dashboard · insilico-org — Claude Code skill: reports and installation
This Claude Code skill turns PLSI data into an HTML dashboard with charts. It runs when PLSI requests a state, plan, research, or target summary.
MIT ★ 93 updated 2026-11-15
npx skills add Insilico-org/plsi --skill dashboard -a claude-codeWhat dashboard does
The skill builds a DashboardPayload that defines the dashboard contents, then sends it to plsi report to render a single HTML file. The PLSI skill making the request determines the scope. Available blocks include tables, time-series charts, scales, timelines, and more.
It can create summaries for a trajectory, current state, plan review, clinician visit, research topic, or one or more targets. A physician sees a patient report before the patient does. PLSI and the plsi report command are required; the HTML is read-only, and clicks do not send choices back to PLSI.
What it helps with
- create an HTML summary of current state
- compare a plan with observed results
- show research evidence and open questions
- compare selected targets in a table
Example prompts for the agent
- Create a dashboard showing how I am doing now.
- Show how the plan compares with observed results.
- Compare these targets in a dashboard.
How to install dashboard
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill dashboard -a claude-code2Manually: put the files in place
~/.claude/skills/dashboard/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/dashboard/
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill dashboard -a codex2Manually: put the files in place
.agents/skills/dashboard/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill dashboard -a cursor2Manually: put the files in place
.cursor/skills/dashboard/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Insilico-org/plsi skills/visualizer/dashboard2Manually: put the files in place
.github/skills/dashboard/SKILL.md1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill dashboard -a gemini-cli2Platform CLI
gemini skills install https://github.com/Insilico-org/plsi --path skills/visualizer/dashboardManually: put the files in place
.devin/skills/dashboard/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about dashboard
How do I install dashboard in Claude Code?
Run npx skills add Insilico-org/plsi --skill dashboard -a claude-code. The skill file is at ~/.claude/skills/dashboard/SKILL.md.
What does dashboard do?
It prepares PLSI report data and renders it as a standalone HTML dashboard with relevant tables and charts.
What does dashboard need to run?
It requires PLSI and the plsi report command to render the DashboardPayload as HTML.
Source text
SKILL.md · MIT
---
name: dashboard
description: Use as PLSI Visualizer when PLSI asks for a custom dashboard (trajectory brief, state now, plan review, life arrow, clinician brief, panel, research topic, target dossier or comparison) - build a DashboardPayload whose scope the requesting skill defines, render it with plsi report and return it to PLSI with a matching chart.
---
# PLSI dashboard
A dashboard is one self-contained HTML file rendered from a DashboardPayload
(`contracts/DashboardPayload.schema.json`); the scope below decides what goes in, the requesting skill decides which
scope. Rendering is read-only: never ask before building one. Clicks in the HTML never reach PLSI; every choice the
user makes about a dashboard is the question card PLSI sends with it.
## When to use
- `nostradamus/trajectory` returned `trajectory.json` (Plan flow, "where am I heading"): scope **trajectory brief**.
- The user asks how they are doing now, how a plan went, or for their life arrow.
- `orchestrator/physicians` prepared a visit summary, or `visitprep/brief` a pre-visit brief: scope **clinician
brief**.
- `panel/overview` returned `panel.json` (physician role): scope **panel**.
- The physician asks for a report for a patient (physician role, `panel/clinic`): scope **patient report**, built
for that patient and shown to the physician first.
- `lab/research` shows a research topic's progress or its finish (`research_status.json`): scope
**research topic**.
- Target Hunter handed over one target (`target_<gene>.json`): scope **target dossier**; two to eight for "which of
these": scope **target comparison**.
## Blocks
`markdown {text}` (no raw HTML) · `kv {items:[{key, value}]}` · `table {columns, rows}` (cell: text, number or
`{text, href, level}`) · `timeseries {series:[{name, unit, points:[{t, v, ref_low, ref_high, flag}], predictions:[{t, expected, low, high}]}]}` ·
`list {items:[{text, href}]}` · `timeline {now, events:[{t, t_end, label, kind: past|predicted, level: info|watch|high, probability,
detail, href}]}`.
`parts {items:[{label, value, tone}], unit}` (one whole in at most 6 parts, instead of a pie) · `meter {value, max,
label, tone}` (one share of its total) · `scale {unit, marks:[{label, value, tone}]}` (values on one number line, such as
ages) · `bars {unit, left, right, items:[{label, value, note}]}` (signed amounts around zero). A `kv` item with
`hero: true` (and `unit`) is the view's one large figure. Any block may name an `icon` (`patient`, `panel`, `visits`,
`lab`, `vitals`, `device`, `medicine`, `condition`, `diet`, `plan`, `follow_up`, `file`, `trend`, `alert`, `note`, `age`).
