---
# ==============================================================================
# auto-remediate.yml — Confidence-gated safe SRE remediation with jev-ops
# Requires TYPESAFE_API_KEY in the environment of the managed hosts.
# ==============================================================================
- name: SRE Autonomous Diagnostics and Gated Remediation
  hosts: webservers
  gather_facts: false

  tasks:
    - name: Collect last 50 error journal entries
      ansible.builtin.command: "journalctl -p err --since '-10m' -n 50 --no-pager"
      register: journal_output
      changed_when: false

    # A quiet journal is the healthy case: skip analysis instead of failing on empty input.
    - name: Evaluate diagnostic health via jev-ops
      ansible.builtin.command: "jev-ops analyze linux --json"
      args:
        stdin: ""
      register: jev_result
      changed_when: false
      when: journal_output.stdout | trim | length > 0

    - name: Parse typed decisions
      ansible.builtin.set_fact:
        jev_decisions: ""
      when: jev_result is not skipped

    - name: Display diagnostic decision
      ansible.builtin.debug:
        msg: >
          Category: 
          (confidence 0%),
          Severity: /5
      when: jev_decisions is defined

    # POLICY GATE: Only auto-restart the service for application faults, with >= 95%
    # confidence, when attention is required. Restarting nginx does not fix disk errors.
    - name: Auto-remediate web service
      ansible.builtin.service:
        name: nginx
        state: restarted
      when:
        - jev_decisions is defined
        - jev_decisions.decisions.needs_attention.value
        - jev_decisions.decisions.category.value == 'application'
        - (jev_decisions.decisions.category.confidence | float) >= 0.95

    # ESCALATION: Notify on-call if severity is high or the model reports low confidence
    - name: Escalate to On-Call Slack channel
      ansible.builtin.uri:
        url: ""
        method: POST
        body_format: json
        body:
          text: >
            🚨 jev-ops needs a human on .
            Category: 
            (0% confidence),
            Severity: /5.
      when:
        - jev_decisions is defined
        - jev_decisions.decisions.needs_attention.value
        - (jev_decisions.decisions.severity.value | int) >= 4
          or (jev_decisions.decisions.category.confidence | float) < 0.60