chatops-skill

name	chatops-skill
description	Enables the agent to communicate with human security analysts for notifications and high-stakes confirmations (Human-in-the-loop).
references	["https://developers.google.com/workspace/chat/design-interactive-card-dialog"]

ChatOps and Human Interaction Implementation

You are equipped with the capability to send rich notifications and action requests to human security analysts via ChatOps (Google Chat / Webhooks). Use this skill when you need a human to perform a task, confirm a state-changing action, or stay informed about a critical event.

Instructions:

1. Critical Actions (HITL): Before performing any irreversible or high-risk state-changing actions (e.g., isolating a host, blocking a user account, or wiping a machine), you MUST propose the action to a human analyst for confirmation using the request_human_confirmation tool.
2. Incident Notifications: When you identify a confirmed CRITICAL or HIGH severity incident, immediately notify the human team using the notify_human_incident tool.
3. Custom Alerts: Use the send_chatops_card tool for general-purpose notifications or status reports that require a structured card format.
4. Transparency: When you send a notification or a request to a human, include this in your orchestrator synthesis (e.g., "I've requested a human confirmation for the host isolation action...").
5. Human Context: Always provide clear rationale and evidence when requesting human input, so the analyst has the necessary context to make a decision.

Tools Summary:

• request_human_confirmation: Propose a specific action with context for approval/denial (legacy flow).
• notify_human_incident: Alert the team to a confirmed incident with severity and IDs.
• send_chatops_card: Send any custom card by raw widget layout.
• verify_user_travel: Send an impossible travel confirmation card directly to a user.
• request_triage_approval: Send a pre-formatted approval card for host isolation or emergency patching.
• deliver_report: Send a card providing a secure download link for a finalized triage report.
• generic_notification: Dispatcher for any ai_* chatops card template by its Python filename.

Example Card Layout Patterns

When using send_chatops_card, follow these modernized patterns for the sections argument to ensure a premium analyst experience:

1. IOC Enrichment Layout (Columns + materialIcon)

[
  { "widgets": [
    { "columns": { "columnItems": [
      { "widgets": [ { "decoratedText": { "topLabel": "CrowdStrike", "text": "APT28 / Fancy Bear", "startIcon": { "materialIcon": { "name": "flag" } } } } ] },
      { "widgets": [ { "decoratedText": { "topLabel": "VirusTotal", "text":             
      
                              
  

2. Runtime / Threat Detection Layout (Grouped Metrics)

[
  { "widgets": [
    { "columns": { "columnItems": [
      { "widgets": [ { "decoratedText": { "topLabel": "Target Pod", "text": "auth-api-88x", "startIcon": { "materialIcon": { "name": "layers" } } } } ] },
      { "widgets": [ { "decoratedText": { "topLabel": "CPU Usage", "text": "100%"            
      
          
        
                          
                 
     
  

3. User/Entity Context (Location side-by-side)

[
  { "widgets": [
    { "columns": { "columnItems": [
      { "widgets": [ { "decoratedText": { "topLabel": "Current Login", "text": "London, UK", "startIcon": { "materialIcon": { "name": "place" } } } } ] },
      { "widgets": [ { "decoratedText": { "topLabel": "Previous Login", "text":             
      
                                
  

Reference Scenarios for ChatOps

The following scenarios are pre-defined as high-value for human interaction. Use the tools indicated for each:

Access & Identity (Use request_human_confirmation)

• ai_credential_reset_approval: Request approval before forcing a password reset on a high-value account.
• ai_privilege_access_v2: Propose and approve privilege access requests using the newer V2 card format.
• ai_privileged_session_recording: Notify and request logic for initiating recording on a sensitive session.
• ai_stale_account_cleanup: Propose deletion of identified stale or dormant accounts.
• ai_suspicious_login_location: Request user/analyst confirmation for logins from new geo-locations.
• ai_user_privilege_audit: Request privilege review for users with excessive permissions.
• impossible_travel_verification: Request verification from the affected user regarding an impossible travel event.
• temp_admin_request: Propose and approve temporary administrative/break-glass access.
• mfa_api_key_alert: Notify and request MFA enforcement for vulnerable service keys.
• ai_security_group_audit: Propose modifications to firewall groups or IAM policies after a security audit.

Containment & Remediation (Use request_human_confirmation)

• ai_brute_force_source_block: Request approval to block an IP address at the firewall after brute force detection.
• ai_data_exfiltration_block: Request to block an egress point once exfiltration is suspected.
• ai_firewall_bypass_request: Request approval for a temporary firewall bypass rule.
• ai_malicious_container_kill: Request approval to terminate a compromised K8s pod or container.
• ai_suspicious_process_kill: Request approval before killing a process on a critical server.
• ai_wipe_host_approval: MANDATORY HITL: Never wipe a host without explicit human approval.
• ai_data_exfiltration_block: Request to block an egress point once exfiltration is suspected.
• ai_malicious_domain_sinkhole: Propose redirection of malicious domain traffic to a sinkhole.
• host_isolation_approval: Propose isolating an infected host from the network.
• vulnerability_patch_approval: Seek approval for applying emergency security patches to production systems. For testing/demos, use the Ivanti Endpoint Manager (CVE-2026-1603) example.

Alerting & Intel (Use notify_human_incident / send_chatops_card)

• ai_threat_intel_sharing: Use send_chatops_card to share new IOCs discovered during an investigation.
• ai_compliance_violation_alert: Notify humans of identified policy or compliance drifts.
• ai_dns_exfiltration_detection: Notify humans of anomalous DNS patterns indicative of tunneling.
• ai_forensic_image_approval: Request permission to take a forensic disk/memory image.
• ai_threat_hunt_hypothesis: Present a new threat hunting hypothesis to human analysts for feedback.
• brute_force_alert: Notification to the team of a detected brute force attack.
• impossible_travel_alert: Notification of a potential impossible travel login event.
• ioc_enrichment_card: Visual card showing multi-vendor intelligence for an IP, Domain, or Hash.
• malware_sandbox_report: Summary of automated sandbox analysis (static and dynamic behavior).
• phishing_report_summary: Overview of user-reported phishing attempts and identified risk.
• ai_canary_token_deployment: Propose the deployment of honeytokens/canaries in sensitive environments.
• shadow_it_discovery: Alert on newly discovered unmanaged cloud resources or applications.

Operational Workflow (Use send_chatops_card / request_human_confirmation)

• ai_draft_comms_review: Propose a draft email/notification for a human to review before sending.
• ai_playbook_selection: Ask the human which strategy to prioritize if multiple playbooks apply.
• ai_incident_summary_confirm: Request a human to review your final incident summary before closure.
• ai_incident_closure_confirm: Final sign-off required from an analyst before officially closing an incident.
• ai_incident_retrospective_request: Trigger a post-mortem or retrospective task after a significant incident.
• ai_network_scan_approval: Request approval before initiating an active network vulnerability scan.
• ai_sensitive_log_access: Request approval for a security analyst to access highly sensitive logs.
• bulk_deletion_verification: Require human dual-control for any bulk deletion of logs or records.
• ai_false_positive_tuning: Propose logic changes to detection rules to reduce noise.
• ai_user_interview_request: Ask an analyst to interview a user to confirm suspicious (but potentially benign) activity.
• ai_data_classification_request: Request a human to classify sensitive data found in an unconventional location.
• ai_vulnerability_revalidation: Request a human to manually verify that a reported vulnerability has been fixed.
• forensics_evidence_ready: Notify the team that requested forensic evidence collection has finished.