Use when adding or editing QA skills in seed-skills/ or getting them onto the live qaskills.sh catalog, e.g. "add N new skills", "create a seed skill for X", "seed the database", "the skill page is empty", "skill 404s on the site".
npx skills add https://github.com/PramodDutta/qaskills --skill add-seed-skills
Pipeline: seed-skills/<slug>/SKILL.md -> validator -> seed.ts upsert into the PROD database -> live catalog. These are PRODUCT catalog skills (Zod schema with testingTypes/languages), not Claude Code workflow skills; never confuse the two formats.
packages/web/src/db/seed.ts parses frontmatter with regex, not a YAML library:
tags: [a, b, c]. YAML block lists (- item) parse as EMPTY arraystestingTypes >= 1, languages >= 1packages/shared/src/constants/ (testing types, frameworks, languages, domains, agent slugs); check there before inventing oneTemplate (model the body on seed-skills/playwright-e2e/SKILL.md):
---
name: Human Readable Skill Name
description: One line, 10-500 chars, what the skill teaches an agent to do.
version: 1.0.0
author: thetestingacademy
license: MIT
tags: [tag-one, tag-two]
testingTypes: [e2e]
frameworks: [playwright]
languages: [typescript]
domains: [web]
agents: [claude-code, cursor, github-copilot, windsurf, cline]
---
# Skill Title
Real instructions: principles, project structure, code samples, checklists.
This body becomes fullDescription: it renders on the skill page and is what
the CLI downloads. Aim for 100+ lines of substance; an empty or thin body
is a broken product page.
ls /Users/promode/qaskills/seed-skills | grep -i "<core-term>"
curl -s -o /dev/null -w '%{http_code}\n' "https://qaskills.sh/api/skills/<slug>" # want 404
seed-skills/<slug>/SKILL.mdFrontmatter per the contract above, then the full body.
cd /Users/promode/qaskills
pnpm --filter @qaskills/shared build && pnpm --filter @qaskills/skill-validator build
for d in <slug-one> <slug-two>; do
node packages/skill-validator/dist/cli.js "seed-skills/$d/SKILL.md" || echo "INVALID: $d"
done
Also eyeball that arrays are inline and the description is one line (the validator uses the real YAML parser and can pass files the seed regex still mangles).
Never use the DATABASE_URL from .env.local. It points at a stale non-prod database; seeding it changes nothing on the live site. This mistake has burned a full session before.
curl -s 'https://qaskills.sh/api/skills?limit=1' and record the top-level total.vercel env pull .env.vercel-prod --environment=production yourself (worked from the agent as of 2026-07-07; the file is gitignored). If the pull is blocked, ask the user to run that exact command and wait.cd /Users/promode/qaskills
export DATABASE_URL='<prod-url-without-quotes>'
pnpm --filter @qaskills/web db:seed
seed.ts is an upsert (onConflictDoUpdate on skills): safe to re-run, it will not delete the live rows that exist only in prod. Never substitute custom SQL, and never run UPDATE/DELETE against prod without explicit user approval.
curl -s 'https://qaskills.sh/api/skills?limit=1' # total must equal baseline + N
curl -s "https://qaskills.sh/api/skills/<slug>/content" | head -20 # frontmatter + body, not empty
curl -s -o /dev/null -w '%{http_code}\n' "https://qaskills.sh/skills/thetestingacademy/<slug>" # 200 (adjust author segment if different)
total unchanged, or slug 404s => you seeded the wrong database. STOP. Do not retry blindly; re-verify which URL was exported and report the mismatch.
git -C /Users/promode/qaskills add seed-skills/<slug-one> seed-skills/<slug-two>
git -C /Users/promode/qaskills commit -m "feat(skills): add <N> <theme> seed skills"
git -C /Users/promode/qaskills push origin main
No web redeploy is needed for catalog changes (pages read the DB), but commit so the repo stays the source of truth.
| Symptom | Cause | Fix |
|---|---|---|
| Skill live but tags/types empty | Block-list arrays or multi-line values in frontmatter | Convert to inline arrays, single lines, re-seed |
| Skill page renders only the short description | Missing/empty markdown body | Write the body, re-seed (upsert refreshes fullDescription) |
| Live total did not grow | Seeded the stale .env.local DB | Step 4.2, re-seed with the real prod URL |
| Validator passes but seed drops fields | Validator parses real YAML, seed.ts regex does not | Obey the format contract, not just the validator |
| Connection error on seed | Quoted URL or Node 24 | Strip quotes; use Node 20 |
.env.local "because it is right there"total comparisonMulti-agent autonomous startup system for Claude Code. Triggers on "Loki Mode". Orchestrates 100+ specialized agents across engineering, QA, DevOps, security, data/ML, business operations, marketing, HR, and customer success. Takes PRD to fully deployed, revenue-generating product with zero human intervention. Features Task tool for subagent dispatch, parallel code review with 3 specialized reviewers, severity-based issue triage, distributed task queue with dead letter handling, automatic deployment to cloud providers, A/B testing, customer feedback loops, incident response, circuit breakers, and self-healing. Handles rate limits via distributed state checkpoints and auto-resume with exponential backoff. Requires --dangerously-skip-permissions flag.
Use when working with error debugging multi agent review
Build evaluation frameworks for agent systems. Use when testing agent performance systematically, validating context engineering choices, or measuring improvements over time.
Diagnoses and debugs A2A agent communication issues including agent status, message routing, transport connectivity, and log analysis. Use when agents aren't responding, messages aren't being delivered, routing is incorrect, or when debugging orchestrator, coder-agent, tester-agent communication problems.
Use when working with error debugging multi agent review
Rapidly creates atomic, focused skills optimized with evidence-based prompting, specialist agents, and systematic testing. Each micro-skill does one thing exceptionally well using self-consistency, program-of-thought, and plan-and-solve patterns. Enhanced with agent-creator principles and functionality-audit validation. Perfect for building composable workflow components.
Ultimate multi-agent framework for Google Antigravity. Orchestrates specialized domain agents (PM, Frontend, Backend, Mobile, QA, Debug) via Serena Memory.
This skill should be used when the user asks to "evaluate agent performance", "build test framework", "measure agent quality", "create evaluation rubrics", or mentions LLM-as-judge, multi-dimensional evaluation, agent testing, or quality gates for agent pipelines.
Take pramoddutta/add-seed-skills from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.