To fix the problem, avoid logging patient_id (even partially) in clear text. The log line should use either a fully redacted placeholder (e.g., [REDACTED]) or a non-reversible surrogate that does not expose the identifier. This keeps operational semantics (indicating that patient data is being accessed) without revealing PHI.

The best minimal change is to update the print at line 84 so it no longer interpolates patient_id[:3]. For example, log a generic message such as 📂 Accessing patient data: [REDACTED PATIENT ID] or, if correlation is needed, use a separate correlation/access ID that does not derive directly from patient_id. Because we must not assume extra infrastructure, we will choose a simple redacted placeholder. No new imports are required.

Concretely, in packages/agent-mesh/examples/03-healthcare-hipaa/main.py, in the access_patient_data method, replace:

print(f"📂 Accessing patient data: {patient_id[:3]}***")

with a version that does not reference patient_id, e.g.:

print("📂 Accessing patient data: [REDACTED PATIENT ID]")

All other uses of patient_id in this method are for internal policy checks and auditing; since we are only allowed to change the highlighted snippet and we don’t see the implementation of _audit_phi_access, we will leave those as-is.

In general, to fix clear‑text logging of sensitive data, you either (1) avoid logging the sensitive values at all, or (2) log only safe, aggregated, or de‑identified derivatives (such as counts or categories) and ensure tainted objects are not passed into log formatting. Here, the problematic print statement is part of a demo/test harness that iterates over potentially tainted findings. Although it currently only logs severity, CodeQL treats the whole f object as tainted. The safest change that preserves functionality is to stop interpolating values derived from f and instead log a generic message that does not depend on the tainted findings contents, while still indicating that a finding was detected.

Concretely, in packages/agent-os/examples/healthcare-hipaa/main.py around lines 785–787, we will change:

• The use of f["severity"] and the conditional icon based on severity to a fixed, non‑tainted message (e.g., “Finding detected” with a generic icon).
• This preserves the visible behavior of iterating over findings and showing that there are as many lines as findings, but removes the data flow from the tainted f object into the log text.

No new imports or helper methods are required; we only modify the print logic within the loop.