Agent
Entrypoint
Entrypoint is a bash script that runs inside the agent's sandbox after secrets load and before the runtime boots. Use it to prepare the environment the agent needs before it starts working.
How to use entrypoint
- 01Open the agent's Entrypoint area.
- 02Write the bash setup script.
- 03Save your changes.
- 04Run a task and confirm the runtime has the expected dependencies.
What entrypoint is for
- Install dependencies: Run package managers, install system packages, or build generated assets.
- Clone or prepare code: Fetch a repository, check out a branch, or create folders the agent expects.
- Configure the sandbox: Write config files, set permissions, or warm caches before the harness starts.
Write an entrypoint
Keep the script idempotent: it should be safe to run more than once. In the UI, leaving the editor empty means the agent has no entrypoint. In the CLI manifest, set exactly one of entrypoint.file, entrypoint.commands, or entrypoint.text.
set -euo pipefail
if [ ! -d repo ]; then
git clone https://github.com/acme/internal-tools.git repo
fi
cd repo
npm ci
npm run build
Use secrets
Secrets are available before the entrypoint runs, so the script can use them for package registries, private repositories, or setup commands. Read them from environment variables and avoid printing them to logs.
set -euo pipefail
git clone "https://oauth2:${GITHUB_TOKEN}@github.com/acme/private-repo.git" repo
cd repo
npm ci
Best practices
- Keep setup fast; slow entrypoints delay every task startup.
- Make commands repeatable and safe if the sandbox already has the file or package.
- Use Secrets for tokens and private configuration.
- Avoid destructive commands unless the agent needs a clean environment every time.
- Prefer pinned package versions for predictable behavior.