Index of every page on this site: llms.txt
All skills
anthropics avatar

/first-run

@4a4224b
by Anthropicanthropics/claude-quickstarts21k stars
3,501
Run it yourself

Guide a new user through their first computer-use agent run (env check, safe browser-only task, then open the trajectory viewer).

Use this Skill: https://skilld.dev/gh/anthropics/claude-quickstarts/first-run

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈35 tokens for metadata: the name and description. ≈785 when used: this file.

Description uses 1.8% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈141 characters, or 1.8%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

First run

Walk the user through running this demo for the first time. Be conversational and pause for confirmation between phases; the goal is that they understand what each step does, not just that the commands succeed.

0. Orient

Briefly say what is about to happen: you will check their setup, run one safe browser-only task that does not touch their mouse or keyboard, and then open the trajectory viewer so they can see what the model saw.

1. Environment check

Never read the contents of .env. Do not cat, grep (without -q), Read, or otherwise display it — the user's API key must never appear in this conversation or in your context. Only check for its presence.

  • Confirm a virtual environment is active and dependencies are installed: python -c "import computer_use, playwright; print('ok')". If that import fails, point them at the Installation section of README.md and stop.
  • Confirm .env exists and the API key is set without ever printing it: [ -f .env ] && grep -q '^ANTHROPIC_API_KEY=.' .env && echo 'API key: set' || echo 'API key: missing' If missing, copy .env.example to .env and ask the user to open it in their own editor and paste their key into it (do not ask them to paste the key into chat, and do not edit .env for them). Wait until they confirm.
  • Run the macOS permission preflight non-fatally: python -c "from computer_use.preflight import check_and_warn; check_and_warn(require=False)" and explain any warnings. These permissions are only needed for the desktop computer tool, which the first task does not use, so it is fine to continue without them for now.

2. Run a safe browser-only task

Explain that this task uses only the headless browser tool, so nothing on their screen will move. Then run:

CU_ENABLE_COMPUTER_USE_TOOLS=false python -m computer_use \
  "Navigate to https://en.wikipedia.org/wiki/Special:Random three times, \
   take a screenshot each time, then summarize what you saw."

While it runs, narrate what the streamed output means (dim thinking text, the cyan tool-call lines, the yellow tool-result lines, the [usage] line). When it finishes, point out the trajectory: runs/... path that was printed.

3. Open the trajectory viewer

python -m streamlit run dev_ui/trajectory_viewer/app.py

Tell them to pick the newest run in the sidebar, expand the System prompt section to see exactly what the model was told, and scroll through the turns to see each screenshot inline.

4. Next steps

Offer two follow-ups and let them choose:

  • Try the desktop tool: once Screen Recording and Accessibility are granted, run python -m computer_use "open TextEdit and type hello world". Warn them again that this will move their real mouse and keyboard.
  • Explore the dev UIs: the tool panel at python -m uvicorn dev_ui.tool_panel.server:app --reload lets them call each tool by hand without the model in the loop.

End by pointing at the Configuration and Effective caching and context pruning sections of README.md for the deeper material.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning2mo3 checks · Risk SAFE
  • Gen Agent Trust Hub2mo

    This skill provides a guided introduction for new users, focusing on environment setup and a safe initial task. It includes proactive security practices, such as explicitly protecting API keys from exposure and restricting the agent's capabilities during the first run.

  • Socket2mo

    No alerts

  • Snyk2mo

    Risk: MEDIUM · 1 issue

Signed by skilld at 4a4224b. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last week.

Activeupdated 6 months ago

README badge

README badge for anthropics/claude-quickstarts/first-run