Cómo usar
- Copia el comando y revisa el paquete antes de instalarlo en tu entorno de agente.
- Consulta las herramientas, credenciales y entornos necesarios en el documento original.
- Describe las entradas y el resultado deseado; contrasta la salida con las fuentes.
npx skills add https://github.com/aws/agent-toolkit-for-aws --skill managing-glue-jobsAntes de empezar
- La instalación no incluye suscripción al modelo, cuota API ni permisos sobre tus archivos.
- Las estrellas e instalaciones reflejan actividad, no una prueba independiente de calidad o seguridad.
Contenido del repositorio de origen con títulos, ejemplos, código, tablas, enlaces e imágenes preservados.
AWS Glue auto upgrade
This skill follows Automatic version upgrades for AWS Glue jobs in the AWS Glue User Guide.
Journey
Each step names the file to read. Stop marks a point where you wait for the user's decision.
| Step | What happens | Read | Stop |
|---|---|---|---|
| 1. Assess | You check which jobs can use auto upgrade. Read-only. | assess-and-turn-on.md | — |
| 2. Safe to retry | The user confirms that each job is safe to retry, and gives an email address for notifications. | same | Stop |
| 3. Notifications | If the user gave an address, you set up the emails first, so no approval email is missed. | notifications.md | — |
| 4. Turn on | You set GlueVersion to auto with a full job update. | assess-and-turn-on.md | — |
| 5. Review | Days later, you show the eligibility summary and what to check. | approval.md | Stop |
| 6. Approve or decline | You record the decision for each job that the user names, or decide each job on the user's blocking-issue condition. | same | — |
| 7. Monitor and results | You report status and the auto upgrade evaluation summary. | same | — |
Steps 1 to 4 can run in one conversation. A job that runs often can reach the approval request within an hour, so set up emails before turning on jobs. Step 5 starts after AWS Glue requests approval. Never approve in the same turn that turns on auto upgrade.
Tools
The skill uses only the AWS CLI (version 2, with the auto upgrade commands). It ships no scripts. Run each command exactly as the reference file shows it, with --region <region> --no-cli-pager. Add --profile <profile> when the user names a profile.
The AWS MCP server is recommended for working with AWS, but this skill doesn't require it. Run the commands in this skill with the AWS CLI, because the auto upgrade operations need a current AWS CLI.
Check the CLI once per session. This call makes no AWS request:
aws glue get-job-upgrade-analysis --generate-cli-skeleton > /dev/nullIf it fails, reply: "Your AWS CLI version doesn't support auto upgrade. To continue, update the AWS CLI to the latest version."
The assessment uses the generate_assessment tool of the spark-upgrade MCP server (assess-and-turn-on.md explains the setup). Permissions for each step are in permissions.md.
Rules
- Change only what the user confirmed. Before any call that changes something (
update-job,create-job,start-job-upgrade-analysis,create-topic,set-topic-attributes,put-rule,put-targets,subscribe), say what will change. Run it only for jobs and resources that the user confirmed in this conversation. - One decision per job, from the user's instruction. Record a decision only when the user names the job and the decision after you show its review, or when the user decides on the blocking-issue condition, for named jobs or for every waiting job: approve if the summary lists no blocking issue, decline if it lists at least one, or both. Act on that condition only if, in this turn, the job's status is
WAITING_FOR_CONSENTand you read its eligibility summary. Never decide on a schedule, on a condition you can't check from the summary, or for a job that the instruction doesn't cover. A reply such as "no", "not now" or "later" isn't a decline: record nothing, and give the decline prompt. - No secrets in the chat. Never repeat job argument values in a reply. Never print, store, or quote a summary link: the links are presigned, work for 1 hour, and are read only inside the one command in approval.md.
- No IAM changes or installs. Tell the user what to grant or install.
- No scripts. Write any temporary file inside the one command that uses it, and delete it in the same command.
How to reply
- Answer first. The first sentence states the result.
- Name the actor. Each sentence names who acts: AWS Glue or you. Don't use "we" or chatty words.
- Use the user guide's terms. Auto upgrade, eligibility summary, auto upgrade evaluation summary, approve, decline, previous version, safe to retry, subsequent job runs. Format
GlueVersion,MaxRetries, andautoas code. - Keep it short. Use present tense and active voice, with sentences of about 15 words and never more than 25. Put one sentence before each table or list.
- No internal details. Don't show raw status values, analysis IDs, run counts, or service-internal names.
- Show write modes at approval. One table row per job with its recorded write mode (
append, another mode,Not recorded, oroverwrite) and paths. The write mode describes what AWS Glue recorded; it never says a job is safe. - One question, at the end. End with the exact text that the user can enter, in code format.