## Scope
| Scope | From | Blocks, in order |
|---|---|---|
| trajectory brief | `nostradamus/trajectory` | markdown (the brief, unchanged) · one timeseries block with two series per marker item, "<label>: current course" and "<label>: with plan" (each with the recorded points incl. `now`, so both start at today; `current` / `plan` level points as predictions; same prefix and unit share one axis, the plan dashed) · table "Two courses" (item, now, current course, with plan, basis; risks as `expected (low-high) %` by horizon) · list (sources, linked) |
| scenarios | `nostradamus/trajectory` (Scenarios mode) | markdown (the brief, unchanged) · one timeseries block per measurement item, one series per scenario ("<label>: <scenario>", the baseline solid, the others dashed) · one timeseries block "Chance of being alive" with each scenario's survival curve (age on the axis) · table "Scenarios" (scenario, each item at the horizon as `expected (low-high)`, life expectancy, median age at death, chance of reaching 80 and 90, years vs baseline, basis) · list (sources, linked) |
| state now | `consultant/assess` result | kv (as of, maps, gaps count) · table (coordinate, value, unit, reference range, date) · list (gaps and next measurements) |
| plan review | `reviewer/review` result | kv (plan, review date, decision proposed) · timeseries (observed vs predicted, with interval) · table (prediction, expected, interval, observed, score); with a `modify` proposal append the trajectory brief's markdown, timeseries and Two courses table |
| life arrow | `orchestrator/arrow` | timeline · markdown (the story) |
| clinician brief | `orchestrator/physicians` (visit) or `visitprep/brief` | markdown (one line: a patient-made summary for discussion, not a diagnosis or referral; from `visitprep/brief`: a pre-visit brief for the physician, not a diagnosis or order) · kv (as of, maps, active plans, open gaps) · table "Recent results" (coordinate, value, unit, date, reference range, change since previous) · table "Active plans" (action, start, prediction with interval and review date, score so far) · list "Questions to ask" (from `visitprep/brief`: "To cover"); a timeseries only for a trend a question is about |
| panel | `panel/overview` | kv (as of, patients, per group count) · table "Needs attention" (label, group, reason, since, next), red flags first · table "On track" (label, last update) |
| research topic | `lab/research` (`plsi research status --full --project <id>`) | markdown (the question in chief; the answer so far with its verdict in words and what limits it: `limited_by`, `ruled_out`, `nothing_found_for`; no answer yet: "No answer yet") · timeline "Moves" (one past event per log row with a `move`: label the move and its question, detail what it cost (`cost.delta`) and what it changed (created, closed, verdicts flipped), level `watch` when not `productive`; `day` and `stop` rows as info events) · kv (verdict, depth, spent today of the allocation, spent this week of the ceiling, moves by kind, what the spend bought from `progress.segments`) · table "Evidence" (hypothesis, for or against, origin, design, findings, source ids from `sources`; the `set_aside` findings as "set aside") · table "Open frontier" (question, verdict, gap, why) · list "Closed directions" (each with its reason). Never a percentage of the research done |
| target dossier | Target Hunter (`plsi targets show`) | markdown (the verdict with its three ratings and the rationale from `assessment`, or "Not assessed yet") · kv (gene, name, Ensembl id, UniProt, indication, gathered, development level with its `evidence`, band with `band_basis`, furthest stage here and anywhere) · table "What the records say" (read, value, what it means: unread sources first as "not read", then structures (co-crystals, read of deposited), chemistry, genetics, safety and essentiality, aging (animal and human apart), assets in development (a floor), trials from the assessment's `harvested_assets`) · table "Risks" (dimension, rating, evidence, mitigation) · list "Open questions" (`open_questions` and `disagreements`) · list "Sources" (every id the assessment cites, linked to its record) |
| target comparison | Target Hunter (`plsi targets show` per target) | markdown (the question, then one line: "Compared side by side, never ranked: the rows share no unit") · table (one column per target in the order named, one row per read: band, furthest stage here, modalities, co-crystal, chemical matter, genetics, aging link, assets, verdict, the three ratings, falsifier; a last column "differs" marks each row whose values are not all equal; an unassessed target's judgment rows say "not assessed") |
| patient report | the physician, for one patient (`panel/clinic`) | markdown (the physician's own message, unchanged, or none) · table "Since your last review" (results dated after the roster row's `reviewed` date, each with its previous value; a value replaced by a new version of a report says "corrected", not "changed"; no `reviewed` yet: "First report: there is no earlier review to compare with") · one timeseries per trend the physician picked, with one plain line under them: "Each result is compared with the range printed on its own lab report; blank months had no data" · table "Your recent results" (test, value, unit, date, the range and the lab's flag printed on that report) · table "Your medicines" only when the physician asked (name as written, current / past / not sure: a past or unclear one is never shown as current) · list "Next steps" (only what the physician stated) · last, one line: "A summary of your records that your physician reviewed with you; not a diagnosis or a treatment plan." Plain words a patient reads without help: no codes, map or coordinate ids, hashes, model names, confidence labels or internal terms; no number that is not a recorded value or one the physician stated; nothing about other patients |
Order is fixed: the takeaway first, pictures next, tables and sources last. At most 6 blocks.
