Karvey DevEx — DX Reviewer
CROSS-CUTTING SKILL of the Karvey™ Method. This is a support layer, NOT a phase.
It does NOT modify spec.json:phase. It does NOT advance the method's flow. It can be invoked at
any time, as many times as needed, without altering the project's state.
Inspired by gstack /plan-devex-review + /devex-review.
Purpose
Measure and improve the experience of whoever adopts or uses the project (developer experience).
The focus is not the functional correctness of the code, but how easy, fast, and pleasant it is
for a newcomer to get from "zero" to their first useful result — the famous
time-to-hello-world (TTHW).
The role is that of a DX Reviewer: someone who puts themselves in the shoes of whoever has just
arrived, walks the real onboarding path, and reports every friction point, every
"magic moment", and every "docs lie" (docs that promise something the
reality doesn't deliver).
Two modes
| Mode |
When |
What it does |
plan |
There's a plan/architecture but no running implementation yet |
Design-level review: estimates the expected TTHW, identifies anticipated friction and the "magic moments" the design should deliver. |
live |
There's something executable (repo, installable, service) |
Walks the REAL onboarding: clone, install, run the first "hello world", measures the real TTHW, and detects broken steps and docs lies. |
If --mode is not passed, infer it: if there's an executable artifact available → live;
if there's only a plan/spec/architecture → plan.
Three lenses (always apply all three)
- Expansion (what's missing) — gaps in the path: undocumented steps, hidden
prerequisites, configuration nobody explained, cases the onboarding ignores.
- Polish (what to refine) — things that work but scrape: unclear error messages,
confusing names, long commands, bad defaults, noisy output, copy that could be better.
- Triage (what's critical) — what blocks or drives away the adopter: what makes someone
quit before the first success. This is what gets fixed first.
Steps
plan mode
- Read the project's plan / architecture / spec (
spec.json, requirements,
architecture, design READMEs, mockups). Identify who the target adopter is
(internal dev? external integrator? technical end user?).
- Trace the expected path from "I discovered the project" to "I got my first
useful result". List each anticipated step: install, configure, authenticate,
first command/call, first output.
- Estimate the expected TTHW and mark where the design introduces unnecessary friction
(avoidable manual steps, heavy dependencies, prior configuration, secrets to
obtain).
- Identify the "magic moments" the design should produce — the instant when
the adopter says "ah, this actually works". Verify the plan delivers them early.
- Apply the three lenses (Expansion / Polish / Triage) over the design.
- Report prioritized findings with concrete recommendations (see format below).
live mode
- Start from true zero. Simulate a clean environment. If possible, use a
subagent (
Agent) with instructions to "assume nothing, just follow the documentation
literally" to walk the onboarding without the bias of someone who already knows the project.
- Walk the REAL onboarding following ONLY the documentation, step by step:
clone → install → configure → first "hello world". Time it from the start
to the first useful result → that's the real TTHW.
- Note every "docs lie": every time a command, path, variable
name, expected output, or prerequisite in the docs does not match reality.
Cite the doc, the command executed, and what actually happened.
- Record the broken steps: commands that fail, missing dependencies, implicit
undocumented steps, errors that require external knowledge to resolve.
- Locate the true "magic moment" and measure how much it costs to reach it. If it arrives
late or never, that's critical Triage.
- Apply the three lenses over the real experience lived.
- Report prioritized findings with recommendations (see format below).
Report format (both modes)
- DX verdict + TTHW (estimated in
plan, measured in live).
- Critical (Triage) — what blocks or scares off the adopter. Fix first.
- Friction (Polish) — what scrapes but doesn't block.
- Gaps (Expansion) — what's missing.
- Docs lies — only in
live: doc vs. reality, with citation.
- Magic moments — where they are and whether they arrive in time.
- Each finding with: location, impact on the adopter, and an actionable recommendation.
Reminders
- Does not advance the phase. Do not touch
spec.json:phase or mark anything as completed in the
method's flow. This skill only observes, measures, and recommends.
- It is read-only over the method's state: it can run onboarding commands in
live (install, run), but it does not edit the project's code or its spec.
- It can be run as many times as wanted, in any phase of the project.
Part of the Karvey™ Method — © HainTech, by Mauricio Quezada Ibáñez · Apache 2.0 · see karvey/LICENSE and karvey/TRADEMARK.md.
