jmandel

Produce a clean output/proposed-questionnaire.json or output/suggestions.json with stable linkIds, exact baseVersion, and the current typed proposal vocabulary. Use whenever you're about to suggest a structural change to the form.

Write product-facing replies and organize run outputs in the canonical Formtastic artifact layout. Always applies to every turn because reply.md is the visible chat response and all structured output depends on the run directory contract.

Create organized, copyright-safe research artifacts under output/research/. Always applies to evidence, guideline, citation, article, source-review, or literature turns.

Search Cochrane Reviews for systematic-review evidence and effect sizes. Use when the user wants the strongest evidence-base summary for a treatment, screening, or diagnostic question. Produces structured effect-size estimates with confidence intervals.

Find clinical practice guidelines from authoritative bodies (NICE, USPSTF, ACP, AHRQ, GIN). Use when the user asks "what does the guideline say" or wants to anchor a question or threshold to a recommended practice. Produces guideline-typed citations with the canonical URL of the recommendation.

Search PubMed via NCBI E-utilities to ground questionnaire items in primary literature. Use when the user asks for evidence, citations, or "what does the research say about X." Produces an organized research bundle with PubMed metadata, abstracts when available, and source-grounded notes.

Run schema + HL7 FHIR Validator + clinical-sanity validation on Questionnaire.json through Formtastic's host-managed validation endpoint. Use before publish, before a big propose-edit, or when the user asks "is this form valid?"

Rewrite summary.md so future turns don't need to re-read all of conversation.jsonl. Use periodically (every ~20 turns) or when the user explicitly asks "summarize what we've done so far."

Transcribe local audio or video recordings into self-contained HTML, JSON, and Markdown using Mistral/Voxtral diarization, then assign human speaker names after inspecting raw speaker IDs. Use when the user asks to transcribe call recordings, meeting recordings, MP4/M4A/MP3/WAV/WebM files, generate transcript HTML/Markdown, archive call logs, or map diarized speakers to names such as Josh, Dave, James, Sean, etc.

Create slide-image decks from markdown, outlines, or detailed presentation briefs, then assemble them into HTML and PowerPoint. Use this skill when the user wants generated presentation slides, a prompt set for slide images, a deck-level visual system, image-model slide generation through Codex-native tools or OpenRouter/Gemini scripts, or assembly/export of existing slide images.

Monitor and guide a Codex (or other AI coding agent) session running in a tmux pane. Use this skill when the user wants to babysit, monitor, supervise, or watch over a Codex/AI agent session, or when they mention tmux + codex together, or ask you to keep an eye on another AI agent working on implementation tasks. Also use when the user says things like "watch codex", "check on codex", "keep codex on track", or "babysit the implementation".

Read, load, and make sense of an Epic "EHI Export" — the thousands-of-TSV-tables-plus-schema-HTML dump a patient receives from an Epic-based health system (folders named EHITables, EHITables Schema, Rich Text, Media). Use this whenever you are handed such an export, or any folder of Epic Clarity/ Chronicles-style tables (PAT_ENC, ORDER_MED, HNO_INFO, ZC_/CLARITY_ master files, *_HX history tables, CSN/PAT_ID identifiers), or when a task involves loading one into SQLite, mapping its schema, finding the clinical notes/messages, or interpreting Epic's data-modeling conventions (CSN contacts, *_DATE_REAL dates, _C_NAME categories, base+supplement tables, denormalized _NAME companions). The foundational skill of the tower: ehi-deep-dives builds on it, and import-ehi feeds it a redacted raw/.

Build a deep dive from an Epic EHI export — the whole loop: ideate the view, explore the record for what supports it, abstract the data into a clean view model (structured + narrative, in one JSON), then build a bespoke static bun/React/D3 web app on that view model. Use this whenever making a clinical or financial deep dive / analysis app from such an export (after reading-epic-ehi-export has the data understood and in SQLite). A deep dive is the tool a clinician would want to understand a condition's course, or a billing analyst a claims history — NOT an infographic, dashboard, or slide deck; designed around what that expert needs, narrative-led, with the data interrogable through clean evidence (never raw rows). The "abstraction to power a deep dive" is step 3 here, not a separate skill (see reference/abstracting-to-the-view-model.md). Trigger on "build a deep dive", "make the <condition>/<billing> analysis app", "turn this into a deep dive".

