dba-ops-skill/dba-ops-runbook
Troubleshoot database/infra errors, compose commands/scripts, write runbook tutorials, capacity planning for DBA, SysOps, DevOps. Covers PostgreSQL, MongoDB, MySQL, ClickHouse, Apache Kafka, RabbitMQ, Linux log management, cron, logrotate. Uses MCP connectors (context7, deepwiki, ClickHouse Docs) for up-to-date official docs. Command and Script output is caveman-compressed (telegraphic prose, byte-exact code); Tutorial and Plan stay full prose. Trigger whenever the user mentions DBA, SysOps, DevOps, or infrastructure — error diagnosis, shell commands/scripts, cron expressions, log rotation, capacity estimation, migration planning, access/auth troubleshooting, or any operational database/infrastructure task. Also trigger on pasted database errors, stack traces, or log snippets. Trigger phrases: "compose a tutorial", "write a runbook", "fix this cron", "how to restore/enable", "compose a command/script", topic sizing, retention, partitions, replication, cluster planning.
83
91%
Does it follow best practices?
Impact
80%
1.12xAverage score across 2 eval scenarios
Advisory
Suggest reviewing before use
README.md
dba-ops-runbook
A Claude skill for production-grade DBA, SysOps, and DevOps operational tasks — error diagnosis, runbook authoring, capacity planning, and infrastructure scripting — grounded in official documentation via MCP connectors.
What It Does
This skill turns Claude into a senior infrastructure engineer that fetches current official documentation before answering. It covers:
- Databases: PostgreSQL, MongoDB, MySQL/MariaDB, ClickHouse
- Messaging: Apache Kafka, RabbitMQ
- Infrastructure: Ansible, Authentik, Linux administration
- Operations: cron scheduling, log rotation, shell scripting, systemd
Output Modes
The skill adapts output format to the request:
| Mode | Trigger | What You Get |
|---|---|---|
| Command | "compose a command", "how to delete" | One-liner with inline comment, no boilerplate |
| Script | "compose a script", "write a script" | Complete executable with shebang, error handling, comments |
| Tutorial | "compose a tutorial", "write a runbook" | Prerequisites → numbered steps → verification → pitfalls |
| Plan | sizing, migration, partitions, retention | Constraints → math → config values → monitoring caveats |
MCP-First Architecture
Every answer begins with an MCP documentation lookup. The skill routes each technology to the best available connector:
| Technology | Primary MCP | Fallback |
|---|---|---|
| PostgreSQL | context7 | — |
| MySQL / MariaDB | context7 | — |
| MongoDB | context7 | MongoDB Docs |
| ClickHouse | ClickHouse Docs (fetch_generic_url_content) | Clickhouse |
| Apache Kafka | kafka Docs | context7 |
| Ansible | ansible-documentation Docs | — |
| Authentik | authentik Docs (fetch_generic_url_content) | — |
| Linux / cron / bash | Training knowledge | context7 (optional) |
When no MCP is connected, the skill distinguishes between version-sensitive topics (warns the user) and stable topics (proceeds from training knowledge without caveat).
MCP Quirks Handled
Several dedicated MCP servers index GitHub repos rather than rendered documentation. The skill contains workarounds for ClickHouse Docs, MongoDB Docs, and Authentik Docs — falling back to fetch_generic_url_content with constructed doc-site URLs when search tools return only READMEs.
Design Principles
No sycophancy. The skill is configured to provide technically correct answers, not agreeable ones. When the user's approach is flawed, the skill pushes back with evidence. Argumentation quality targets doctoral-dissertation level.
No chat history. Every response is self-contained. The skill never references prior conversation or cross-chat context — only the current request, MCP docs, and training knowledge.
Exemplary code quality. All scripts and config snippets are treated as Open Source Reference Application material — clear structure, proper error handling, defensive coding, meaningful comments, no reliability-for-brevity tradeoffs.
Hardcoded values over variables. Commands use realistic paths and usernames (find /opt/kafka/logs) instead of $VARIABLE placeholders. The user knows their environment.
Version awareness. When a feature may be version-gated, the skill states the minimum version or flags uncertainty and suggests checking the changelog.
Installation
Download the .skill file from Releases and install it in Claude's skill directory.
dba-ops-runbook/
└── SKILL.mdRequired MCP Connectors
For full functionality, connect these MCP servers in your Claude environment:
- context7 — general-purpose docs (PostgreSQL, MySQL, MongoDB, Kafka, Linux)
- ClickHouse Docs or Clickhouse — ClickHouse documentation
- kafka Docs — Apache Kafka documentation
- MongoDB Docs — MongoDB documentation (fallback; context7 preferred)
- ansible-documentation Docs — Ansible documentation
- authentik Docs — Authentik documentation
The skill degrades gracefully when connectors are absent — it warns on version-sensitive topics and proceeds from training knowledge on stable ones.
Examples
"Compose a command to clean up Kafka logs older than 7 days"
"Write a runbook for enabling ClickHouse dead letter queue"
"Fix this cron: 0 3 * * * mongosh --eval 'db.collection.aggregate([{$out:"backup"}])'"
"Size a Kafka cluster for 50k msg/s at 1KB average, 7-day retention"
"This MongoDB error: Unauthorized not authorized on admin to execute command"License
MIT