Agent Skills: Mission Control Promotion Evaluation Skill

Mission Control Promotion Evaluation Skill

Core Purpose

Systematically evaluate the health of a Mission Control environment as a platform (not demo workloads) to support release and promotion readiness decisions. This skill queries the live environment using MCP tools and produces a structured diagnostic report.

Important Distinction

This evaluation focuses on Mission Control platform health — the components that make MC work (canary-checker, config-db, mission-control deployments, core health checks, scrapers, jobs). It does NOT evaluate demo workloads or user-created resources unless they indicate a platform problem.

Known expected-fail checks: Some health checks have the label Expected-Fail=true. These are intentional test checks and should be excluded from failure counts and findings.

Parameters

When invoked, check if the user specified:

• time_window: Lookback period (default: 24h)
• target: Environment to evaluate (ask user if not specified)

Evaluation Procedure

Execute these phases sequentially. After each phase, record component status and findings.

Initialize a running JSON result conforming to @skills/promotion-eval-mission-control/schema.json with:

{
  "verdict": "READY",
  "evaluated_at": "<current ISO timestamp>",
  "time_window": "<window>",
  "target": "<target environment>",
  "components": {},
  "findings": [],
  "recommendations": []
}

Catalog Type Reference

These are the confirmed MissionControl catalog types:

• MissionControl::ScrapeConfig — config scrapers
• MissionControl::Playbook — playbook definitions
• MissionControl::Notification — notification rules
• MissionControl::Job — background jobs
• MissionControl::Canary — canary check definitions
• MissionControl::Connection — external connections
• MissionControl::Topology — topology definitions

Phase 1: Health Check Pipelines

Goal: Determine if health checks are running and passing.

1. Get failing checks directly: view_failing-health-checks_mission-control with withRows=true and select=["id","name","type","status","severity","last_transition_time","description"]
2. Get total check count: list_all_checks for baseline metrics
3. Filter out expected failures: Exclude checks with label Expected-Fail=true from failure counts
4. Drill into real failures: For each genuinely unhealthy check (not expected-fail), call get_check_status(id, limit=10) to retrieve recent execution history. Classify as:
   • Transient: Occasional failures mixed with passes
   • Persistent: Consistently failing across recent executions
5. Assess staleness: From the check list, identify checks where updated_at is older than the time window

Metrics to record:

• total_checks: Total number of health checks
• healthy_count: Number currently healthy
• unhealthy_count: Number currently unhealthy (excluding expected-fail)
• expected_fail_count: Checks labeled Expected-Fail
• persistent_failures: Number failing consistently
• stale_count: Number not updated within time window
• health_rate: Percentage healthy (excluding expected-fail from denominator)

Verdict logic:

• PASS: No persistent failures, stale_count == 0, health_rate > 95%
• WARN: Some transient failures OR 1-2 stale checks OR health_rate 80-95%
• FAIL: Any persistent failures OR stale_count > 2 OR health_rate < 80%

Phase 2: Config Scrapers

Goal: Verify config scrapers are active and producing fresh data.

1. Find scraper configs: search_catalog with type=MissionControl::ScrapeConfig and select=["id","name","health","status","updated_at"]
2. Check freshness: For each scraper, check updated_at timestamp. Flag any not updated within expected schedule (typically 1h)
3. Check scraper errors: view_mission-control-system_mission-control with withPanels=true — this returns scraper error counts and a list of scrapers with errors
4. Review recent changes: search_catalog_changes with type=MissionControl::ScrapeConfig created_at>now-{window} for config changes

Metrics to record:

• total_scrapers: Number of scraper configs found
• active_count: Scrapers updated within expected window
• stale_count: Scrapers not recently updated
• error_count: From system view scraper errors panel

Verdict logic:

• PASS: All scrapers active, no errors
• WARN: 1-2 scrapers slightly stale OR minor errors
• FAIL: Any scraper missing updates for > 2h OR significant errors

Phase 3: Background Jobs & Playbooks

Goal: Check for failed playbook runs and job errors.

1. Get failed job history: view_jobhistory_mission-control with withRows=true and select=["name","status","duration","error","timestamp"] limit=20
2. Get failed playbook runs: get_playbook_failed_runs(limit=10) for recent failures
3. Get recent playbook runs: get_playbook_recent_runs(limit=20) to calculate success rate
4. Drill into failures: For any failed playbook runs, call get_playbook_run_steps(run_id) to understand the failure cause
5. Check playbook catalog health: search_catalog with type=MissionControl::Playbook health=unhealthy to find unhealthy playbook definitions

Metrics to record:

• total_recent_runs: Total playbook runs in window
• failed_runs: Number of failed runs
• success_rate: Percentage of successful runs
• job_errors: Count of job errors from job history view

Verdict logic:

• PASS: success_rate > 95%, no recurring job errors
• WARN: success_rate 80-95% OR some job errors
• FAIL: success_rate < 80% OR critical/recurring job failures

Phase 4: Notification Delivery

Goal: Verify the notification pipeline is functioning.

