daniel-brus

You can investigate local and S3-shipped logs, inspect S3-compatible object storage keys and prefixes, check Redis cache keys and TTLs, run read-only Postgres validation queries, inspect the production Hetzner k3s VPS via whitelisted kubectl commands, merge feature PRs into develop, and manage package import layering as a DAG.

They target developers and operators maintaining the chess_teacher project who need safe, read-only debugging of production systems. Typical users are engineers diagnosing errors, checking cache behavior, validating database contents, reviewing deploy status, or refactoring module structure without risking writes, deletes, or restarts.

Production access uses SSH into the Hetzner VPS with whitelisted kubectl get/describe/logs commands, plus Postgres reads via kubectl exec into the streamlit pod. Credentials come from Doppler ci config, Redis uses REDIS_URL with redis-py, and storage reads use ObjectStorage with S3 or a local S3-compatible backend from .env.

Yes. All inspection skills are explicitly read-only: Redis never writes or deletes, VPS access forbids apply, delete, restart, or ad-hoc shell, and Postgres runs only validation queries. The merge skill never merges to main, restricting changes to the feature-to-develop flow with CI verification.

You need the chess_teacher project environment: a .env with S3 or S3-compatible backend credentials, REDIS_URL for cache access, Doppler ci config for SSH credentials, the DatabaseClient with metadata.yml table keys, and kubectl access to the k3s cluster. Production Postgres queries route through the VPS skill, not local doppler prod config.