feat: Add clinical-note-red-flag-scanner template#296
Open
Vilsee wants to merge 3 commits into
Open
Conversation
Adds a clinical documentation compliance analysis template that accepts a clinical note as text and returns a structured, severity-ranked JSON list of documentation red flags (missing consent, drug interactions, incomplete vitals, ambiguous dosing, etc.) for clinician and compliance reviewer triage. - Single-flow template (API Request -> LLM Analyse Note -> Variables -> API Response) - 10 red-flag categories with 4 severity levels (CRITICAL/HIGH/MEDIUM/LOW) - Uses Gemini (gemini-3.1-flash-lite-preview) for analysis - Structured JSON output with per-flag reasoning, location, and remediation - Includes comprehensive agent.md, README with tested example I/O, and safety constitution
Contributor
|
Warning Review limit reached
Next review available in: 37 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Repository UI (base), Organization UI (inherited) Review profile: ASSERTIVE Plan: Pro Plus Run ID: 📒 Files selected for processing (5)
WalkthroughChangesAdds a complete Clinical Note Red-Flag Scanner template with Gemini analysis, structured JSON output rules, a four-node execution flow, safety guardrails, documentation, model configuration, and Lamatic packaging metadata. Clinical Note Red-Flag Scanner
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
:robot_face: AgentKit Structural ValidationNew Contributions Detected
Check Results
🎉 All checks passed! This contribution follows the AgentKit structure. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 13
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@kits/clinical-note-red-flag-scanner/agent.md`:
- Around line 15-19: Revise the standards-alignment section and guardrail claims
in kits/clinical-note-red-flag-scanner/agent.md at lines 15-19 to state only
that the template may support documentation-review workflows, and explicitly
disclaim regulatory compliance certification. Apply the same narrowed wording
and disclaimer to the user-facing compliance statement in
kits/clinical-note-red-flag-scanner/README.md at lines 12-14.
- Around line 29-33: Update the input-note guidance in the agent documentation
to remove the “any length” claim. Define and enforce a maximum supported note
size or implement chunking before scanning, document the actual limit and
behavior for oversized notes, and keep the required clinicalNote input contract
unchanged.
In `@kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts`:
- Around line 159-163: Update the finalText mapping in the Variables node to
reference only LLMNode_453.output.generatedResponse, removing the trailing
literal “{” so the complete model response remains valid JSON.
- Around line 110-115: Update the API Request trigger configuration for nodeName
"API Request" so advance_schema requires clinicalNote to be a non-empty,
non-whitespace string. If the schema cannot express trimming, add an explicit
guard immediately before the LLM node that rejects empty or whitespace-only
clinicalNote values before scanning.
In `@kits/clinical-note-red-flag-scanner/lamatic.config.ts`:
- Around line 9-12: Add the deployment environment key to the mandatory step
configuration identified by id "clinical-note-red-flag-scanner", using the step
ID expected by runtime flow resolution so it can read the deployed flow ID from
process.env. Preserve the existing id and type values.
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.md`:
- Line 13: Update the REGULATORY guidance in the clinical-note red-flag scanner
prompt to require jurisdiction and care-setting context before emitting
regulatory findings. If those inputs are unavailable, restrict findings to
requirements explicitly identified by the caller or documented in the note; do
not make jurisdiction-specific claims from clinicalNote alone.
- Line 5: Update the DRUG_INTERACTION category definition in the system prompt
to avoid claiming definitive or exhaustive detection without an authoritative
interaction source; either specify a versioned, clinically verified interaction
database/tool, or restrict findings to interactions explicitly documented in the
note and label screening as non-exhaustive.
- Around line 23-27: Update the recommendation field instructions and the
prohibition near the medical-advice guidance to require documentation-only
remediation actions. Explicitly prohibit treatment choices, dosing regimens,
diagnoses, and triage or emergency-care advice, and revise the README example to
remove nitroglycerin and emergency-care instructions.
- Around line 15-30: Update the JSON response contract in the system prompt to
explicitly require that flagCount equals flags.length, flag IDs are unique and
sequential starting at FLAG-001, and category and severity values use only the
defined valid codes. Require responses to satisfy these invariants in addition
to the existing structure.