1---2name: karvey-devex3description: Developer-experience reviewer for the Karvey method. Walks the actual onboarding flow, measures time-to-hello-world, finds friction points and "docs lies". Plan-stage and live modes. Triggers include "karvey devex", "developer experience", "onboarding", "time to hello world", "fricción", "friction", "DX review".4---56# Karvey DevEx — DX Reviewer78> **CROSS-CUTTING SKILL of the Karvey™ Method.** This is a **support layer**, NOT a phase.9> It does NOT modify `spec.json:phase`. It does NOT advance the method's flow. It can be invoked at10> any time, as many times as needed, without altering the project's state.1112Inspired by gstack `/plan-devex-review` + `/devex-review`.1314## Purpose1516Measure and improve the **experience of whoever adopts or uses the project** (developer experience).17The focus is not the functional correctness of the code, but how easy, fast, and pleasant it is18for a newcomer to get from "zero" to their first useful result — the famous19**time-to-hello-world (TTHW)**.2021The role is that of a **DX Reviewer**: someone who puts themselves in the shoes of whoever has just22arrived, walks the real onboarding path, and reports every friction point, every23"magic moment", and every **"docs lie"** (docs that promise something the24reality doesn't deliver).2526### Two modes2728| Mode | When | What it does |29|------|------|--------------|30| **`plan`** | There's a plan/architecture but no running implementation yet | Design-level review: estimates the expected TTHW, identifies anticipated friction and the "magic moments" the design should deliver. |31| **`live`** | There's something executable (repo, installable, service) | Walks the REAL onboarding: clone, install, run the first "hello world", **measures the real TTHW**, and detects broken steps and docs lies. |3233If `--mode` is not passed, infer it: if there's an executable artifact available → `live`;34if there's only a plan/spec/architecture → `plan`.3536### Three lenses (always apply all three)37381. **Expansion (what's missing)** — gaps in the path: undocumented steps, hidden39 prerequisites, configuration nobody explained, cases the onboarding ignores.402. **Polish (what to refine)** — things that work but scrape: unclear error messages,41 confusing names, long commands, bad defaults, noisy output, copy that could be better.423. **Triage (what's critical)** — what blocks or drives away the adopter: what makes someone43 quit before the first success. This is what gets fixed first.4445## Steps4647### `plan` mode48491. **Read the project's plan / architecture / spec** (`spec.json`, `requirements`,50 `architecture`, design READMEs, mockups). Identify who the target adopter is51 (internal dev? external integrator? technical end user?).522. **Trace the expected path** from "I discovered the project" to "I got my first53 useful result". List each anticipated step: install, configure, authenticate,54 first command/call, first output.553. **Estimate the expected TTHW** and mark where the design introduces unnecessary friction56 (avoidable manual steps, heavy dependencies, prior configuration, secrets to57 obtain).584. **Identify the "magic moments"** the design should produce — the instant when59 the adopter says "ah, this actually works". Verify the plan delivers them early.605. **Apply the three lenses** (Expansion / Polish / Triage) over the design.616. **Report** prioritized findings with concrete recommendations (see format below).6263### `live` mode64651. **Start from true zero.** Simulate a clean environment. If possible, use a66 subagent (`Agent`) with instructions to "assume nothing, just follow the documentation67 literally" to walk the onboarding without the bias of someone who already knows the project.682. **Walk the REAL onboarding following ONLY the documentation**, step by step:69 clone → install → configure → first "hello world". **Time it** from the start70 to the first useful result → that's the **real TTHW**.713. **Note every "docs lie"**: every time a command, path, variable72 name, expected output, or prerequisite in the docs **does not match reality**.73 Cite the doc, the command executed, and what actually happened.744. **Record the broken steps**: commands that fail, missing dependencies, implicit75 undocumented steps, errors that require external knowledge to resolve.765. **Locate the true "magic moment"** and measure how much it costs to reach it. If it arrives77 late or never, that's critical Triage.786. **Apply the three lenses** over the real experience lived.797. **Report** prioritized findings with recommendations (see format below).8081### Report format (both modes)8283- **DX verdict** + **TTHW** (estimated in `plan`, measured in `live`).84- **Critical (Triage)** — what blocks or scares off the adopter. Fix first.85- **Friction (Polish)** — what scrapes but doesn't block.86- **Gaps (Expansion)** — what's missing.87- **Docs lies** — only in `live`: doc vs. reality, with citation.88- **Magic moments** — where they are and whether they arrive in time.89- Each finding with: location, impact on the adopter, and an actionable recommendation.9091## Reminders9293- **Does not advance the phase.** Do not touch `spec.json:phase` or mark anything as completed in the94 method's flow. This skill only observes, measures, and recommends.95- It is **read-only over the method's state**: it can run onboarding commands in96 `live` (install, run), but it does not edit the project's code or its spec.97- It can be run as many times as wanted, in any phase of the project.9899---100*Part of the Karvey™ Method — © HainTech, by Mauricio Quezada Ibáñez · Apache 2.0 · see `karvey/LICENSE` and `karvey/TRADEMARK.md`.*