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
{
"context": "Tests whether the agent uses Check mode when asked to verify Kafka connectivity: output must be bare probe commands only, with no prose preamble, postamble, explanatory lines, or comment lines. Multiple probes must appear on separate lines, and commands must not contain invented hostnames or IP addresses.",
"type": "weighted_checklist",
"checklist": [
{
"name": "No prose preamble",
"description": "check_output.txt does NOT begin with any introductory or explanatory text before the first command (no lines matching patterns like 'here are', 'sure,', 'to check', 'you can run', 'the following', 'below are', 'these commands')",
"max_score": 10
},
{
"name": "No prose postamble",
"description": "check_output.txt does NOT end with any explanatory, concluding, or advisory text after the last command (no trailing lines like 'you may need to', 'note that', 'hope this helps', 'adjust as needed', 'let me know')",
"max_score": 10
},
{
"name": "No mid-output explanatory lines",
"description": "check_output.txt contains NO lines that consist of plain English prose — every non-blank line is a shell command, not an explanation (no lines like 'this checks whether...', 'this command will...', 'first, run...')",
"max_score": 10
},
{
"name": "No comment lines",
"description": "check_output.txt contains NO lines that begin with `#` (no inline shell comments explaining what any command does)",
"max_score": 10
},
{
"name": "Two or more probe commands",
"description": "check_output.txt contains at least 2 distinct non-blank command lines",
"max_score": 10
},
{
"name": "Separate lines for each check",
"description": "The port-listening check and the advertised.listeners config check appear on separate lines — they are NOT chained into a single line using `&&`, `;`, or `|` in a way that combines both checks as one compound statement",
"max_score": 10
},
{
"name": "Port 9092 probe",
"description": "At least one line in check_output.txt contains `9092` alongside a network inspection tool (ss, netstat, nc, nmap, lsof, or telnet) to check whether Kafka is listening on that port",
"max_score": 15
},
{
"name": "advertised.listeners probe",
"description": "At least one line in check_output.txt contains the exact string `advertised.listeners` to inspect the Kafka broker configuration",
"max_score": 15
},
{
"name": "No invented hostname as target",
"description": "No command line in check_output.txt targets an invented hostname or fabricated IP address as a connection destination (no patterns like `myhost:9092`, `broker1`, `kafkaserver`, `kafka.example.com`, or a made-up IP followed by `:9092`)",
"max_score": 10
}
]
}