- Around line 37-42: Update the evidence guidance near the flag-generation rules
so `detail` and `location` use redacted or paraphrased evidence rather than
reproducing exact patient text. Explicitly prohibit names, MRNs, DOBs, contact
information, and other identifiers while preserving enough non-identifying
context to support each flag.
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_user_1.md`:
- Around line 1-2: Update the clinicalNote interpolation in the clinical-note
scanner prompt to prevent raw PHI from being sent to Gemini. Require a
pre-redacted or de-identified note, or enforce a documented compliant
zero-retention provider configuration before forwarding the content; do not rely
on prompt-level confidentiality instructions.
- Around line 1-2: Update the clinical-note prompt content around
triggerNode_1.output.clinicalNote to clearly delimit the note as untrusted data
and explicitly instruct the model to ignore any commands or instructions
contained within it, while preserving the required JSON, confidentiality, and
safety behavior.
In `@kits/clinical-note-red-flag-scanner/README.md`:
- Around line 1-8: Move the “# Clinical Note Red-Flag Scanner” H1 above the
deploy badge HTML so it is the README’s first content and satisfies MD041. Keep
the existing badge markup unchanged below the heading.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI (base), Organization UI (inherited)
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 83786c0b-8632-4fb4-8f70-44be325cb8c6
📒 Files selected for processing (9)
kits/clinical-note-red-flag-scanner/.gitignorekits/clinical-note-red-flag-scanner/README.mdkits/clinical-note-red-flag-scanner/agent.mdkits/clinical-note-red-flag-scanner/constitutions/default.mdkits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.tskits/clinical-note-red-flag-scanner/lamatic.config.tskits/clinical-note-red-flag-scanner/model-configs/clinical-note-red-flag-scanner_llmnode-453_generative-model-name.tskits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.mdkits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_user_1.md
Comment on lines
+23
to
+27
| "severity": "CRITICAL | HIGH | MEDIUM | LOW", | ||
| "title": "Brief flag title", | ||
| "detail": "Specific explanation of what is missing or problematic", | ||
| "location": "Where in the note this issue was found (or 'ABSENT' if the issue is something missing entirely)", | ||
| "recommendation": "Specific action to remediate this flag" |
Contributor
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | 🏗️ Heavy lift
Mission-critical: keep remediation guidance documentation-only.
The recommendation field asks for specific actions while Line 41 prohibits medical advice; the README example already supplies a nitroglycerin regimen and emergency-care instructions. Require documentation actions only, and explicitly prohibit treatment choices, dosing regimens, diagnoses, and triage advice.
Also applies to: 41-42
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.md`
around lines 23 - 27, Update the recommendation field instructions and the
prohibition near the medical-advice guidance to require documentation-only
remediation actions. Explicitly prohibit treatment choices, dosing regimens,
diagnoses, and triage or emergency-care advice, and revise the README example to
remove nitroglycerin and emergency-care instructions.
Comment on lines
+37
to
+42
| - Be specific. Cite the exact text or absence that triggered each flag. | ||
| - Do not fabricate flags. If the note is well-documented, return fewer flags. | ||
| - Sort flags by severity (CRITICAL first, LOW last). | ||
| - If no flags are found, return an empty flags array with a positive summary. | ||
| - Do not provide medical advice. You are reviewing documentation completeness, not making clinical judgments. | ||
| - Treat all patient data as confidential. Do not repeat PII unnecessarily in your output. No newline at end of file |
Contributor
There was a problem hiding this comment.
🔒 Security & Privacy | 🟠 Major | ⚡ Quick win
Redact PII from evidence fields.
“Cite the exact text” conflicts with the PII-minimization rule. Require redacted/paraphrased evidence and prohibit repeating names, MRNs, DOBs, contact details, or other identifiers in detail and location.
🧰 Tools
🪛 markdownlint-cli2 (0.23.0)
[warning] 42-42: Files should end with a single newline character
(MD047, single-trailing-newline)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.md`
around lines 37 - 42, Update the evidence guidance near the flag-generation
rules so `detail` and `location` use redacted or paraphrased evidence rather
than reproducing exact patient text. Explicitly prohibit names, MRNs, DOBs,
contact information, and other identifiers while preserving enough
non-identifying context to support each flag.
