Create Skill for medical assitante triage - #230
Create Skill for medical assitante triage#230Michael Pereira (michaelfp) wants to merge 13 commits into
Conversation
There was a problem hiding this comment.
Pull request overview
Adds a new medical-triage-assistant submission to the skills gallery, providing an agent-facing triage workflow (limits-first), supporting reference protocols, and a standardized pre-report template intended for review by licensed clinicians.
Changes:
- Introduces the core
SKILL.mdwith strict non-diagnostic boundaries, red-flag escalation behavior, document review rules, and a 5-level urgency classification workflow. - Adds agent reference materials for the triage protocol and fallback lab reference values.
- Adds a human-facing
README.md,metadata.jsongallery metadata, and a fixed pre-report template with a mandatory disclaimer.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.
Show a summary per file
| File | Description |
|---|---|
| submissions/medical-triage-assistant/SKILL.md | Defines the agent’s runtime workflow, guardrails, escalation rules, and required disclaimer usage. |
| submissions/medical-triage-assistant/references/triage_protocol.md | Provides structured interview steps, red-flag criteria, and the 5-level urgency model used by the skill. |
| submissions/medical-triage-assistant/references/lab_reference_values.md | Supplies fallback adult reference ranges and required flagging language when lab reports omit ranges. |
| submissions/medical-triage-assistant/assets/pre_report_template.md | Adds the standardized pre-report template including the mandatory non-diagnostic disclaimer. |
| submissions/medical-triage-assistant/README.md | Human-facing overview, intended deployment notes, and folder structure explanation. |
| submissions/medical-triage-assistant/metadata.json | Registers the submission in the gallery with platforms/tags/author metadata. |
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.
Comments suppressed due to low confidence (1)
submissions/medical-triage-assistant/references/triage_protocol.md:55
triage_protocol.mdreferenceslab_reference_values.mdwithout thereferences/path, while the rest of the submission (e.g.,SKILL.md) usesreferences/lab_reference_values.md. If the agent/tooling resolves file references from the submission root, this can break the lookup and make the fallback table unreachable.
- **The laboratory's own printed reference range always prevails.** Use `lab_reference_values.md` only when the report does not show a range, and state that a general reference was used.
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.
Suppressed comments (1)
submissions/medical-triage-assistant/references/triage_protocol.md:55
triage_protocol.mdreferenceslab_reference_values.mdwithout thereferences/prefix, but the actual file lives atreferences/lab_reference_values.md(as referenced elsewhere in this submission). This can cause the agent to look for a non-existent path when it needs the fallback ranges.
- **The laboratory's own printed reference range always prevails.** Use `lab_reference_values.md` only when the report does not show a range, and state that a general reference was used.
|
Thanks for this, Michael Pereira (@michaelfp) — it's carefully written, and the "never diagnose / every output needs physician review" guardrails are clearly deliberate. My hesitation is about the deployment framing rather than those guardrails. Because the skill ingests clinical data (lab reports, symptoms) and produces an urgency classification plus flagged out-of-range values, its function looks like clinical decision support / SaMD — and that classification follows intended use, not a "not a medical device" label in the text. The gallery's runtimes (Copilot Studio / Cowork / Scout) aren't certified medical environments: no BAA/HIPAA-eligibility by default, no MDR/SaMD or ANVISA clearance, no clinical quality system. So the current README, which points readers to "production use" after a compliance review, reads as more production-ready than the platform can actually support. Rather than block the concept, I'd like to keep it clearly framed as a demo so nobody deploys it against real patients unaware. Could you add explicit caveats in both
Once those caveats are in, I'm happy to take another look. Thanks again for the contribution! |
|
adilei I have updated the Skill.MD, Readme.md and Metadata.json with a warning to only use as demonstration |
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.
Suppressed comments (1)
submissions/medical-triage-assistant/SKILL.md:8
SKILL.mdshould stay agent-facing; this long blockquote is largely adoption/regulatory context (and it references Scout even though the submission targets only Copilot Studio/Cowork inmetadata.json). Per.github/copilot-instructions.md:114-156, move the detailed regulatory/platform notes toREADME.mdand keep only a short runtime safety reminder here.
