The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for processing untrusted data through the ingestion of clinician-authored intervention records.
Ingestion points: The skill reads local JSON files matching specific templates (e.g., source_fact_manifest.json, clinician_authored_intervention.json) via the read_json and load_package functions in scripts/_common.py.
Boundary markers: Mandatory draft notices ("DRAFT — NOT MEDICAL ADVICE") and clinician sign-off attestations are programmatically enforced in scripts/_common.py and required by the instructions in SKILL.md.
Capability inventory: The skill's capabilities are restricted to local filesystem operations (atomic JSON writes and directory creation) using the Python standard library; it lacks network access, subprocess execution, or dynamic code evaluation (eval/exec) capabilities.
Sanitization: Strict structural validation is performed by the _check_bounds function in scripts/_common.py, which limits object nesting depth to 20, total JSON nodes to 20,000, and string character counts to 8,000, effectively mitigating resource exhaustion and complex injection attempts.
[EXTERNAL_DOWNLOADS]: The documentation references official government and regulatory domains (e.g., hhs.gov, fda.gov, nice.org.uk) for clinical guidance and safety program requirements.
These references target well-known health and regulatory organizations for provenance documentation and do not involve automated downloads or code execution.

treatment-plans