AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

clinical-reports

Creates safety-bounded draft structures and runs local deterministic checks for clinical case, diagnostic, trial, safety, and aggregate research reports. Use only with synthetic, de-identified, or aggregate inputs and verified source-fact manifests; every output requires qualified review.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--clinical-reports at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 12

references/data_presentation.md

Aggregate Data Presentation

Present verified outputs without changing their meaning. Do not calculate a clinical conclusion, select an analysis, or repair source discrepancies.

Mandatory table metadata

Every table should identify:

  • artifact and analysis purpose;
  • data cut and source-output version;
  • analysis population/set;
  • treatment/group labels;
  • denominator for every group and row when it varies;
  • whether a value is a subject count, event count, observation count, or estimate;
  • units, time point/window, and summary statistic;
  • missing, unknown, not assessed, suppressed, or not applicable values;
  • coding dictionary/version/language where applicable;
  • statistical method and multiplicity status only when copied from a verified output;
  • provenance manifest and reviewer status.

Counts and denominators

  • Show n/N (%), not a percentage alone, unless the governing output specifies another form.
  • For a proportion of distinct participants, require 0 <= n <= N and N > 0. This bound does not apply to recurrent event counts; preserve the verified count or exposure-time rate and label its denominator separately.
  • Recalculate only for consistency checking; do not silently replace the reported percentage.
  • State the rounding rule and tolerance.
  • Do not add subgroup percentages when the subgroup denominator is unknown.
  • Do not infer that denominators are randomized, treated, evaluable, or safety populations.
  • Keep event counts distinct from subjects affected; event counts can exceed the number of subjects.
  • Do not sum non-mutually-exclusive categories.

Dates and time

  • Use YYYY-MM-DD calendar dates for the date checker; it rejects compact and week-date forms so precision and representation are not silently changed.
  • Preserve source timezone and precision.
  • Distinguish event date, collection date, database cut, report date, and verification date.
  • Flag start-after-end and conflicting dates.
  • Do not impute a missing day, month, timezone, or chronology.
  • For de-identified case reports, use an authorized relative timeline; do not distort intervals.

Units and precision

  • Preserve source units.
  • Require a unit for every dimensional quantity.
  • Use one verified unit per comparable series or expose the mismatch.
  • Do not convert or normalize without an authorized traceable conversion rule.
  • Preserve clinically meaningful precision; do not create extra significant digits.
  • Identify SD, SE, CI, IQR, range, and denominator explicitly.

Missing and suppressed values

Keep these states distinct:

  • missing;
  • not_collected;
  • not_assessed;
  • unknown;
  • not_applicable;
  • suppressed_for_privacy;
  • zero.

Never convert a blank to zero. State small-cell suppression rules and ensure totals or complementary cells do not reveal suppressed values.

Adverse-event tables

For term-level aggregate tables:

  • use each required CSV header exactly once, with no surrounding whitespace; duplicate headers can overwrite values during dictionary parsing and are rejected;

  • state MedDRA version and language;

  • state analysis set and denominator;

  • state counting rule from the verified SAP/output;

  • show subjects affected and event count separately;

  • do not add p-values or causal labels;

  • do not interpret between-group differences;

  • preserve threshold rules exactly;

  • have safety and statistical reviewers verify deduplication, hierarchy, and population.

Figures

No figure is mandatory. Create one only when requested, supported by verified aggregate data, allowed by the target guidance, and reviewable without external image generation.

A CONSORT flow diagram is part of CONSORT 2025 reporting, but every count and reason must come from verified trial outputs. A missing count remains missing; do not create a decorative or inferred diagram.

Deterministic consistency check

consistency_checker.py can inspect:

  • ISO date and range ordering;
  • unit-label consistency;
  • n/N (%) arithmetic;
  • component-total arithmetic.

It reports mismatches and review needs. It does not change the input, choose the correct source, or validate clinical/statistical meaning.