wordpress-development skill

A single skill for WordPress development, with Skillsmith-based local evals.

Skill structure

The skill lives under skills/wordpress-development/. It uses a single-entry-point layout: skills/wordpress-development/SKILL.md is the root document that routes the model to topic-specific reference files. Currently the main topic is the Interactivity API, covered by skills/wordpress-development/references/interactivity-api.md and five sub-references under skills/wordpress-development/references/interactivity-api/.

Prerequisites

• Node ≥ 20.17
• Docker — required by @wordpress/env to boot a local WordPress instance during evals
• Dependencies — run npm install from the repo root; the postinstall script automatically installs the Playwright Chromium browser

Environment setup

Copy .env.example to .env and fill in the keys you need. The default Skillsmith configuration uses only claude-code-provider agents, so a logged-in Claude Code install needs no API keys. If you want to test other providers, .env.example documents ANTHROPIC_API_KEY, OPENAI_API_KEY, and GOOGLE_GENERATIVE_AI_API_KEY.

Skillsmith auto-loads .env from the directory you invoke it in (using Node's built-in process.loadEnvFile). Shell environment variables take precedence over .env.

Running evals

Single scenario (the normal dev workflow)

npx skillsmith <scenario-dir>
# equivalently
npm run skillsmith -- <scenario-dir>

Scenario names are the bare directory names under eval/scenarios/ (e.g. counter, async-fetch). For example:

npx skillsmith counter

This runs one scenario against the single configured testing agent and writes results under .skillsmith/<runId>/.

Full matrix (owner's manual step only)

npm run skillsmith

Running the full scenarios × testing-agents matrix is expensive and is not an agent task. See the cap below.

Report location

All run output lands under .skillsmith/<runId>/:

• Per-agent report: iteration-N/<scenario>/<agent>/report.json
• Per-scenario report: iteration-N/<scenario>/report.json
• Run-level summary: report.json, run.json, summary.txt
• Playwright results: iteration-N/tests-report.json

Agent cap (R12)

Agents must never run the full eval matrix; an agent runs at most one scenario against one testing agent, and the full matrix is the owner's manual step. Always pass a single scenario directory as a positional argument so the run stays capped:

npx skillsmith <one-scenario-dir>

Running npm run skillsmith with no argument executes the entire scenarios × testing-agents matrix. That is expensive and is reserved for the owner to run by hand after setup — it is never an agent task.