TRAIL COORDINATES 47.62N · 122.33W
FIELD NOTES · 01/10
ANALOGY
VOCABULARY
OUR STORY
SKILLS WE USE
LIVE DEMO
EVERYDAY AI CODING
Skills & Plugins
for Testers
for Testers
A trail map through AI-assisted QA — written once by someone who's been here before, followed by everyone after.
Maneesh Maddala
MINISTRY OF TESTING — AI CHAPTER
30 JULY 2026
30 JULY 2026
SECTION 02
FIELD NOTES · 02/10
AGENDA
Where We're Headed
01
The Analogy
02
The Vocabulary
03
The Problem
04
Our Story
05
Skills We Use
06
Live Demo
FIELD NOTES · 03/10
THE ANALOGY
Two people step into
a forest, same moment.
a forest, same moment.
One has a trail map. The other has instinct, and figures it out step by step. Same forest, same start — very different afternoon.
A Skill is like the trail map.
Written once, followed every time.
Your team never starts from zero.
Your team never starts from zero.
INSTINCT
TRAIL MAP
FIELD NOTES · 04/10
THE VOCABULARY
Two ways to say the same thing.
TERMKITCHENCODE
/command
Recipe card — pulled out on demand
function call — explicit, returns a result
skill
Chef's training — always shapes every dish
auto-imported utility — always in scope
plugin
The whole kitchen — station, staff, recipes
an npm package — bundles + runs setup hooks
MCP server
The delivery dock — live ingredients, real time
an external API your agent calls at runtime
agent
The head chef — plans the menu, runs the line, adapts as it goes
an autonomous process — plans, uses tools, runs to done
FIELD NOTES · 05/10
THE VOCABULARY
Same idea, more of it.
SKILL
One tool. One job.
A SKILL.md, plus whatever references, assets, or scripts it needs — all built to do one task well.
e.g. grill-me
PLUGIN
One box. Many tools.
A bundle of skills (and other pieces) packaged together — installed as a single unit.
e.g. Superpowers — 14 skills bundled
FIELD NOTES · 06/10
THE PROBLEM
Every engineer was re-explaining
the same things to Claude.
the same things to Claude.
Naming conventions. Retry strategy. Definition of done. Slightly different, every single time.
FIELD NOTE
This is a solved problem.
Before building a Skill from scratch — check if the community already solved it. Same instinct as not writing your own assertion library.
CHECK
FIRST
FIRST
FIELD NOTES · 07/10
OUR STORY
The pipeline Agent we built
ASANA
JIRA
FIGMA
TESTRAIL
PLAYWRIGHT
PR REVIEW
“The most powerful skill you'll ever install is the one you wrote yourself.”
WHAT WE TRIED
One skill to run the whole pipeline end to end — no human in the loop.
WHERE IT BROKE
No guardrails — it went wild: random test cases, automation nobody asked for.
WHAT WE CHANGED
Trained it on our conventions — naming, locators, structure — plus guardrails and evals.
WHAT IT DOES NOW
Runs the pipeline semi-autonomously. Full-day of work → about 2 hours.
FIELD NOTES · 08/10
SKILLS WE USE
What's actually in our toolbox
PLUGIN
Superpowers
TDD discipline + review gate — 14 skills bundled
SKILL
grill-me
Interviews before building, one question at a time
PLUGIN
Ponytail
YAGNI ladder — stops over-abstracted code
SKILL
Graphify
Codebase → knowledge graph for risk-based scoping
SKILL
Claude Code setup
Our own onboarding conventions for every new repo
SKILL
find-skills
Discovers relevant public Skills — you're not searching blind
PLUGIN
claude-mem
Persists context across sessions — no starting from zero
Skill = SKILL.md + its references, assets, and scripts.
Plugin = a bundle, installed together.
Plugin = a bundle, installed together.
FIELD NOTES · 09/10
LIVE DEMO
One repo with Skills. One without.
“Build a Playwright test framework for the Sauce Demo app covering login, cart, checkout, and error handling — 8 to 10 tests across the critical flows.”
Same prompt, both repos — deliberately generic, structure left to each repo's own judgment.
repo-a — no skills
spec/
└─ everything.spec.ts // one giant file
// login repeated inline, no fixture
// asserts value exists, not that it's correct
repo-b — grill-me installed
pages/
tests/
fixtures/
// interviews first — structure, conventions
// assertions reflect your answers
FIELD NOTES · 10/10
END OF TRAIL
“Skills are how you stop explaining yourself to AI and start directing it.”
Q & A
Maneesh Maddala
MINISTRY OF TESTING — AI CHAPTER
30 JULY 2026
30 JULY 2026