# AI agent skills for Deadline Cloud
<a name="examples-skills"></a>

For background on working with AI agents, see [Using AI agents with Deadline Cloud](ai-agents.md). The [skills](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/skills) directory provides the following Deadline Cloud AI agent skills (`SKILL.md` files) that you can load into an agent so it follows a known-good procedure:
+ [deadline-cloud-job](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/skills/deadline-cloud-job) — Walks the agent through creating an Open Job Description job bundle for Deadline Cloud, optionally with a conda package recipe for custom dependencies. Covers designing parameters and steps, writing the template, validating with `openjd check`, iterating locally with `openjd run --tasks`, and submitting to a farm.
+ [conda-builder](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/skills/conda-builder) — Walks the agent through creating or updating a [rattler-build](https://prefix-dev.github.io/rattler-build/) conda recipe for a DCC, mimicking the structure of existing recipes in the samples repository. Recipes build inside an Amazon Linux 2023 (AL2023) Docker container to match the GLIBC version on service-managed fleet workers, and the agent indexes the result into a local conda channel for testing. The skill currently covers Blender. Per-DCC guides are in the `dcc/` subdirectory.
+ [host-config-from-installer](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/skills/host-config-from-installer) — Walks a customer through installing any Windows `.exe` software locally, verifying it works, uploading the installer to Amazon Simple Storage Service (Amazon S3), and converting the verified install steps into a PowerShell host configuration script for Windows service-managed fleet workers. Use it when the software isn't covered by a more specific skill (for example, the 3ds Max host configuration).