synthetic-users
Skill5 starsChanged 27 days agoMITView on GitHub
Pilot the v1 research guide against synthetic personas —every trait marked as evidence, public data or assumption, every answer naming the trait it rests on, at least one skeptic— to find broken questions before spending real fieldwork, and write the Diagnosis + Script v2. Manually triggered, fourth station of the discovery pipeline; it validates nothing.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- miguelclavel/ux-research-skillstheir own version, first seen Nov 8, 2026
- The-Utopia-Studio/skillstheir own version, first seen Oct 13, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- The-Utopia-Studio/skills★ 6same name
- miguelclavel/ux-research-skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository12 other files it gives its agents.
- cpoSkill
- ctoSkill
- grillSkill
- ideation-engineSkill
- insight-engineSkill
- market-researchSkill
- prdSkill
- product-builder-resolverSkill
- product-metricsSkill
- product-strategySkill
- ux-researchSkill
- visual-designerSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 35
Every simulated persona carries its traits with an explicit origin, marked one by one:
- Very specific: line 47
Search (datasetsearch.research.google.com).
- Distinctive: line 64
No "I love it!" without a cost next to it: Every enthusiastic answer from a synthetic persona is accompanied by its cost or its friction: …
- Distinctive: line 71
Every question of the v1 script is reviewed against this list.
- Distinctive: line 108
Defects found and rewrites: every defect from the taxonomy that showed up, with its rewrite.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
synthetic-users — the Diagnosis + Script v2
Fourth station of the pipeline. It pilots the guide that ux-research built against synthetic
personas —simulated profiles with explicit traits— before spending the first minute of real
fieldwork. Generic domain example: a B2B appointment-management app could simulate an administrator
of a small clinic and a skeptical one who already tried and abandoned a similar tool.
When it is invoked
| When | From / to |
The Research Plan from ux-research already has a v1 guide | Entry from ux-research (.agents/skills/ux-research/SKILL.md) |
| The Diagnosis + Script v2 is written | Exit: real fieldwork — interviews with real people, the work the operator picks up; it is not a skill of this pack |
| After real fieldwork | The raw answers are converged by insight-engine (.agents/skills/insight-engine/SKILL.md), and from there the pipeline reaches the strategic layer that prd writes (.agents/skills/prd/SKILL.md) |
Rule number one
This validates NOTHING. It opens the session with this sentence and repeats it on closing: no answer from a synthetic persona is evidence that something works with real users. It serves one purpose —finding defects in the guide before spending fieldwork time— and no other. A Diagnosis that reads as if it validated the hypothesis is badly written, no matter how many times the rule was repeated in the text.
The synthetic personas
It is run with the pressure method of .agents/skills/grill/SKILL.md#method — cited by path, never copied —
while building each profile: every trait proposed is pressured for its origin before entering the
profile.
Every simulated persona carries its traits with an explicit origin, marked one by one:
| Mark | Origin |
| (evidence) | Real research already done on this segment —an interview, a usage figure, something the Discovery Brief or the Market Brief brings |
(public data: <source>) | A public dataset that describes a population the segment belongs to, cited by name and link |
| (assumption) | A construction with no research and no public data behind it |
A trait with no mark does not enter the profile.
A trait is searched for in public data before it is marked as an assumption. The catalogs to
search are Kaggle (kaggle.com/datasets), Hugging Face (huggingface.co/datasets) and Google Dataset
Search (datasetsearch.research.google.com). If a dataset covers the trait, it is marked
(public data: <source>); if none does, it stays (assumption).
Public data is never marked as evidence. A public dataset describes a general population, not the segment of this product: counting it as evidence would make a synthetic persona look more validated than it is.
At least one skeptical persona, with a reason: someone who already tried something similar and abandoned it, someone with a business objection (the cost, the process change), or someone who simply does not have the problem the hypothesis assumes. A set of personas that only confirm the hypothesis is no use for finding defects: it is useful for confirming them.
