Current in-situ privacy solution approaches are inadequate in protecting sensitive information. They either require extra configuration effort or lack the ability to configure user desired privacy settings. Based on in-depth discussions during a design workshop, we propose PrivacyShield, a mobile system for providing subtle just-in-time privacy provisioning. PrivacyShield leverages the screen I/O device (screen digitizer) of smartphones to recognize gesture commands, even when the phone's screen is turned off. Based on gesture command inputs, various privacy-protection policies can be configured on-the-fly. We develop a novel stroke-based approach to address the challenges in segmenting and recognizing gesture command inputs, which helps the system in achieving good usability and performance. PrivacyShield also provides developers with APIs to enable just-in-time privacy provisioning in their applications. We have implemented an energy efficient PrivacyShield prototype on the Android platform, including smartphones with and without a low-power co-processor. Evaluation results show that our gesture segmentation algorithm is fast enough for real-time performance (introducing less than 200ms processing latency) and accurate (achieving an accuracy of 95% for single-character gestures and 89% for even three-character gestures). We also build a non-touch-screen-based just-in-time privacy provisioning prototype called the wrist gesture method. We compare the performance of the two prototypes by doing a 6-week field study with 12 participants and show why a simplistic solution falls short in providing privacy configurations. We also report the participants' perceptions and reactions after the field study.