Engineering
SRE triage bot: skill + box + Grafana/Sentry/GitHub MCP
SRE agent demos need a custom environment and seven hand-written tools first.
What hurt
刘屿之外 (@Lucian_Lc7) contrasted Anthropic's SRE agent demo (define agent, environment, session, upload logs, write tools for metrics/deploys/diff) with Grok Bot: create a triage bot, encode on-call steps as a Skill, drop logs on the box, connect Grafana/Sentry/GitHub via MCP, then ask what caused the latency spike. Tip with method — no invented skill body.
What to install or create
One triage bot. Your on-call checklist becomes a Skill. Box reads logs/greps. MCP for Grafana/Sentry/GitHub when available. One concrete incident question per run.
Bots
SRE triage
Follows on-call Skill. Reads logs on the box. Uses MCP for metrics/issues/diff. No deploy without yes.
Plugins
Browser
Optional
Files
On-call Skill. Incident log drop folder.
Routines
None on the first run. Watch one manual pass first.
The prompt
This is a tip, not a pasteable system prompt. The author did not publish one. Follow the first-run steps.
First run
- 1Read https://x.com/Lucian_Lc7/status/2098939622307029495.
- 2Create SRE triage. Write your real on-call steps into a Skill (inputs, checks, output format, approvals).
- 3Connect Grafana/Sentry/GitHub MCP when you have them. Drop one incident log bundle on the box.
- 4Ask one concrete question (e.g. what caused this latency spike?). Review the cited evidence.
- 5Only after clean read-only runs, allow deploy/rollback behind your yes.
Approval fence
Read-only on first incidents. Deploys/rollbacks need your yes.
What to tweak
Steal the Skill + box + MCP stack. Do not invent the Anthropic tool functions.
Source
刘屿之外 @Lucian_Lc7 · Original post
Not affiliated with xAI. We did not write this prompt. Third-party template terms still apply.