- Fix trailing brace in variablesNode mapping - Add envKey to lamatic.config.ts - Add minLength validation to flow advance_schema - Narrow compliance claims and add data privacy warnings in README and agent.md - Add token limit to agent.md - Restrict system prompt DRUG_INTERACTION and REGULATORY claims - Enforce documentation-only recommendations and PII redaction - Add prompt injection defense to user prompt - Enforce strict JSON invariants - Fix markdownlint warnings
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
kits/clinical-note-red-flag-scanner/README.md (3)
67-69: 🔒 Security & Privacy | 🟡 Minor | ⚡ Quick winKeep the public example fully synthetic and non-identifying.
The sample uses patient-like clinical data and demonstrates reproducing MRN/DOB fields, contradicting the redaction requirements. Replace it with clearly synthetic, de-identified content and avoid showing identifier reproduction.
Also applies to: 87-89
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@kits/clinical-note-red-flag-scanner/README.md` around lines 67 - 69, Replace the clinicalNote example in the README with clearly synthetic, non-identifying content that does not resemble real patient data or include names, demographics, dates of birth, MRNs, or other identifiers. Update the related example at the referenced reproduction section as well, removing any demonstration of identifier reproduction while preserving the red-flag scanner usage context.
105-116: 🔒 Security & Privacy | 🟠 MajorRemove treatment and triage advice from the example output.
The sample recommends medication dosing and emergency-care instructions, while the system contract permits documentation-only remediation. Revise the example to use actions such as documenting missing fields, not treatment or return-to-ED guidance.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@kits/clinical-note-red-flag-scanner/README.md` around lines 105 - 116, Update the example output entries FLAG-003 and FLAG-004 in the README so their recommendation text only instructs users to document missing clinical fields or plan details. Remove medication dosing guidance and emergency-care or return-to-ED instructions while preserving the existing finding context.
73-80: 🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick winAlign the documented response shape with the runtime contract.
This section says
resultcontains a JSON object and shows the report without the outerresultfield, while the flow and agent documentation defineresultas a JSON string. Show the actual wrapper and serialization, or change the implementation and all documentation consistently.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@kits/clinical-note-red-flag-scanner/README.md` around lines 73 - 80, Update the “Example Output” section to match the runtime contract: show the outer result field and represent its value as the serialized JSON string returned by the flow. Keep the documented summary and flagCount payload consistent with the implementation and related flow/agent documentation.
♻️ Duplicate comments (1)
kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts (1)
114-114: 🩺 Stability & Availability | 🟠 MajorEnforce the documented maximum note size.
The runtime accepts arbitrarily large notes while the documentation promises a 10,000-token limit. Add bounded validation or chunking before the single LLM request.
kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts#L114-L114: add a maximum-size guard or chunking path to the trigger contract.kits/clinical-note-red-flag-scanner/agent.md#L33-L33: document the enforced limit and oversized-input behavior.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts` at line 114, Enforce the documented 10,000-token maximum in the clinical note trigger contract around the advance_schema definition by adding bounded validation or chunking before the single LLM request. In kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts:114, reject or appropriately process oversized clinicalNote values; in kits/clinical-note-red-flag-scanner/agent.md:33, document the enforced limit and the behavior for oversized input.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.md`:
- Line 13: Update the REGULATORY guidance in the system prompt to require
jurisdiction and care setting before making jurisdiction-specific findings. When
those inputs are unavailable, limit findings to requirements explicitly
identified by the caller or clinical note and label them non-jurisdictional.
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_user_1.md`:
- Line 4: Update the prompt input around clinicalNote so raw clinical notes
cannot be forwarded directly to Gemini without a data-control boundary. Require
and validate pre-redacted/de-identified content, or verify a compliant
zero-retention provider configuration before executing the node; otherwise block
execution rather than interpolating the complete note.
