Skills not loaded when prompt is passed via -p

[nsgundy]

Describe the bug

I have installed the playwright-cli skill. The skill is not used when I use -p to execute the prompt:

copilot --model gpt-5-mini --allow-all-tools -p "take screenshot of github.com"

Affected version

0.0.398

Steps to reproduce the behavior

1. Install playwright-cli skill from https://github.com/microsoft/playwright-cli using the /plugin method
2. In an empty directory (i.e. /tmp/test) run

copilot --model gpt-5-mini --allow-all-tools -i "take screenshot of github.com"

Output

Fetching GitHub homepage HTML so a best-effort textual capture can be provided (no headless browser available here); calling web fetch and reporting intent in parallel.

● web_fetch
  └ Here is the raw content:

Running a headless browser to capture a full-page PNG of https://github.com because it produces an accurate screenshot.

● Take screenshot of github.com using headless browser
  $ BROWSER="$(command -v chromium-browser || command -v chromium || command -v google-chrome || 
  command -v google-chrome-stable || true)"
  if [ -n "$BROWSER" ]; then
    "$BROWSER" --headless --disable-gpu --no-sandbox --screenshot=/tmp/github_screenshot.png 
  --window-size=1280,1024 "https://github.com" || true...
   └ 36 lines...

read_bash
   └ 31 lines...

stop_bash
   └ 1 line...

● Check screenshot file exists
  $ ls -l /tmp/github_screenshot.png || true
   └ 2 lines...

Tried taking a screenshot but no headless browser produced an image; environment likely lacks a Chrome/Chromium binary or necessary system services. If a screenshot is still needed, allow use of an alternative method (e.g., using an external screenshot API or grant access to a system with a browser) and specify preferred output filename.

Expected behavior

I expect to see the same behavior as when running in interactive mode where the skill gets loaded as expected (same cwd as above).

copilot --model gpt-5-mini --allow-all-tools -i "take screenshot of github.com"

Output

❯ take screenshot of github.com

◐ **Clarifying tool invocation**

  I’m working through how to effectively use the report_intent function alongside multi_tool_use.parallel, which only allows tool 
  functions. It’s important to follow the guideline that says the report_intent must come first in tool calls. I also need to 
  include a brief explanation before invoking the tools, making sure not to use "I will" statements. So, I’ll need to craft a 
  concise pre-tool message that explains my actions clearly while adhering to the limitations.**Setting up tool calls**

  I need to initiate multi_tool_use.parallel by first calling functions.report_intent with a brief intent string like "Taking 
  screenshot." Next, I’ll use functions.skill with the skill "playwright-cli." However, I must consider that this skill may require
   additional context later, as it could ask for details like the URL or output location. I’ll start with a simple pre-tool 
  message, then call the tools and handle the responses from the skill agent as needed. This should be straightforward!

● skill(playwright-cli)

● Open GitHub and take screenshot
  $ playwright-cli open https://github.com && playwright-cli resize 1280 800 && playwright-cli screenshot && ls -1t *.png | head -n1
   └ 12 lines...

● Screenshot saved to .playwright-cli/page-2026-01-29T11-52-46-390Z.png.

Additional context

Executed in a Docker development container based on Ubuntu 20.04 with Nodejs 18 installed.