.1 — plan ready bottom-anchor + Stay dismiss

Fixed

• Plan ready dialog no longer floats mid-screen (bottom-anchored overlay, max 50% height)
• Stay in plan mode / Esc dismiss once without agent_end immediately reopening the dialog

Install

pi install npm:@georgedong32/permission-modes@2.4.1

v2.4.0 — CC-aligned auto classifier

Highlights

• CC-aligned auto security: deny-and-continue, fail-closed classifier, denial tracking
• Classifier tool stage (default): forced classify_result tool call, API thinking disabled
• Model ref suffix: provider/model@tool overrides classifier.stage
• Dangerous allow stripping in auto mode, injection probes, tighter auto tiers

Install

pi install npm:@georgedong32/permission-modes@2.4.0

Full changelog: CHANGELOG.md