OPEN • PORTABLE • HUMAN-GOVERNED

Find what the documents disagree about.

Portable Agent Skills that check a Clinical Study Report, protocol, or briefing document against its own sources — and report what they found without pretending to clinical judgment they have not earned.

No account, backend, analytics, or document upload to this site.

01Source Preflight
02Mechanical Findings
03Defect Register
04Human Review Gate

THE MAP · 167 TASKS

A job, then a task, then a skill.

Writing it out that way was uncomfortable. You see how much of the job nobody has built anything for.

The profession made navigable in the browser - no AI, no install. Open a domain, then a subdomain, to reach a task page. Coverage is regenerated from the ledger whenever the map is built.

Skills review, reconcile, verify, structure, and flag. Qualified humans decide, approve, sign, and submit.

One subdomain: Study delivery / Design. Each task maps to a skill.
The live job map: three bands and fifteen domains, each row coloured by whether a skill carries it. Study delivery opened to Design. Covered work sits next to the gaps.

HOW IT WAS BUILT

Map the job. Then decide what needs a skill.

The recording is the same sequence as the library: write the profession down, split skill from context from human decision, build what can run, write the refusals, and leave the gaps on the map.

Play it. It does not autoplay - it is a minute long on purpose, so each step can be read.

How the library was built: map the job, split what needs a skill, write the refusals, publish the gaps.
The job model grouped into Band A foundations, Band B development, and Band C cross-cutting. A request to pick a phase 3 dose, stamped refused. Skills never select a dose, sign off, or submit. The map lists uncovered work next to covered work. The gaps ship with the wins.

151 PACKAGES — 151 RELEASED, 0 BUILT

Honest status, not a marketing count.

01

Released — 151

Each package passed its assigned evidence gate. That is not a clinical-validation claim. See the claim ledger for the live numbers.

Read the claim ledger →
02

Built — 0

A built package exists and validates, but qualification has not passed. None currently sit in that state.

Why that distinction →
03

Paste a block

Every package ships a generated PASTE.md. Paste it into an ordinary chat with only the source material permitted in that environment.

Browse the skills →

SYNTHETIC DEMONSTRATION

Every finding cites where it came from.

A review reads the documents you permit, compares values across them, and reports each disagreement with both sides and their locations. It never decides which value is right — that is the reviewer's job, and the tool says so.

Shown here: build-work-context, the released context utility. Fictional throughout. Missing details stay unknown rather than being invented.

ClinPharm PMx Skills is not medical advice, clinical decision support, a validated GxP system, or a replacement for scientific and regulatory judgment. No skill recommends a dose.

Synthetic demonstration showing a fictional user attaching a chat paste block, answering a short interview, and receiving a reusable professional work-context file.

MEASURED, NOT MARKETED

What was actually tested.

12 planted defects Expert-keyed fixture: 5 Critical, 4 Major, 3 Minor
6/6 script recall Deterministic path, 0 of 5 false-positive traps triggered
0 unqualified Stated as built, never counted as released

The 6/6 result is a historical deterministic-path diagnostic, not a current qualification result. No model-path qualification result is currently accepted. Review the prompts, raw outputs, rubric, scores, and limitations, or read the fixtures and expert keys.

WHERE THE SKILLS APPLY

Clinical pharmacology, end to end.

BUILD THE NEXT SKILL

Which review should become portable next?