AgentSkillsCN

investigate-service-issues

调试 systemd 服务故障——查看服务状态、查阅日志、排查启动问题、定位失败单元。可选择自动重启。适用于用户输入“服务宕机”、“检查 systemd”、“调查服务”或“aa-workflow 未运行”时使用。

SKILL.md
--- frontmatter
name: investigate-service-issues
description: Debug systemd service failures - status, journal logs, boot issues, failed units. Optional auto-restart. Use when user says "service down", "check systemd", "investigate service", or "aa-workflow not running".

Investigate Service Issues

Debug systemd service failures and investigate issues. Checks status, journal logs, boot-level issues, failed units. Optional auto-restart.

Inputs

InputTypeDefaultPurpose
service_namestring-Service to check (e.g., aa-workflow-mcp). Empty = all failed
sincestring1hLog time window (1h, 30m, 1d)
auto_restartboolfalseRestart failed services
check_all_failedbooltrueList all failed systemd units

Persona

Load devops persona (systemd tools: systemctl, journalctl).

Workflow

1. Bootstrap

  • •persona_load("devops")

2. System Info

  • •hostnamectl_status() — hostname, OS
  • •timedatectl_status() — time sync (NTP)

3. Failed Units

  • •systemctl_list_units(pattern="--failed") — all failed units
  • •systemctl_list_unit_files(pattern="aa-workflow-*") — workflow units

4. Service-Specific (if service_name)

  • •systemctl_is_active(unit="{service}.service")
  • •systemctl_is_enabled(unit="{service}.service")
  • •systemctl_status(unit="{service}.service")
  • •journalctl_unit(unit="{service}.service", lines=50)

5. Broad Log Analysis

  • •journalctl_logs(priority="err", since=since, lines=30)
  • •journalctl_boot(boot="0", priority="warning", lines=20)

6. Ollama Check

  • •ollama_status() — inference server

7. Auto-Restart (if auto_restart and service_name)

  • •systemctl_daemon_reload()
  • •systemctl_restart(unit="{service}.service")
  • •systemctl_status(unit="{service}.service") — verify

8. Analyze

Parse for: inactive service, not enabled, failed units count, OOM, permission denied, connection refused, time not synced, restart success.

9. Log

  • •memory_session_log("Investigated service issues", ...)

10. Failure Patterns

  • •segfault/core dump → binary compatibility
  • •"too many open files" → ulimit, LimitNOFILE
  • •learn_tool_fix(...) for recurring patterns

Output

Report: issues, recommendations, service status, recent logs, failed units, restart result, system info.

Quick Actions

  • •systemctl_restart(unit="{service}.service")
  • •journalctl_unit(unit="{service}.service", lines=100)
  • •skill_run("manage_local_services", '{"action": "status"}')

Chains To

  • •manage_local_services