Import an Epic EHI Export (the "Requested Record" zip) into a working tree and produce a REDACTED, same-shape subset that is safe to commit and publish. Goes zip → raw.unredacted/ (full PHI, gitignored) → raw/ (redacted, committable) via: discover identifier values from the structured columns, fuzzy-redact them everywhere (notes included), use agents to catch free-text-only secrets, and PROVE the result clean. Use this whenever you want to publish or share more than a hand-picked artifact — closer to the full record — without leaking PHI.

Help a patient request their complete Electronic Health Information (EHI) Export from their healthcare provider. Supports Epic and 70+ other certified EHR vendors. Explains what EHI is, why it matters, identifies the provider's EHR system, generates a vendor-specific appendix, guides through gathering details, finding forms, and producing a ready-to-submit PDF package.

Connect to a user's health records from their patient portal via SMART on FHIR. Use when the user wants to access or ask questions about their personal health data.

Testing toolkit for the FHIR Writing Clinical Notes specification at connectathons. Use when the user needs to test FHIR DocumentReference write operations, validate conformance with the Writing Clinical Notes spec, or participate in a FHIR connectathon for clinical notes. Includes templates, OAuth helpers, and automated test scenarios for both provider-authored and patient-asserted notes.

Convert a directory of clinical forms (PDFs, HTML pages, screenshots, scanned images, vendor templates, plain text — anything that looks like a patient intake form, PROM, follow-up form, or clinician checklist) into a directory of FHIR R4 Questionnaire JSON resources that follow HL7 SDC (Structured Data Capture) idioms. Use whenever the user has a folder of clinical forms they want represented as FHIR Questionnaires, or asks to "make these into FHIR", "convert this PROM", "turn these intake forms into Questionnaire resources", or similar — even if they don't say "SDC" or "FHIR" explicitly, if the source material is clinical questions / rating scales / intake packets and the desired output is structured form definitions, use this skill.

Build and host browser-only clinical "applets" that get rich SMART-on-FHIR data access and approved LLM access WITHOUT ever holding the OAuth token, network authority, or DOM. The model is one trusted wrapper that does the SMART launch once and safely runs many interchangeable, equally-sandboxed applets inside it — so a hospital CIO can enable a single wrapper and let clinicians experiment freely but safely. Use this skill whenever the user wants to: build a SMART on FHIR app or dashboard; run untrusted, third-party, or LLM-generated UI code that still needs patient data; sandbox/isolate a web app so it can't exfiltrate a token or phone home; create a clinician-facing applet platform or "app store"; add another applet to such a wrapper; or load applet code dynamically from a URL safely. Reach for this even if the user just says "let clinicians build their own tools safely," "I have an LLM that writes patient-facing widgets," or "I need a SMART app but I don't trust the app with the token."

Coordinate external Copilot CLI agents over FHIR spec work, using durable run state, isolated worktrees, bounded concurrency, foreground tick monitoring, retry recovery, and workflow-specific chunk contracts.

Turn meeting notes (or any notes/agenda/transcript) into a short, funny, on-the-nose AI theme song, and optionally play it into a video call as warm-up music. Use whenever the user wants a theme song, jingle, anthem, or hype track made from notes, an agenda, a sprint/standup/retro, or a project — or wants warm-up/hold music for a meeting — or mentions "minuet". Acts as a collaborative producer: brainstorm the song's concept (genre, mood, vocals, length, comedic angle) with the user, then drive the bundled `minuet` CLI to compose lyrics and generate audio via the Gemini API (Lyria) and pipe it through a virtual microphone. Trigger even if the user never says "minuet" — e.g. "make a song from our standup notes", "I want hype music for the team call", "turn these minutes into a jingle".

Set up Toolgate in existing projects so coding agents can run selected CLIs through a sandboxed, gated command surface.

Create Reci-pop structured recipe pages: decompose recipes into mise en place, step-local quantities, metric/unit-toggle data, scaling-friendly ingredient rows, covered-time process layout, storyboard prompts, generated image assets, and static recipe HTML. Use whenever the user asks to convert a recipe into a structured cooking page, model recipe process flow, generate recipe illustrations, prepare recipe JSON/assets for a static site, or reason about prep ordering and covered cooking time.