openelis-user-manual

---

name: openelis-user-manual description: Author an OpenELIS Global user-manual entry (Word doc, Confluence-bound) for a feature or workflow, with an overview, numbered steps, and real cropped screenshots. Use whenever Casey wants to write/draft/update a user manual page, document a feature or workflow for the manual, or produce a how-to for OpenELIS. Captures media via the openelis-screenshots skill and builds the document with the docx skill. Hard rule: document only what is actually BUILT and verified on a live instance — never spec-only behavior.

OpenELIS user-manual authoring

Produces a polished, Confluence-bound manual entry for an OpenELIS feature/workflow. Pairs with two other skills: openelis-screenshots (captures the real screens + walkthrough video) and docx (builds the Word document). Output is a draft for Casey to review, then publish to the Confluence user manual.

The one rule that matters most

Document only shipped, verified behavior — not the spec. Many OpenELIS designs are specced but not yet built (e.g., environmental regulation-driven test suggestion is designed but, as of v3.2.1.x, selecting a compliance standard does NOT auto-add tests — staff pick tests manually). Before writing any step, confirm it on a live instance. If a capability is only in the FRS/spec, either omit it or mark it clearly as "planned / not yet available" — never describe it as if a user can do it today. When unsure, drive the live instance (via openelis-screenshots) and watch what actually happens; log behavior to confirm.

Workflow

1. Scope & verify. Identify the feature and the exact live route(s) (see openelis-screenshots/routes.md). Drive the real instance and confirm each step works as you'll describe it. Note required fields, gates (e.g., a Save button that stays disabled until X), and any behavior that differs from the spec. Pick the right instance: clinical flows → testing.openelis-global.org; environmental/vector → indonesiademo.openelis-global.org.
2. Capture media with the openelis-screenshots skill: one labeled screenshot per pertinent step plus the walkthrough video. For long forms, capture a full-page PNG and section offsets, then crop to the pertinent section so text is readable (recipe below). Save real demo data where allowed so screens show populated states, not empty forms. Mask PII.
3. Structure the document (see Outline). Lead with an Overview that gives context — what the feature is, who uses it, and when/why — before the steps.
4. Build the .docx with the docx skill (US Letter, Arial, themed headings, header/footer with page numbers, centered figures with italic captions). Validate it.
5. Deliver for review. Present the doc + video. Publish to Confluence only on Casey's explicit go (then it can be created/updated via the Atlassian connector in storage format).

Outline (use this shape)

• Title (feature/workflow name).
• At a glance — a compact label/value table BEFORE the steps (this is a job aid, not just feature docs). Five fields, in order:
  • Purpose — the task in the user's terms ("Register a new patient so they can be linked to test orders").
  • You'll end up with — the outcome/output ("A saved patient record, searchable when creating orders").
  • Who does this — the typical role, ALWAYS with the caveat that access is permission-based and configurable: "Reception staff. Access is set by user permissions and can be configured by your administrator." Don't publish an exact role/permission matrix — OpenELIS access is configurable per deployment (admin is largely all-or-nothing); naming precise permissions won't hold across sites.
  • Before you start — the data/materials needed (IDs, the physical samples, a chosen standard, etc.).
  • Where — the menu path (Main menu → … → …).
• Overview / context (optional, for richer flows) — 2-4 sentences naming the workflow shape; skip if the "At a glance" block already says enough.
• Step-by-step — numbered steps, imperative voice, one cropped screenshot per pertinent step with a figure caption. Call out required fields and any gates ("Save & Next stays disabled until…").
• Field reference — a table of the key fields and notes.
• Tips / notes — shortcuts, gotchas, and how anomalies are handled (e.g., NCE).
• Keep claims to verified behavior; if you reference something not-yet-built, label it explicitly.

Cropping recipe (readable section images from a long form)

In the capture spec: after filling the form, take a full-page screenshot and record each section heading's document offset:

await page.screenshot({ path: `${OUT}/_full.png`, fullPage: true, animations: 'disabled' });
const offsets = await page.evaluate(() => {
  const names = ['Lab Number','Sampling Site','Default Collection Conditions','Per-Sample Manifest'];
  const els = [...document.querySelectorAll('h2,h3,h4,div,label,span')];
  const out = { _dpr: devicePixelRatio };
  for (const n of names) { const e = els.find(x => (x.textContent||'').trim() === n && x.getBoundingClientRect().height < 60);
    if (e) out[n] = Math.round(e.getBoundingClientRect().top + scrollY); }
  return out;
});
require('fs').writeFileSync(`${OUT}/_offsets.json`, JSON.stringify(out));

Then crop in the sandbox with PIL (multiply offsets by _dpr; crop from one section's top to the next section's top):

from PIL import Image, json
im = Image.open('_full.png'); o = json.load(open('_offsets.json')); d = o['_dpr']
order = ['Lab Number','Sampling Site','Default Collection Conditions','Per-Sample Manifest']
ys = [(n, int(o[n]*d)) for n in order if n in o]
for i,(n,y) in enumerate(ys):
    y2 = ys[i+1][1] if i+1 < len(ys) else im.height
    im.crop((0, max(0,y-20), im.width, y2)).save(f'{n.lower().replace(" ","-")}.png')

Crops are full-resolution, so the section text stays legible in the doc.

Document styling (docx)

• Page: US Letter (12240×15840 DXA), 1" margins. Font: Arial. Themed H1/H2 color per domain (clinical blue 15487F, environmental green 1B5E20).
• Header rule + footer with menu-path breadcrumb and page number.
• Images centered; cap very tall images by height (~800px) and wide ones by width (~600px); prefer cropped sections over full-page where legibility matters. Italic gray captions.
• Required ImageRun type + altText. Validate with the docx skill's validate.py.