---
Outside diff comments:
In `@kits/clinical-note-red-flag-scanner/README.md`:
- Around line 67-69: Replace the clinicalNote example in the README with clearly
synthetic, non-identifying content that does not resemble real patient data or
include names, demographics, dates of birth, MRNs, or other identifiers. Update
the related example at the referenced reproduction section as well, removing any
demonstration of identifier reproduction while preserving the red-flag scanner
usage context.
- Around line 105-116: Update the example output entries FLAG-003 and FLAG-004
in the README so their recommendation text only instructs users to document
missing clinical fields or plan details. Remove medication dosing guidance and
emergency-care or return-to-ED instructions while preserving the existing
finding context.
- Around line 73-80: Update the “Example Output” section to match the runtime
contract: show the outer result field and represent its value as the serialized
JSON string returned by the flow. Keep the documented summary and flagCount
payload consistent with the implementation and related flow/agent documentation.
---
Duplicate comments:
In `@kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts`:
- Line 114: Enforce the documented 10,000-token maximum in the clinical note
trigger contract around the advance_schema definition by adding bounded
validation or chunking before the single LLM request. In
kits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.ts:114,
reject or appropriately process oversized clinicalNote values; in
kits/clinical-note-red-flag-scanner/agent.md:33, document the enforced limit and
the behavior for oversized input.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI (base), Organization UI (inherited)
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: fe830c49-c5c0-4bba-bc7e-97d970a0f9b7
📒 Files selected for processing (6)
kits/clinical-note-red-flag-scanner/README.mdkits/clinical-note-red-flag-scanner/agent.mdkits/clinical-note-red-flag-scanner/flows/clinical-note-red-flag-scanner.tskits/clinical-note-red-flag-scanner/lamatic.config.tskits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.mdkits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_user_1.md
| 7. **FOLLOW_UP** — Missing follow-up plan, discharge instructions, or continuity-of-care documentation | ||
| 8. **ASSESSMENT** — Missing or incomplete clinical assessment/diagnosis relative to documented symptoms | ||
| 9. **IDENTITY** — Missing or incomplete patient identification markers (MRN, DOB, provider signature/attestation) | ||
| 10. **REGULATORY** — Documentation gaps that may violate specific regulatory requirements (CMS, Joint Commission, state-level mandates, subject to applicable jurisdiction and care setting) |
Contributor
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major
Mission-critical: require jurisdiction before regulatory findings.
The flow accepts only clinicalNote, so the model cannot reliably determine which CMS, Joint Commission, or state requirements apply. If jurisdiction and care setting are unavailable, restrict findings to requirements explicitly identified by the caller or note and label them non-jurisdictional.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@kits/clinical-note-red-flag-scanner/prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.md`
at line 13, Update the REGULATORY guidance in the system prompt to require
jurisdiction and care setting before making jurisdiction-specific findings. When
those inputs are unavailable, limit findings to requirements explicitly
identified by the caller or clinical note and label them non-jurisdictional.
- Replace realistic PHI with synthetic placeholders in README example - Strip remaining clinical advice from FLAG-003 and FLAG-004 examples in README - Wrap JSON example output in stringified result field in README - Enforce 45000 character limit in flow advance_schema - Document enforced token limit and rejection behavior in agent.md - Strengthen REGULATORY caveat in system prompt (advisory only, caller provides jurisdiction) - Add hard PHI input contract to user prompt
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a Clinical Note Red-Flag Scanner template — a single-flow API pipeline that accepts a clinical note as text and returns a structured, severity-ranked JSON list of documentation red flags for clinician and compliance reviewer triage.
Problem Statement
Clinical documentation frequently ships with silent gaps — missing informed consent language, undocumented drug-interaction risk, incomplete vitals/history, or ambiguous dosing instructions — that create real patient-safety and regulatory exposure (aligned with EU AI Act Article 12 documentation/audit requirements, HIPAA documentation-integrity expectations, and CMS/Joint Commission accreditation standards).
