Generates and verifies 21 CFR Part 11-style audit trails — who/what/when, electronic signatures, and tamper-evidence — for OpenMed pipelines in GxP and clinical-trial (GCP) settings. Use when the user runs OpenMed in a regulated/validated environment and needs an attributable, time-stamped, tamper-evident record of each processing action, electronic-signature manifestations, or computer-system-validation (CSV) evidence. Trigger keywords: 21 CFR Part 11, Part 11, audit trail, electronic signature, e-signature, GxP, GCP, GLP, GMP, CSV, computer system validation, data integrity, ALCOA, tamper-evident, contemporaneous. Pairs adjacent to OpenMed: maps directly onto OpenMed deidentify(audit=True) -> signed AuditReport with .sign(key)/.verify(key), whose repro_hash + HMAC give the tamper-evidence and attribution Part 11 expects. This is a compliance-enablement aid, not a validation certification.
74
91%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Passed
No findings from the security scan
In FDA-regulated GxP work (GCP clinical trials, GLP, GMP) any electronic record used to support a regulatory decision must meet 21 CFR Part 11: it has to be attributable (who), contemporaneous and time-stamped (when), describe what changed, be tamper-evident, and — where a signing event occurs — carry a controlled electronic signature. These map onto the ALCOA+ data-integrity expectations (Attributable, Legible, Contemporaneous, Original, Accurate, +Complete/Consistent/Enduring/Available).
OpenMed's deidentify(..., audit=True) already emits a deterministic,
PHI-free AuditReport that you can .sign() (HMAC-SHA256) and later
.verify(). That gives you the tamper-evidence and attribution primitives;
this skill wraps them in the who/when/what/e-signature envelope Part 11 wants.
This is a compliance-enablement aid. Part 11 compliance also requires validated systems (CSV), SOPs, and access controls that live outside any single library — a QA/validation lead signs off.
AuditReport with who/when/what + an
e-signature manifestation (meaning, signer, timestamp).| Part 11 expectation | 21 CFR cite | OpenMed mechanism |
|---|---|---|
| Tamper-evident, accurate copies | 11.10(b),(c) | AuditReport.to_json() + repro_hash over the canonical payload |
| Audit trail: what changed, when | 11.10(e) | AuditReport.spans (action per identifier), input_hash/deidentified_text_hash, openmed_version, manifest_hash |
| Operational/authority checks; attribution | 11.10(d),(g) | AuditSignature.key_id (signer/key identity) + your envelope's user id |
| Signature manifestation (name, date, meaning) | 11.50 | Your envelope fields signer, signed_at, meaning |
| Signature/record linking, non-repudiation | 11.70, 11.200 | HMAC-SHA256 over the canonical payload via .sign() / .verify() |
The HMAC binds the signature to that exact report content: any later edit to a
span, hash, or field changes repro_hash, so .verify() fails — that is the
tamper-evidence.
import openmed, json, datetime as dt
note = "Subject S-014 (DOB 1962-08-09) reported headache on 2024-05-01."
# 1) Produce the deterministic, PHI-free audit record for this processing step.
report = openmed.deidentify(note, policy="hipaa_safe_harbor", audit=True)
# 2) Sign it with a controlled release key (stored in a vault / HSM, never in code).
report.sign(b"<release-hmac-key>", key_id="omv-signer-2026")
# 3) Wrap in a Part 11 envelope: who / when / what / signature meaning.
trail = {
"record": report.to_dict(), # tamper-evident, no PHI
"who": "j.smith@sponsor.example", # authenticated user (your IdP)
"when": dt.datetime.now(dt.timezone.utc).isoformat(),
"what": "PHI de-identification of source narrative (study X, subject S-014)",
"signature_manifestation": { # 21 CFR 11.50
"signer_printed_name": "Jane Smith",
"meaning": "reviewed and approved",
"signed_at": dt.datetime.now(dt.timezone.utc).isoformat(),
},
"system": {"openmed_version": report.openmed_version,
"manifest_hash": report.manifest_hash},
}
with open("part11_trail.json", "w") as fh:
json.dump(trail, fh, indent=2, sort_keys=True)
# 4) Later — verify integrity (optionally bind to the exact source/output text).
ok = report.verify(b"<release-hmac-key>", original_text=note)
assert ok, "AUDIT TRAIL TAMPERED OR KEY MISMATCH"audit=True to get the deterministic record.key_id.
Never embed the key in source or the trail..verify(key, original_text=..., deidentified_text=...)
to confirm neither the record nor the bound texts changed.auditing-deidentification-runs
(deidentify(audit=True) → AuditReport) is the source of the signed,
PHI-free trail this skill envelopes.auditing-safe-harbor-checklist documents that the
18 identifier categories were handled — useful as a CSV artifact.enforcing-nophi-logging ensures the surrounding
application logs don't leak identifiers into the trail.checking-hipaa-compliance — Part 11 audit controls and the
HIPAA Security Rule audit-controls standard (164.312(b)) reinforce each other.AuditReport gives tamper-evidence and
attribution, but Part 11 also requires validated systems (CSV), SOPs, training,
and access controls you implement around it. Don't claim "Part 11 compliant"
from the audit object alone..sign() is a deliberate step; signature
is None until called. Empty/None keys are rejected.key_id, and never store the key with the trail.AuditReport is hash-and-offset only; don't
reintroduce identifiers in the what/who free-text fields.report.to_json().openmed/core/audit.py.80da98c
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.