Confluence publishing — automated, images and all (WORKING)

Alongside the .docx, emit a Confluence storage-format XHTML + an images/ folder; the publisher uploads the PNGs as page attachments and embeds them. Verified end-to-end (the Atlassian MCP can write a page body but cannot upload attachments, so we use the REST publisher).

Storage XHTML format: prose + plain captions, with each image slot a token on its own line: <p>{{IMG:01-search.png}}</p> (one per screenshot; the matching PNG lives in images/). The publisher replaces each token with <ac:image><ri:attachment ri:filename="01-search.png"/></ac:image>.

Publish (run on Casey's Mac; token lives in ~/.openelis/confluence.env, never in chat):

cd ~/Documents/OpenELIS\ QA/docs-manual/publish
# new draft for review (parent 1189609473 = "New User Manual sections to be verified", space OG):
python3 confluence-publish.py --create --space OG --parent 1189609473 \
  --title "<Feature> (draft — verify)" --xhtml <section>.xhtml --images <section>/images
# update an existing page instead:
python3 confluence-publish.py --page <pageId> --xhtml <section>.xhtml --images <dir>

Images are auto-sized with ac:width (default max 1020px; small images keep native size — never upscaled) so screenshots don't render at full resolution; override with --max-width. Re-runs are idempotent (attachments matched by filename, page version bumped). --dry-run previews. The publisher ships inside this skill at publish/confluence-publish.py (also in openelis-work/docs-manual/publish/). Always publish a draft under 1189609473 for verification first; promote to the live manual only after Casey signs off. The script + a worked Add-a-Patient example live in openelis-work/docs-manual/publish/ (see its README).

Fallbacks: Word import (Confluence → Insert → Import Word document on the .docx — images come inline, needs no token) or manual drag (publish prose with empty <p/> slots + a grey [ drop image: ... ] hint per figure; reader drags the PNG in).

Terminology: the left navigation is the Main menu — write "Main menu → …", never "left menu", "side menu", or "sidebar".

Caption rule (all paths): captions describe the screen for the reader only — never include "captured", "captured automatically", or tool/provenance wording; that would force a cleanup pass. Link the new page from the relevant manual index; keep the walkthrough video as a linked asset.

Provenance & staleness (feeds the Doc Freshness Tracker)

Every page you author must register a provenance + drift contract so it can be watched for going stale. Add/extend an entry in OpenELIS QA/docs-manual/contracts.json:

• id, title, manualDoc, captureSpec
• base (the instance the manual documents) and capturedVersion (OpenELIS version captured against)
• routes[] — each with path and anchors[]: the on-load headings / field labels / buttons the manual relies on (e.g., "Lab Number", "Save & Next", "New Patient"). Pick anchors visible on initial load so the check doesn't false-positive on interactive-only elements.

tests/docs/drift.check.spec.ts visits each route and asserts the anchors exist; it writes docs-manual/drift-report.json with per-section status: ok (anchors present, version matches) · review (instance version ≠ capturedVersion) · drift (a documented control is missing → the UI changed, re-capture and re-verify) · error (route failed). The weekly openelis-docs-drift-weekly scheduled task re-runs it, commits the report to DIGI-UW/openelis-work (docs-manual/), and the OpenELIS Doc Freshness Tracker artifact renders it in its "Manual section UI drift" card. So when this skill ships a page, it also makes that page self-monitoring: if the screen later changes, the tracker flags the exact section and missing control. Re-run the capture and rebuild the doc, then bump capturedVersion.

Review gate (run before publish — fresh eyes, not the author)

Run this on every page before publishing. Do it as re-derive, then diff — ideally with a review subagent or a second person, not the author (the author anchors on what they built). First re-derive the ideal page from scratch: state the goal, the minimal real workflow to achieve it, and the steps and screenshots that workflow actually needs. Then diff the draft against that and apply these gates:

1. One-sentence workflow test. Write the workflow as a single sentence; every step must serve it. Cut anything that is setup belonging to another page, or unrelated configuration. (A page titled for feature X should not be two-thirds prerequisite admin config.)
2. First real step. Mark where the task actually begins and delete everything before it — menu navigation, and prerequisite configuration that is documented elsewhere. The first real step is the first action unique to this workflow.
3. Step ↔ screenshot fidelity. Each step has exactly one screenshot that shows that step's action or state. No generic landing pages, no empty / "no records" screens, no mismatched selection. If a screenshot does not show what its step describes, it is wrong even if it is a real screenshot.
4. Verified at the workflow level. Every step was driven live end-to-end. A step you could not reach (a gated dialog, a feature toggle that is off) is a blocker to resolve before publishing, never prose carried over from an older draft or the spec.
5. Process gates. Authored through this skill's Outline (At-a-glance first); published as a draft under the review parent (1189609473) first, never written straight over a live page; drift contract registered in contracts.json.

The highest-leverage habit is gates 1-2: treat the existing page as a suspect to prune, starting from the goal — not as a baseline to fix in place.

Definition of done

Overview gives real context; every step is verified against the live build (no spec-only claims); images are cropped and legible with captions; PII masked; a walkthrough video accompanies it; the doc validates; and it's delivered for review before any publish. The Review gate above has passed (re-derive-then-diff, step/screenshot fidelity, workflow-level verification, draft-first).