WellAlly Health: a local-first personal health record built on Claude Code slash commands

The problem WellAlly Health targets: scattered records and cloud-only health apps

The README frames the problem as fragmentation. Lab results sit in a hospital portal, discharge summaries arrive as paper or photographs, imaging reports are separate again, and the patient's own memory fills the gaps. The stated alternative offered by commercial apps is uploading everything to a vendor cloud. WellAlly Health takes the opposite position: the README describes it as a local-first personal health record (PHR) and personal health information system (PHIS) whose data lives in plain JSON and Markdown files on your disk, with no database server and no account.

The audience is narrow and specific. You need Claude Code installed and opened in the project directory, according to the Quick Start section. The README also states that the project is built on Claude Code's command-line tooling and uses AI image recognition (it names the GLM tool mcp__4_5v_mcp__analyze_image) for document understanding. So the practical user is someone comfortable running a CLI agent, who has a pile of medical paperwork to structure, and who cares enough about where the data sits to accept a command-line workflow instead of a phone app.

Two disclaimers in the README matter more than the feature list. The project is not affiliated with, endorsed by, or associated with Anthropic or Claude.ai; it is an independent project by WellAlly Tech. And it is not a medical device: the README says all analysis is for personal reference only and must never replace professional medical advice.

How the command layer and the data directory fit together

There is no application server in the architecture described. The repository is organised around .claude/, which holds commands/ and specialists/, plus a data directory the README calls my-his/. Each slash command is a Markdown file: save-report.md, query.md, profile.md, radiation.md, surgery.md, discharge.md, medication.md, interaction.md, consult.md, specialist.md. The README claims 60+ commands in total. The specialists directory holds what the README calls 16 specialty plus MDT coordinator Skills.

Data flow is file-to-file. A command such as /save-report takes an image path, the AI layer extracts structured values, and the result is written into a typed subdirectory: 生化检查/ for biochemical tests organised by YYYY-MM, 影像检查/ for imaging data and image backups, 手术记录/ for surgery history, 出院小结/ for discharge summaries. Profile data goes to data/profile.json, radiation records to data/radiation-records.json, allergies to data/allergies.json. Drug interactions live in data/interactions/ with an interaction-db.json and an interaction-logs/ directory.

The interaction database is the most interesting design choice, because it is data rather than model behaviour. The README describes a five-level severity warning system labelled A, B, C, D and X, where X means absolute contraindication, and says checks cover drug-drug, drug-disease, drug-dose and drug-food pairs. Radiation tracking is also modelled rather than merely logged: the README mentions BSA-adjusted dosing and a decay model. Those two are the parts of the project that look like deliberate engineering instead of prompt packaging.

Installing WellAlly Health and saving a first lab report

The README gives one prerequisite and no package installation step: Claude Code installed and opened in this project directory. There is no npm install instruction in the Quick Start, and no release artefacts were retrieved, so treat the repository checkout as the distribution. The package.json declares engines of node >=18.0.0 and python >=3.8, which tells you what the surrounding scripts expect.

The Quick Start then walks through setup and ingestion. The first command sets body metrics and birthdate, which later calculations depend on:

/profile set 175 70 1990-01-01

The README's comment for this line says height in centimetres, weight in kilograms, birthdate. After running it, the README's directory listing suggests you should find data/profile.json holding those values.

Ingestion is a single command per document. The README shows a photograph of a lab report going in by path, and a discharge summary going in with an @ reference:

/save-report /path/to/image.jpg
/discharge @医疗报告/出院小结.jpg

The expected result is structured output under 生化检查/ for the lab report and 出院小结/ for the discharge summary, with the original image kept alongside the extracted data in the imaging case. Radiation and surgery entries are recorded in natural language rather than fixed fields:

/radiation add CT chest
/surgery Gallbladder removal surgery in August last year due to gallstones
/query all

The README positions /query all as the way to read everything back, and /consult as the entry point to a multi-disciplinary review. Because the storage format is open, the honest check on whether ingestion worked is to open the generated JSON yourself rather than trust the confirmation message.

