Agent-Skills.md

Agent Skills: Create Promotion Evaluation Template

>

UncategorizedID: flanksource/claude-code-plugin/promotion-eval-create

Repository

flanksource/claude-code-plugin
flanksource

Install this agent skill to your local

pnpm dlx add-skill https://github.com/flanksource/claude-code-plugin/tree/HEAD/skills/promotion-eval-create

Skill Files

Browse the full folder contents for promotion-eval-create.

Download Skill

Loading file tree…

skills/promotion-eval-create/SKILL.md

Skill Metadata

Name
promotion-eval-create
Description
>

Create Promotion Evaluation Template

Core Purpose

Guide the user through a structured interview to build a custom promotion evaluation skill for their system. The output is a complete SKILL.md file they can use to evaluate environment health and release readiness.

Use @skills/promotion-eval-mission-control/SKILL.md as the reference implementation — it demonstrates the structure, phased evaluation approach, verdict logic, and report format that this skill generates for other systems.

Workflow

Step 1: Identify the System

Ask the user:

  1. What system or platform are you evaluating? (e.g., Kubernetes cluster, AWS environment, SaaS application, database cluster, CI/CD pipeline)
  2. What is the evaluation for? (e.g., release promotion, environment readiness, disaster recovery validation, compliance check)
  3. What environment(s) will be evaluated? (e.g., staging, production, specific cluster names)

Step 2: Define Components

Ask the user to list the health components they want to evaluate. Suggest categories based on their system type:

For Kubernetes-based systems:

  • Deployments/StatefulSets health
  • Pod status and restarts
  • Node health and capacity
  • Ingress/networking
  • PersistentVolume status
  • CronJob success rates

For cloud infrastructure (AWS/GCP/Azure):

  • Compute instance health
  • Database connectivity and replication
  • Load balancer targets
  • Certificate expiry
  • Storage utilization
  • Network connectivity

For SaaS/application systems:

  • API endpoint health
  • Database query performance
  • Queue depth and processing rates
  • Error rates and latency
  • Authentication/SSO status
  • Third-party dependency health

For CI/CD pipelines:

  • Build success rates
  • Test pass rates
  • Deployment success rates
  • Artifact availability
  • Environment provisioning

For each component, ask:

  • What tool or API provides the health data? (MCP tool, HTTP endpoint, CLI command, database query)
  • What metrics matter? (counts, rates, durations, thresholds)
  • What are the PASS/WARN/FAIL thresholds?

Step 3: Define Parameters

Ask about configurable parameters:

  • time_window: What lookback period? (default: 24h)
  • target: How is the environment identified?
  • Any custom parameters specific to their system?

Step 4: Define Verdict Logic

Confirm the overall verdict mapping:

  • READY: Which components must PASS?
  • CAUTION: Which components can WARN without blocking?
  • NOT_READY: Which component failures are blocking?
  • Are any components optional (can SKIP without affecting verdict)?

Step 5: Generate the Skill

Using the gathered information, generate a complete SKILL.md following this structure:

---
name: promotion-eval-<system-name>
description: >
  Evaluates <system> health for <purpose>.
  Checks <component list summary>.
  Use for <trigger scenarios>.
allowed-tools: <list of MCP tools or other tools needed>
---

# <System> Promotion Evaluation Skill

## Core Purpose
<One paragraph describing what this evaluation does>

## Parameters
- **time_window**: Lookback period (default: <default>)
- **target**: <how environment is identified>
<any custom parameters>

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

Initialize a running JSON result conforming to the schema:
<JSON template with verdict, components, findings, recommendations>

---

### Phase N: <Component Name>
**Goal**: <what this phase checks>

1. <Step-by-step tool calls or queries>

**Metrics to record**:
- <metric>: <description>

**Verdict logic**:
- PASS: <criteria>
- WARN: <criteria>
- FAIL: <criteria>

---
<repeat for each component>

## Report Generation
<Markdown report template>
<JSON output template>

## Overall Verdict Logic
<Component-to-verdict mapping>

## Error Handling
- If a tool call fails, record the component as SKIP with a note
- Do not let one phase failure block subsequent phases
- Reuse data across phases when possible

Step 6: Review and Refine

Present the generated skill to the user and ask:

  • Does the component list look complete?
  • Are the thresholds appropriate?
  • Should any phases be added or removed?
  • Are the tool references correct?

Iterate until the user is satisfied.

Output

Write the final SKILL.md to the user's chosen location (default: current project's .claude/skills/ directory).

If the evaluation uses MCP tools, also note which MCP servers need to be configured.