## Steps
1. **Collect** the requesting bot's file named in PLSI's request (`work/<thread>/trajectory.json` etc.) and the recorded data:
```bash
plsi export --instance <instance>
plsi due --instance <instance>
```
Scopes **research topic**, **target dossier** and **target comparison**: only the file PLSI names
(`research_status.json`, `target_<gene>.json`), written from the commands in the Scope table; render them on the
user's own instance (relation `self`), `subject_id` the topic name or the genes.
Scope **panel**: `panel.json` and `plsi instances list` instead; the payload is rendered on the physician's own
instance and holds patient labels and the physician's own handles (roster `handle`) only, never full names. Its
values stay in that report until a patient is
forgotten: name the report path in the `result` so `memoryhole/forget` can find and delete every panel naming
the patient.
2. **Write** `<PLSI_HOME>/work/<thread>/dashboard.json`, copying values from those files only:
```json
{"kind": "plsi_dashboard", "schema_version": 1, "title": "Where you are heading", "subject_id": "<subject_id>",
"generated_at": "2026-10-21T09:00:00Z", "blocks": [
{"type": "markdown", "title": "In brief", "data": {"text": "..."}},
{"type": "timeseries", "title": "Biological age gap", "data": {"series": [
{"name": "Biological age gap: current course", "unit": "years", "points": [{"t": "2026-09-14", "v": 2.1}],
"predictions": [{"t": "2027-01-19", "expected": 2.1, "low": 0.6, "high": 3.6}]},
{"name": "Biological age gap: with plan", "unit": "years", "points": [{"t": "2026-09-14", "v": 2.1}],
"predictions": [{"t": "2027-01-19", "expected": 1.1, "low": -0.9, "high": 2.6}]}]}}]}
```
3. **Render** and keep the `uri`:
```bash
plsi report --instance <instance> --payload <PLSI_HOME>/work/<thread>/dashboard.json
```
Scope **clinician brief**: also render the same payload with `--format pdf` (a static A4 copy for e-mail) and
return the PDF path next to the HTML one in step 5. Scope **patient report**: render the PDF only (it lands in
the patient's own `reports/`); the `result` names the work for Data Controller (`controller/egress`, kind
"patient report"); PLSI shows only a cleared PDF to the physician on one card (Approve for the patient / Edit /
Don't share), and only an approved file goes to `panel/clinic` step 5.
4. **Chart for chat.** Draw the same data as one PNG following `visualizer/charts` (trajectory brief: both courses
on one axis per item, bands for intervals, the plan course lighter and dashed). The research and target scopes
have no PNG: the HTML report is the file.
5. **Return** to PLSI in one `result`: the brief, the PNG path and the HTML report path (`uri`). PLSI sends them, in
this order, as one turn: the brief with the PNG, the HTML report as a chat attachment and, last, in the Plan flow,
the single adoption card (PLSI's `orchestrator/conduct` step 7). If rendering fails, say so in the `result`; PLSI
never sends the adoption card without the PNG and the HTML report.
## Rules
- Never in a dashboard: names, places, employers, raw record text or inbox file contents, numbers not in a bot's file
or the export, points without their interval, intervention names beyond the plan's action labels (the research
and target scopes: beyond those in the topic's or the dossier's own file), secrets.
- Level-only items go in the table as "not predicted", never as a drawn line.
- `change_from_baseline` points go in the Two courses table only, never on a level timeseries.
- Target scopes never show a composite score, a probability of success, a rank, a "best" or a total across rows; a
sample carries its denominator (`n of N`, read of deposited); a source in `unread` is "not read", never 0.
- One dashboard per thread step; regenerate rather than edit an old report.
© Insilico-org · MIT · Insilico-org/plsi
Similar skills
All skills for Agent setup: memory, context, orchestration →dashboard by other authors comes first.