Where WellAlly Health breaks down: no releases, stub scripts, and an agent dependency

The most concrete limitation is versioning and testing. package.json reports version 2.0.0 and describes the project as featuring 28+ interconnected health modules, but no releases were retrieved for the repository. More telling, the npm scripts are placeholders. test:unit, test:integration, test:cover, lint, lint:fix, format, format:check and typecheck each echo a message and exit 0. The test script chains test:unit and test:integration, so npm test succeeds without executing a test runner. Anyone evaluating this as software should read that as a statement about verification, not about intent.

The second constraint is the Claude Code dependency. Every workflow in the README is a slash command executed inside that tool. If Claude Code changes its command loading, or if you stop paying for or using it, the interface layer is gone. The files survive, which is the point of the plain JSON and Markdown format, but the automation does not.

The third is scope discipline. This is a record organiser with analysis prompts attached. The README says plainly that it is not a medical device and that analysis is for personal reference only. If you want a tool that a clinician will read, sign, or act on, this is the wrong category of software, and the drug interaction severity levels in a local JSON file carry no regulatory weight. Finally, no release notes or upgrade path appear in the README, so a schema change in profile.json or the interaction database has no documented migration story.

WellAlly Health compared with a self-hosted records server

The obvious alternative for someone who wants private health records is a self-hosted server such as an open source personal health record platform with a database, a web interface and user accounts. The difference in approach is not cosmetic. A server-based PHR gives you a query language, concurrent access from multiple devices, and a schema enforced by the database. It also gives you a service to patch, back up and secure, and a running process that can be reached over the network.

WellAlly Health inverts those trade-offs. There is no database server and no account, per the README, so the attack surface is your filesystem and whatever the AI agent does with a document you hand it. Backups are file copies. Portability is total: the README describes the format as open JSON plus Markdown, with no lock-in. What you give up is everything a database provides, including referential integrity between a medication record and the interaction log that references it, and any multi-user story at all.

The comparison that matters for a buyer is against doing nothing structured. A folder of scanned PDFs and a notes app is genuinely private and needs no Claude Code. WellAlly Health's argument is that extraction and cross-referencing are worth the setup, particularly the interaction checks and the radiation dose model, which are tedious to maintain by hand in a spreadsheet.

Maintenance status, licence and what an upgrade actually costs

The repository is not archived, and the last push was on 2026-07-16. That is the only maintenance signal available here. No releases were retrieved, and the CHANGELOG.md at the repository root is the place the README's structure implies change history lives, though its contents are not documented in the README. The npm scripts do not run real checks, so there is no continuous verification to point at either.

Licence is MIT, per both the README badge and the license field in package.json, with copyright attributed to WellAlly Tech. MIT is permissive: you can use, modify and redistribute the code, including commercially, provided the copyright notice and permission notice are preserved. That is the licence text, not legal advice, and it says nothing about the medical disclaimers, which are a separate matter of how you use the output.

Upgrade cost is the part worth thinking about before you commit records. Commands are Markdown files under .claude/commands/, and the specialists live under .claude/specialists/, so a pull that changes command definitions can change behaviour without touching your data. Your data sits in data/ with a fixed layout: profile.json, radiation-records.json, allergies.json, interactions/, medications/, and the Chinese-named record directories. If a future version renames or restructures those paths, the README documents no migration command. Copying data/ before you pull is the only reversible step available.

Editorial conclusion

Adopt WellAlly Health if you already run Claude Code, keep your own backups, and want structured lab, medication and radiation records in files you can read without the tool. Skip it if you need a clinical decision support product, a mobile app, a team deployment, or anything with a validated release process: package.json version 2.0.0 is the only version marker, no releases were retrieved, and the test, lint and typecheck scripts exit 0 without running a tool. Before committing real records, run /profile set once and open data/profile.json to confirm the height, weight and birthdate landed where you expect, then copy the data directory somewhere outside the repository.

Frequently asked questions

Official sources

1. huifer/WellAlly-health on GitHub
2. Issues
3. License: MIT
4. Project website
5. README