1. Get notification send history: view_notification-send-history_mission-control with withRows=true and select=["id","age","resource_name","resource_current_health","title","notification"] limit=20
2. Get notification stats from system view: The view_mission-control-system_mission-control panel (already fetched in Phase 2) includes notification counts by status (SENT, SILENCED, REPEAT-INTERVAL, etc.)
3. Find notification configs: search_catalog with type=MissionControl::Notification and select=["id","name","health","status"]
4. Check for error notifications: For each notification config, call get_notifications_for_resource(resource_id, status=error, since=now-{window})

Metrics to record:

• total_notification_configs: Number of notification rules
• sent_count: From system view
• silenced_count: From system view
• error_count: Notifications with error status
• delivery_rate: sent / (sent + error) percentage

Verdict logic:

• PASS: No delivery errors, system view shows sends happening
• WARN: Some errors but delivery_rate > 95%
• FAIL: delivery_rate < 95% or notification system appears down

Phase 5: System & Event Queue

Goal: Check overall system health indicators, database, and event queue.

1. System overview: view_mission-control-system_mission-control with withPanels=true (reuse from Phase 2 if already fetched)
   • Check scraper errors, notification stats, agent resource counts
2. Database health: view_mission-control-database_mission-control with withPanels=true
   • Check DB size, active users, DB connections
3. Connection health: list_connections to verify external integrations are configured

Metrics to record:

• db_size_bytes: Database size
• db_connections: Active connections
• active_users: User count
• total_connections: Number of configured connections

Verdict logic:

• PASS: DB healthy, connections configured, no concerning metrics
• WARN: High DB connections or large DB size growth
• FAIL: Database unreachable or critical system errors

Phase 6: MC Infrastructure Health

Goal: Verify Mission Control's own Kubernetes resources are healthy.

1. Get MC pods directly: view_mission-control-pods_mission-control with withRows=true and select=["name","namespace","status","health","updated"] — this returns all MC-related pods
2. Find MC deployments: search_catalog with type=Kubernetes::Deployment and name patterns:
   • name=mission-control*
   • name=canary-checker*
   • name=config-db* Use select=["id","name","health","status","updated_at"] for each.
3. Describe unhealthy resources: For any unhealthy MC deployment or pod, call describe_catalog(id) to get full details including error messages
4. Check recent changes: search_catalog_changes with type=Kubernetes::Deployment name=mission-control* created_at>now-{window} (and similar for canary-checker, config-db)
5. Check related configs: For unhealthy resources, use get_related_configs to trace Deployment → ReplicaSet → Pod

Metrics to record:

• total_mc_pods: MC pods found
• healthy_pods: Healthy MC pods
• unhealthy_pods: Unhealthy MC pods
• total_mc_deployments: MC deployments found
• healthy_deployments: Healthy MC deployments
• recent_changes: Changes to MC components in time window

Verdict logic:

• PASS: All MC resources healthy, no concerning recent changes
• WARN: Recent deployments/changes but all healthy, OR minor pod restarts
• FAIL: Any unhealthy MC deployment or persistent pod failures

Report Generation

After all phases complete, produce the final report in two parts:

Part 1: Markdown Report

# Promotion Evaluation Report

**Target**: <target environment>
**Evaluated at**: <timestamp>
**Time window**: <window>
**Verdict**: **<READY|CAUTION|NOT_READY>**

## Summary

| Component | Status | Key Metrics |
|-----------|--------|-------------|
| Health Checks | <PASS/WARN/FAIL> | <health_rate>% healthy, <persistent_failures> persistent failures |
| Config Scrapers | <PASS/WARN/FAIL> | <active_count>/<total_scrapers> active, <error_count> errors |
| Jobs & Playbooks | <PASS/WARN/FAIL> | <success_rate>% success rate, <failed_runs> failures |
| Notifications | <PASS/WARN/FAIL/SKIP> | <sent_count> sent, <error_count> errors |
| System & DB | <PASS/WARN/FAIL> | DB <db_size>MB, <db_connections> connections |
| MC Infrastructure | <PASS/WARN/FAIL> | <healthy_pods>/<total_mc_pods> pods healthy |

## Findings

<For each finding, sorted by severity (critical first)>
### [severity] [component]: [message]
- **Resource**: [name] ([type], ID: [id])
- **Evidence**: [evidence]

## Recent Changes (Risk Factors)

<List recent changes to MC infrastructure that could affect stability>

## Recommendations

<Numbered list of actionable recommendations>

Part 2: Structured JSON

Output the completed JSON object conforming to the schema. Wrap in a code block with language json.

Overall Verdict Logic

Derive the top-level verdict from component statuses:

• READY: All components PASS (or SKIP for non-critical ones)
• CAUTION: Any component is WARN, but none are FAIL
• NOT_READY: Any component is FAIL

Components that must not FAIL for READY: health_checks, config_scrapers, mc_infrastructure Components that may SKIP without affecting verdict: notifications

Error Handling

• If an MCP tool call fails or returns unexpected data, record the component as SKIP with a note
• Do not let one phase failure block subsequent phases — evaluate all phases independently
• Reuse data across phases when the same tool was already called (e.g., system view data)