DD Agent Skills

Portable, agent-agnostic skills for AI coding assistants. All skills use the dd- prefix convention.

Works with: Claude Code, Codex, Cursor, Hermes, GitHub Copilot — zero agent-specific tool calls.

Available Skills

dd-webapp-full-test

Comprehensive web application testing across 6 dimensions. Tell your agent "test my app thoroughly" and get a complete audit.

Step	Output
1. Pre-flight	Detects framework, build tool, installed tools, missing deps
2. Lock dimensions	User confirms which 6 dimensions run, which skip
3. Execution plan	Detailed per-dimension plan — exactly what gets tested, user approves
4. Execute	All dimensions run in order: functional → visual → a11y → security → compatibility → performance
5. Report	TEST-REPORT.md at project root with pass/fail, issues, recommendations
6. Cleanup	Temporary artifacts removed; scripts + report kept

#	Dimension	What it tests
1	Functional	Unit tests, component tests, E2E flows, API contract validation
2	Visual	Screenshot regression, dark mode, responsive breakpoints (24 scenarios)
3	Accessibility	axe-core scans, keyboard navigation, focus management, ARIA labels
4	Security	HTML output safety, input validation, header checks, sensitive data leaks
5	Compatibility	Multi-browser, offline/slow network, multi-language rendering
6	Performance	Core Web Vitals, bundle analysis, API latency, CLS, Long Tasks, cache

Provider note: Some LLM providers or enterprise gateways may flag security test content. The skill handles this gracefully — skips affected tests, notes it in the report, and continues normally.

Stack: Playwright + Vitest + React Testing Library + axe-core.

Files: dd-webapp-full-test/SKILL.md + 3 pre-flight scripts + 4 reference checklists.

Critical Rules (agent-enforced)

The skill instructs agents to follow three hard rules:

1. Never modify business code — write tests to e2e/, write results to TEST-REPORT.md. Never touch application logic, components, styles, or config. All issues go into the report; the user decides what to fix.
2. Test results are facts — PASS, FAIL, or SKIP only. No "this is fine" judgments.
3. Ask before installing — never npm install without explicit user confirmation.

Pre-flight Scripts

Three fallback scripts auto-detect your stack. Agent tries each in order:

Script	Runtime	When
scripts/preflight.mjs	Node.js	Always works (Playwright needs Node)
scripts/preflight.py	Python 3	Fallback
scripts/preflight.sh	Bash	Unix/macOS fallback (timeout-safe)

Output: JSON with framework, build tool, installed tools, missing dependencies.

Supported Frameworks

Auto-detected and pattern-adapted: React, Next.js, Vue 2/3, Angular, Svelte, SolidJS, Vanilla HTML/JS.

---

Installation

Install via Agent

Tell your agent:

"Install dd-webapp-full-test from https://github.com/dojeee/agent-skills"

The agent should execute:

# Clone the repo
git clone https://github.com/dojeee/agent-skills /tmp/agent-skills

# Copy to the correct location for YOUR agent:
# Claude Code (per-project):
mkdir -p .claude/skills
cp -r /tmp/agent-skills/dd-webapp-full-test .claude/skills/

# Hermes (per-user):
mkdir -p ~/.hermes/skills
cp -r /tmp/agent-skills/dd-webapp-full-test ~/.hermes/skills/

# Codex / Cursor (per-project):
mkdir -p .codex/skills
cp -r /tmp/agent-skills/dd-webapp-full-test .codex/skills/

# Verify
ls .claude/skills/dd-webapp-full-test/SKILL.md  # or your agent's path
ls .claude/skills/dd-webapp-full-test/references/

# Cleanup
rm -rf /tmp/agent-skills

Agent-Specific Paths

Agent	Skills Directory	Discovery
Claude Code	<project>/.claude/skills/	Referenced from CLAUDE.md
Hermes	~/.hermes/skills/	Auto-discovered by triggers
Codex / Cursor	<project>/.codex/skills/	Referenced from AGENTS.md
GitHub Copilot	<project>/.github/copilot/	.github/copilot-instructions.md

Verify Installation

cat .claude/skills/dd-webapp-full-test/SKILL.md | head -6
# Should show:
# ---
# name: dd-webapp-full-test
# description: |
#   Comprehensive web app testing across 6 dimensions...

Post-Install — Reference in Project Config

After installing, the agent will ask whether to add a reference to CLAUDE.md or AGENTS.md. If accepted, it appends:

## Testing

Run comprehensive tests with the dd-webapp-full-test skill:

- "test my app thoroughly" — runs 6-dimension audit (functional, visual, a11y, security, compatibility, performance)
- "run all 6 test dimensions" — same
- "comprehensive test" — same
- "security scan" — dimension 4 only

Skill location: `.claude/skills/dd-webapp-full-test/SKILL.md` (or your agent's skill directory).
Tests are written to `e2e/`. Results in `TEST-REPORT.md`.

Naming Convention

All skills use the dd- prefix:

dd-<skill-name>/
├── SKILL.md              # Main instructions: patterns, code templates, pitfalls
├── scripts/              # Pre-flight checks, helpers
└── references/           # Detailed checklists, payload lists, metric definitions

Contributing

Guidelines:

• Prefix with dd-
• Pure English (code examples may contain non-English content for testing)
• No agent-specific tool calls
• Include a "Pitfalls" section with common mistakes
• Include a "CRITICAL RULES" section that forbids agents from modifying business code
• Test your skill against a deliberately buggy app to verify it catches issues
• Be mindful of LLM content filters — avoid putting raw security payloads in the main SKILL.md; use separate data/ files instead (not auto-loaded by agents)

License

MIT