> **DEMONSTRATION ONLY — NOT FOR CLINICAL USE AS AUTHORED.** This skill is illustrative content for learning and prototyping. Triage/urgency classification of real patients may constitute a regulated medical-device function (SaMD / clinical decision support) under local law, and deploying it may be prohibited without the appropriate regulatory clearance (e.g., FDA, EU MDR, ANVISA), a lawful data-processing/consent basis, and clinical governance. The host platforms (Copilot Studio, Cowork, Scout) are not certified medical devices or regulated healthcare environments — no BAA/HIPAA-eligibility, MDR/SaMD or ANVISA clearance by default. Do not use with real patients or real patient data as authored.
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.
Suppressed comments (2)
submissions/medical-triage-assistant/SKILL.md:8
- This line references Scout as a host platform, but this submission’s
metadata.jsondeclares onlyCopilot StudioandCowork. Keeping platform mentions aligned avoids confusing adopters about where the skill is intended to run.
> **DEMONSTRATION ONLY — NOT FOR CLINICAL USE AS AUTHORED.** This skill is illustrative content for learning and prototyping. Triage/urgency classification of real patients may constitute a regulated medical-device function (SaMD / clinical decision support) under local law, and deploying it may be prohibited without the appropriate regulatory clearance (e.g., FDA, EU MDR, ANVISA), a lawful data-processing/consent basis, and clinical governance. The host platforms (Copilot Studio, Cowork, Scout) are not certified medical devices or regulated healthcare environments — no BAA/HIPAA-eligibility, MDR/SaMD or ANVISA clearance by default. Do not use with real patients or real patient data as authored.
submissions/medical-triage-assistant/SKILL.md:17
- Requiring all outputs (including triage summaries and document analyses) to end with the disclaimer from
assets/pre_report_template.mdis internally inconsistent, because that disclaimer is explicitly written for a pre-report (it starts with “This pre-report…”). This will cause non-pre-report messages to include inaccurate boilerplate. Consider limiting the mandatory template disclaimer to the pre-report output only (or introduce a separate generic disclaimer asset for other message types).
3. **Every output is preliminary and non-binding.** Every triage summary, document analysis, and pre-report MUST end with the fixed disclaimer from `assets/pre_report_template.md`, stating that the content requires evaluation by a licensed physician and does not replace a medical consultation.
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.
Suppressed comments (2)
submissions/medical-triage-assistant/SKILL.md:8
SKILL.mdshould stay strictly agent-facing; the long regulatory/“host platforms are not certified…” paragraph reads like human-facing deployment/adoption guidance (and it’s already covered inREADME.md). Per repo submission guidance, keep the runtime file focused on what the agent must do, and move/keep human-facing regulatory notes inREADME.mdinstead.
> **DEMONSTRATION ONLY — NOT FOR CLINICAL USE AS AUTHORED.** This skill is illustrative content for learning and prototyping. Triage/urgency classification of real patients may constitute a regulated medical-device function (SaMD / clinical decision support) under local law, and deploying it may be prohibited without the appropriate regulatory clearance (e.g., FDA, EU MDR, ANVISA), a lawful data-processing/consent basis, and clinical governance. The host platforms (Copilot Studio, Cowork, Scout) are not certified medical devices or regulated healthcare environments — no BAA/HIPAA-eligibility, MDR/SaMD or ANVISA clearance by default. Do not use with real patients or real patient data as authored.
submissions/medical-triage-assistant/README.md:4
- This blockquote references a “Regulatory status” section that doesn’t exist in this README, which makes the guidance misleading for adopters.
> This skill is illustrative content: it shows how a triage-support agent *could* be structured. It has no regulatory clearance of any kind and must not be deployed against real patients or real patient data as authored. See "Regulatory status" below before doing anything beyond demos with synthetic data.
This as skill who can support nurse and doctor to get triage from a patient to provide better information on symptons