# ISO Desk

> Before an ISO readiness pack goes to an auditor or assessor, drop its JSON records - scope and
> applicability intake, document and record register, CAPA records, design/risk traceability,
> supplier controls, FDA QMSR transition evidence, evidence manifest. A free in-browser check runs
> the iso-standards-readiness agent skill's own standard-library checks with the same JSON report
> and exit status as its Python scripts; two metered lanes add judgement: a draft evidence review for
> authorized human assessment, or one drafted controlled document with its register row.

URL: https://iso-desk.skillsafe.ai/
API: https://iso-desk.skillsafe.ai/api.html
Tokens: https://iso-desk.skillsafe.ai/tokens.html
Model: gpt-terra (the balanced GPT tier alias on SkillSafe)
Source skills: @k-dense-ai/iso-standards-readiness (primary) and @k-dense-ai/iso-13485-certification,
which it supersedes (k-dense-ai/scientific-agent-skills, MIT, K-Dense Inc.).
Standards: ISO 13485 (medical device QMS, certification lane), ISO 14971 (device risk management,
no certificate of its own), ISO/IEC 17025 and ISO 15189 (laboratories, accreditation lane).
Contains no ISO or IEC clause text. Not affiliated with ISO, IEC, FDA or any certification or
accreditation body.

## What runs free, in the browser

A JavaScript port of the skill's scripts (skills/iso-standards-readiness/scripts at commit 1e5eeff),
checked byte for byte against CPython 3.12 on generated records (see NOTICE.txt):

- validate_scope_intake.py (--standard): organization roles, activities from the profile's
  vocabulary, sites, products or scope items, markets; an undetermined applicability decision is a
  blocker (HUMAN_DECISION_REQUIRED);
- audit_document_records.py: documents, records and external sources with owners, evidence,
  approvals and source references; effective documents need approval and an effective date;
  superseded references must resolve; a non-current external source is a gap;
- check_capa.py: closed CAPAs need implemented actions and an approved "effective" result
  (CLOSURE_BLOCKED); an ineffective result must reopen;
- check_traceability.py: artifacts and rows linking intended use, hazards, risk controls, design
  inputs and outputs, verification, validation, production and postmarket sources; references must
  resolve; a verified row may hold no open gaps;
- check_supplier_controls.py: high and critical approved or conditional suppliers need their
  risk-based controls implemented or verified (CONTROL_NOT_READY);
- check_qmsr_transition.py: the post-2026-02-02 QMSR basis (Compliance Program 7382.850), the
  14 catalog items and four attestations;
- validate_evidence_manifest.py and gap_analyzer.py (--standard): entries by process domain and a
  per-domain status (not-assessed, evidence-missing, evidence-incomplete,
  evidence-present-for-human-review). The optional --verify-files SHA-256 comparison needs your
  local evidence files and runs only with the skill's own script.

The page adds labelled "page" checks quoting the skill's rules: a design traceability matrix under a
laboratory profile, QMSR evidence under a laboratory profile, a manifest purpose in the wrong lane,
"ISO 17025 / 15189 certified" wording, ISO 14971 certification, ISO 15189:2012, Compliance Program
7382.845, "Amendment 1:2021", ILAC MRA / IAF MLA wording, and forbidden titles.

## Metered lanes (run input field `task`)

- `review`: a JSON "Draft evidence review for authorized human assessment" - declared basis, a
  stance on every blocker (B) and page check (P), findings covering every gap group (G) with actions
  and owner roles, sampling limits, unresolved decisions, the next authorized decision and
  boundaries. Status: blocked, gaps_to_close or ready_for_human_review.
- `draft`: one controlled document (procedure, work instruction, quality-manual section or plan)
  with eight fixed sections and a draft document-register row, which the page re-checks with
  audit_document_records.

Every reply is reconciled in the browser: items answered once, no blocker argued away, coverage of
every gap group, status no looser than the facts, the titling rule, numbers and dates present in what
was sent, and no compliance, certification or inspection-readiness claim.

## Boundaries

Structural evidence checks only. Nothing here certifies, accredits, determines applicability or
compliance, or predicts an audit, assessment or inspection result, and the checks never open the
evidence files themselves.