chore(skills): move agent skills under playwright-core, add init-skills

[pavelfeldman]

Summary

• Move all agent skills (playwright-cli, component testing, trace) to packages/playwright-core/src/tools/skills/<skill-name>/, one build copy rule ships them all.
• Share a single installSkills() helper across playwright cli install --skills, playwright trace install-skill and the new command.
• Add npx playwright init-skills [--loop claude|agents] next to init-agents — installs all Playwright skills.

[github-actions]

Test results for "MCP"

4 failed
❌ [chrome] › mcp/cli-navigation.spec.ts:56 › goto chrome:// page that closes the tab does not crash the response @mcp-windows-latest-chrome
❌ [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate @mcp-macos-latest-chromium
❌ [chromium] › mcp/dashboard.spec.ts:125 › daemon show: closing page exits the process @mcp-macos-latest-chromium
❌ [firefox] › mcp/annotate.spec.ts:230 › should capture annotations via show --annotate @mcp-ubuntu-latest-firefox

7782 passed, 1249 skipped

---

Merge workflow run.

[github-actions]

🟢 CI is clear — all four failures are pre-existing MCP flakes

Hi, I'm the Playwright bot and I took a first look at the CI failures here.

The MCP report has 4 failed tests, but none are caused by this PR — each one flips verdict across many unrelated runs in the aggregated CI results, and this PR only relocates skill files, tweaks help text, and adds the init-skills path. The daemon/annotate/dashboard runtime these tests exercise isn't touched.

Details

Overall: this PR (5e02ce8) moves agent skills under playwright-core/.../tools/skills/, shares an installSkills() helper, and adds npx playwright init-skills. The daemon program.ts change is confined to initWorkspace (the --skills/init-skills flow) and the help string; core daemon, annotate, navigation, and dashboard behavior is unchanged. None of the four failing tests touch skill installation.

Pre-existing flake / infra — all four, with history from the test-results DB (final verdict after retries, on SHAs/PRs unrelated to this PR):

• [firefox] › mcp/annotate.spec.ts:230 › should capture annotations via show --annotate — bimodal: firefox failed 10 of 313 runs, passed 303 (plus 5 within-run rescues), and it also flakes on chromium/chrome/webkit. Not PR-related.
• [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate — bimodal: chromium failed 5 of 318 runs, passed 313 (6 rescues); flakes across firefox/webkit/chrome too.
• [chromium] › mcp/dashboard.spec.ts:125 › daemon show: closing page exits the process — bimodal: chromium failed 7 of 318 runs, passed 311 (3 rescues); also flakes on firefox/webkit.
• [chrome] › mcp/cli-navigation.spec.ts:56 › goto chrome:// page that closes the tab does not crash the response — bimodal: chrome failed 5 of 289 runs, passed 284 (5 rescues), and never failed on chromium.

The failures also spread across four different tests on four different browser/OS combos (chrome-windows, chromium-macos ×2, firefox-ubuntu) with no shared signature tied to the diff — consistent with flake, not a regression.

_{Triaged by the Playwright bot - agent run}

[github-actions]

Test results for "tests 1"

2 flaky

⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [playwright-test] › ui-mode-test-watch.spec.ts:185 › should watch new file `@windows-latest-node22`

49990 passed, 1189 skipped

---

Merge workflow run.