In recent years, demand for mental health care services has outpaced provider availability. This problem is exacerbated in rural and historically underserved communities where specialist care providers may not exist, and patients regularly face lengthy waitlists and significant gaps in care when they are at their most vulnerable. Current AI-powered digital mental health platforms have attempted to fill this void by providing more accessible assessments and longitudinal monitoring that traditional clinical visits cannot achieve. Standardized instruments were originally designed for clinical contexts where providers can clarify and follow up on the nuances of mental health symptomatology. In the absence of this clinical dialogue, patients, especially children and young adults, may misinterpret app-based questions or underreport risk of self-harm, leading to erroneous diagnostic conclusions or lack of intervention. It is essential to develop high-fidelity app-based tools that close gaps in assessment, care, and assist in identifying patients in crisis. Journaling offers a promising alternative to gain deep insights into patient mental status. Increasingly common in therapeutic settings, especially among adolescents, journaling not only provides considerable insight for clinicians, but also benefits patients themselves. Therapeutic journaling has been readily translated to digital mental health applications and platforms, providing a convenient and accessible format for patients to process emotions and document symptoms in real-time — while also capturing experiences patients may forget, deem ‘normal’, or withhold in clinical settings. We have developed the Mirror Journal application to provide such a space, connecting users with an accessible platform for mood ratings and reflection, as well as guided and unguided journaling in a variety of modalities (e.g., video journals, drawing, writing), and safeguards that can direct them towards support when in need. By passively extracting meaningful features of these journal entries related to mental health symptoms, we can provide clinicians with longitudinal metrics that they can use to develop a holistic and reliable picture of patient well-being between clinical visits. In this presentation, we will describe how we use AI agents to deliver clinically-informed evaluations of journal entries by integrating Large Language Models (LLMs) with traditional mood screeners and structured assessments. Our approach simulates the administration of clinical assessments to journaling data, enabling us to extract features of symptom presence and severity. By using validated clinical instruments as the basis for evaluation, features measured with this approach are highly interpretable in a clinical setting. This increases the accessibility of assessment, and provides the opportunity for naturalistic longitudinal symptom evaluation rather than standard of care snapshot in-clinic evaluations. This approach provides the opportunity for evaluation in the lead-up towards, during, or following a crisis, both enhancing our clinical understanding of symptoms during these events, and potentially opening the door for early and preventative intervention. Additionally, this approach can direct users to relevant curated resources based on their current mood and symptoms, such as structured journaling prompts and crisis hotlines, increasing safety without requiring access to clinical care.