Inspiration

When clinical, healthcare, and enterprise teams collaborate inside Jira, sensitive Protected Health Information (PHI), acute crisis language, and adversarial prompt-override attempts frequently slip into tickets and comments unchecked. As a Licensed Professional Counselor with 28 years in mental health, a Ph.D. in Psychology, and an independent AI safety researcher, I built Orchide Clinical Guard to bring clinical-grade behavioral governance directly into everyday Jira workflows.

What it does

Orchide Clinical Guard embeds the 31-Law Clinical & Behavioral Safety Matrix inside Jira issues via Atlassian Forge:

  • Zero-Click Live Issue Screening (@forge/api): Automatically extracts and evaluates Jira issue summary, Atlassian Document Format description, and comments the moment a ticket is viewed, created, or updated (avi:jira:created:issue, avi:jira:updated:issue).
  • Multi-Tier Risk Classification:
    • CLEARED (LOW): No crisis indicators, coercion, or unredacted PHI detected.
    • PHI REVIEW (MODERATE): Flags potential HIPAA/PHI identifiers (SSN, DOB, medical record, diagnosis) for redaction verification.
    • CONTAINMENT TRIGGERED (HIGH): Blocks boundary-erosion and adversarial override patterns (ignore previous, bypass, coercion).
    • INTERVENTION REQUIRED (CRITICAL): Escalates acute self-harm or crisis indicators immediately.
  • Interactive Sandbox Audit Probe: Lets security and compliance reviewers test edge-case prompts directly inside the Jira issue panel without modifying production ticket fields.
  • Immutable Firestore Audit Trail: Streams structured, append-only telemetry records (issueKey, riskLevel, statusBadge, source, findings, timestamp) to Google Cloud Firestore (orchide_jira_audits) for compliance auditing.

How we built it

  • Atlassian Forge (nodejs22.x & UI Kit 2 @forge/react): Built a native jira:issuePanel paired with background issue lifecycle triggers (avi:jira:created:issue and avi:jira:updated:issue) using read:jira-work permissions.
  • 31-Law Guardrail Resolver: Engineered a deterministic clinical and behavioral screening engine in JavaScript that parses Atlassian Document Format (ADF) nodes and evaluates text against clinical safety, HIPAA PHI, and adversarial containment rules.
  • Google Cloud Firestore Telemetry: Connected the Forge resolver to Google Cloud Firestore via REST API (firestore.googleapis.com) to persist real-time, timestamped audit records.

Challenges we ran into

Parsing nested Atlassian Document Format (ADF) JSON structures across live Jira descriptions and comment threads—and ensuring both real-time UI panel scans and background issue triggers wrote clean, verified audit records to Firestore—required careful schema normalization and Firestore security rule configuration.

Accomplishments that we're proud of

  • Achieving zero-click live screening on Jira tickets with immediate visual risk badges (PHI REVIEW, CONTAINMENT TRIGGERED, INTERVENTION REQUIRED).
  • Delivering an end-to-end compliance pipeline from Jira issue creation to immutable Firestore audit logging in under 3 minutes of setup.

What's next for Orchide Clinical Guard

  • Expanding automated Jira comment warnings and Slack/Confluence cross-product alerts when HIGH or CRITICAL containment events occur.
  • Adding configurable organization-level policy thresholds and one-click PHI redaction suggestions directly inside Jira.

Built With

Share this project:

Updates

Submission history