Agents & Skills
Skills vs Agents
| Skills | Agents | |
|---|---|---|
| What | Orchestration scripts | Pure building blocks |
| Invoked by | User (slash command or prompt) | Skills only |
| Invokes | Other skills + agents | Tools (browser, file, bash) |
| Example | /e2e-automate | @playwright-test-planner |
Agents never invoke each other. Skills are the orchestrators — they pre-read files and pass content inline to agents.
Skills (8)
| Skill | Invoked as | What it does |
|---|---|---|
e2e-automate | /e2e-automate | Full 10-phase pipeline |
e2e-plan | /e2e-plan <url> | Phases 4–5 only (explore + validate) |
e2e-generate | /e2e-generate <plan-file> | Phase 7 only (BDD generation) |
e2e-heal | /e2e-heal @ModuleName | Phase 8 only (run + fix failures) |
e2e-run | /e2e-run | Run existing tests without healing |
e2e-validate | /e2e-validate | Validate generated feature files |
e2e-process | /e2e-process <jira-url-or-file> | Phase 3 only — convert Jira tickets or files to a parsed test plan |
e2e-desktop-automate | Natural language in Claude Desktop | Full pipeline via MCP tools |
Agents (9)
@playwright-test-planner
The exploration agent. Opens a real browser, navigates your app, and builds the test plan.
- Tools:
browser_navigate,browser_snapshot,browser_click,browser_type - Output:
seed.spec.js,{module}-plan.md, plannerMEMORY.md - Budget: 20 browser calls max (5 in verification mode when memory is fresh)
@explorer
Combines authentication + browser exploration into a single agent. Authenticates via the configured auth strategy, then navigates the app, discovers UI elements, and writes the seed file and test plan.
@bdd-generator
Converts the test plan into Gherkin .feature files using 3-column data tables (| Field | Value | Type |). Does not write Playwright implementation code — that is @code-generator's job.
@code-generator
Reads the .feature file and the seed file, then fills in the Playwright implementation inside steps.js. Uses processDataTable() + validateExpectations() from stepHelpers.js.
@playwright-test-healer
Diagnoses and fixes failing tests — greps source for broken selectors, applies targeted fixes, retries up to 3 iterations.
@playwright-test-generator
Creates automated browser tests by executing test plan steps live via Playwright MCP tools, reading the recorded action log, and writing a working seed.spec.js.
@execution-manager
Runs the correct Playwright command based on module category and tags:
| Category / Tag | Projects |
|---|---|
@Workflows | --project setup --project precondition --project workflow-consumers |
@authentication in module name | --project auth-tests |
@serial-execution tag in feature file | --project setup --project serial-execution |
All other @Modules | --project setup --project main-e2e |
@input-processor
Handles Phase 3 input conversion — converts Excel, CSV, PDF, JSON, Word, PowerPoint, and HTML files to parsed test plan markdown via the Markitdown MCP.
@jira-processor
Fetches Jira requirements via the Atlassian MCP, analyzes completeness, asks targeted clarifying questions, and outputs refined requirements as a test plan.