fix(skills): narrow npm and npx grants

[Devin Rousso (dcrousso)]

the CLI and trace skills preapprove unrelated npm and npx commands

limit automatic approval to Playwright commands

fix #42745

[github-actions]

Test results for "MCP"

2 failed
❌ [chrome] › mcp/cli-webmcp.spec.ts:188 › webmcp-call disambiguates same-name tools in identical same-origin frames @mcp-macos-latest-chrome
❌ [chrome] › mcp/cli-webmcp.spec.ts:162 › webmcp-call disambiguates duplicate tool names by frame @mcp-ubuntu-latest-chrome

8690 passed, 1474 skipped

---

Merge workflow run.

[github-actions]

🟢 Both failures are a pre-existing WebMCP flake — this PR is clear

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

The two mcp/cli-webmcp.spec.ts failures also fail on plain main and on other PRs, so they're not on you. This PR only edits allowed-tools: frontmatter in two SKILL.md files plus the two tests that assert on it — nothing that the WebMCP frame-registration path touches.

Details

Pre-existing flake / infra

• [chrome] › mcp/cli-webmcp.spec.ts:188 › webmcp-call disambiguates same-name tools in identical same-origin frames and [chrome] › mcp/cli-webmcp.spec.ts:162 › webmcp-call disambiguates duplicate tool names by frame — same story, one flake. Both flip verdict across runs on SHAs this PR can't be responsible for:

  where	sha	which test
  push on main	07f1a615 (this PR's base)	same-name tools, mcp-ubuntu-latest-chrome
  #42788	1f166c07	same-name tools, mcp-ubuntu-latest-chromium
  #42785	fabe18a1	same-name tools, mcp-macos-latest-chrome

  Run history for the two bots that went red here (oldest first, one square per run):

  same-name tools on mcp-macos-latest-chrome
  🟩🟩🟩🟩🟩🟩🟩🟥🟩🟩🟩🟩🟥

  duplicate tool names on mcp-ubuntu-latest-chrome
  🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟥

  The failure signature is the same race every time — the second frame's tool registration hasn't landed when the assertion runs, so the tool list is short:

  Expected substring: "- frame: http://localhost:8907/widget.html (frame 1)"
  Received string:    "### Result
  Found 1 WebMCP tool(s). ..."
  

  and in the duplicate tool names case the call succeeds instead of reporting the ambiguity, because only one frame had registered echo by then. Both are the same underlying timing problem, one bot each.

Why the diff can't reach it

The four changed files are playwright-cli/SKILL.md and playwright-trace/SKILL.md (only the allowed-tools: line and one sentence of prose), plus assertions in tests/mcp/cli-misc.spec.ts and tests/mcp/init-agents.spec.ts. No product code, and nothing shared with the WebMCP tool-registration path under test.

I'm a first pass, so treat this as a strong lean rather than the last word — but the same test going red on the base main SHA is about as clean as this evidence gets.

_{Triaged by the Playwright bot - agent run}