The existing AgentKit registry has a
medical-assistantsymptom-checker chatbot, but nothing that performs structured compliance analysis of clinical documentation. This template fills that gap — it's not a conversational agent, it's a single-pass analysis tool that outputs machine-readable results for dashboards, alerting systems, and CDI (Clinical Documentation Improvement) workflows.What This Template Does
CONSENT,DRUG_INTERACTION,ALLERGY,VITALS,HISTORY,DOSING,FOLLOW_UP,ASSESSMENT,IDENTITY,REGULATORYCRITICAL/HIGH/MEDIUM/LOW) with per-flag reasoning, location references, and remediation recommendationsgemini-3.1-flash-lite-preview) for analysisFlow Architecture
4 nodes, single flow, template type — no UI, no env vars, no external dependencies beyond a Gemini API key.
Tested Example
Input — a 67M chest pain presentation with intentional documentation gaps:
{ "clinicalNote": "Patient: John D., 67M. Chief Complaint: Chest pain x 2 hours. History: HTN, DM2. Current Meds: Metformin 500mg BID, Lisinopril 10mg daily. Exam: BP 158/94, HR 88, RR 18. Lungs clear. Heart: regular rhythm, no murmurs. Assessment: Probable angina. Plan: Start aspirin 81mg daily, order stress test. Prescribed nitroglycerin PRN." }Output — 5 flags across 5 categories, severity-ranked:
Each flag includes
detail,location, andrecommendationfields. Full JSON output is in the README.Files Added
How It Was Built
.tscontains authentic runtime IDs and edge metadatakits/following thearticle-summarisertemplate structure exactlylamatic.config.ts,README.md,agent.md, and flow meta with comprehensive documentationChecklist
kits/article-summariser/exactlylamatic.config.ts— typetemplate, author, tags, deploy/github links.tsexported from Lamatic Studio (authentic runtime metadata)agent.md— overview, purpose, flow details (4 nodes), I/O schemas, guardrailsREADME.md— deploy button, category table, real tested example with full JSON outputconstitutions/default.md— safety guardrails and PII handlingregistry.jsonentries (verified all 79)feat: Add ...conventionclinical-note-red-flag-scanneris unique in the registryAdded
kits/clinical-note-red-flag-scanner/template kit:lamatic.config.ts: Lamatic template configuration (single step wired toCLINICAL_NOTE_RED_FLAG_SCANNER_FLOW_ID), deployment metadata, links, tags..gitignore: ignores.lamatic/,node_modules/,.env,.env.local.README.md: end-to-end usage docs, category list, severity/output schema, and example input/output JSON.agent.md: AgentKit agent documentation describing the API-invoked single-pass red-flag scanning pipeline.constitutions/default.md: baseline safety/privacy/adversarial-input guardrails (no PII retention/logging, refusal to prompt injection, avoid fabricating, professional tone).flows/clinical-note-red-flag-scanner.ts: single-flow API pipeline definition exporting{ meta, inputs, references, nodes, edges }.model-configs/clinical-note-red-flag-scanner_llmnode-453_generative-model-name.ts: Gemini model binding (gemini/gemini-3.1-flash-lite-preview) forLLMNode_453.prompts/clinical-note-red-flag-scanner_llmnode-453_system_0.md: strict system prompt enforcing raw JSON-only output, fixed schema, severity-ranked/uniqueFLAG-00xids, and documentation-only remediation.prompts/clinical-note-red-flag-scanner_llmnode-453_user_1.md: user prompt that injects runtime{{triggerNode_1.output.clinicalNote}}into a<clinical_note>block.Flow behavior (
clinical-note-red-flag-scanner.ts)—node types and how it works:triggerNode(“API Request”): GraphQL-triggered input acceptingclinicalNotewith minLength/pattern validation.dynamicNode(“Analyse Note”,LLMNode_453): calls the LLM using the referenced system/user prompt templates and the Gemini model config; producesgeneratedResponse.dynamicNode(“Variables”,variablesNode_197): mapsLLMNode_453.output.generatedResponseintofinalText.responseNode(“API Response”,graphqlResponseNode): returns JSON withresultset to{{variablesNode_197.output.finalText}}.defaultEdgebetween trigger → LLM → variables → response, plus aresponseEdgeto emit the trigger’s response.