Every simulated answer names the trait it rests on, with that trait's mark: "I wouldn't use it, I already have Jira (rests on: uses Jira daily, public data)". The answer itself is always invented, whatever the mark of its trait. Naming the trait shows which answers stand on something and which are pure construction, and it tells a broken question apart from a badly built persona.
No "I love it!" without a cost next to it. Every enthusiastic answer from a synthetic persona is accompanied by its cost or its friction: what it costs them, what worries them, what they would have to stop doing. An answer made only of enthusiasm is the signal that the guide is built to confirm, not to learn.
The taxonomy of question defects
Every question of the v1 script is reviewed against this list; the defect found is recorded with its rewrite next to it, never merely pointed at:
| Defect | Example | Rewrite |
| Leading question (suggests the answer) | "Don't you think an automatic reminder would be useful?" | "How do you keep from missing an appointment today?" |
| Hypothetical question ("would you use...?") | "Would you use a feature that rescheduled appointments on its own?" | "Tell me about the last time you had to reschedule an appointment" |
| Double question | "Is the dashboard clear and fast to use?" | Split into two questions, one per attribute |
| Question with no behavior behind it | "What do you think about appointment management in general?" | "When was the last time you managed appointments for your team?" |
| Jargon the segment does not use | An internal technical term of the product team | The word the segment uses, taken from prior research |
The Diagnosis + Script v2
It closes by writing the deliverable as dated research of the loaded product node, in:
research/<YYYY-MM-DD>-script-diagnosis.mdDated, never overwritten. The path falls inside content: */products/*/research/*.md
line of the brain's own tree.md: no new glob and no resolver row are needed. Without a product
node, this same file — <YYYY-MM-DD>-script-diagnosis.md — is what lands in the current folder
instead.
The Diagnosis carries these four sections, in this order, each with its literal heading:
## Synthetic personas
## Simulation
## Defects found and rewrites
## Script v2- Synthetic personas: each profile, with its traits marked
(evidence),(public data: <source>)or(assumption), and the skeptical persona identified as such. - Simulation: the simulated answers against the v1 script, each one naming the trait it rests on, every enthusiasm with its cost next to it.
- Defects found and rewrites: every defect from the taxonomy that showed up, with its rewrite.
- Script v2: the corrected guide, ready for real fieldwork.
The Script v2 section opens and closes with rule number one written out: "this validates NOTHING" — before the first question and after the last one.
What this deliverable does not claim
The Diagnosis + Script v2 does not replace real fieldwork: it is the step that reduces how much
broken script reaches the first interview with a real person. Synthesizing what comes out of real
fieldwork —transcripts into findings— is not the job of this skill: that is insight-engine.
Destination
The skill resolves the destination before producing the deliverable, never after. Three cases, two destinations:
- A product node is loaded: the deliverable goes to
research/<YYYY-MM-DD>-<name>.mdof that node. - There is a brain, but no product node — no workspace yet, or a workspace without one: the deliverable is written to a file in the current folder.
- There is no brain at all: the deliverable is written to a file in the current folder.
With no product node, the file is written to the current folder anyway. The closing message names only the path it just wrote and stops there — that is the normal way this skill ends, not an anomaly to qualify.
This skill never asks which destination to use, and it never invents a third one.
The rest of the pack installs one skill at a time. Look at .agents/skills/ first and then .claude/skills/ (a skill present in both
counts once) and offer only the ones that are not there — once per session: the first time this skill closes in the session,
never again on a later close of the same or another deliverable:
grill—npx skills add pedroromeroluna/ai-first-product-skills --skill grillinsight-engine—npx skills add pedroromeroluna/ai-first-product-skills --skill insight-engineprd—npx skills add pedroromeroluna/ai-first-product-skills --skill prdux-research—npx skills add pedroromeroluna/ai-first-product-skills --skill ux-research
The whole pack at once: npx skills add pedroromeroluna/ai-first-product-skills. That command installs what the pack offers;
anything listed above it is installed by naming it.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

