helm-skill-builder

name	helm-skill-builder
description	Generates per-application skill directories under /skills/ with SKILL.md and reference files that guide the helm-generator sub-agent. Use when the coordinator determines no skill exists for the requested application type AND the helm-planner did not auto-generate skills. This is the fallback skill builder for generic chart generation requests. Triggers on keywords: skill builder, create skill, scaffold skill, new app type, unknown chart.
compatibility	Requires write_file, read_file, ls tools. Virtual filesystem rooted at /. Must write to /skills/helm-operator/{app}-chart-generator/.
metadata	{"author":"talkops-ai","version":"2.0"}
allowed-tools	read_file write_file ls

Helm Skill Builder

When to Use

Use this skill as a fallback when:

1. No app-specific skill exists under /skills/helm-operator/{app}-chart-generator/
2. The helm-planner did NOT auto-generate skills (its output does not contain "Skills written for")

If the planner already wrote skills, this agent is NOT needed — the coordinator skips it.

Workflow

Progress:

• Step 1: Read conventions from memory
• Step 2: Determine the template file set
• Step 3: Write SKILL.md
• Step 4: Write reference files
• Step 5: Return summary

Step 1. Read Conventions

read_file /memory/helm-operator/AGENTS.md

Check for any global conventions (naming, labels, security requirements).

Step 2. Determine Template File Set

Based on the application type, determine which templates are needed:

Always create:

• Chart.yaml, values.yaml
• templates/_helpers.tpl, templates/deployment.yaml, templates/service.yaml

Add based on app needs:

• templates/ingress.yaml — when the app exposes HTTP endpoints
• templates/hpa.yaml — when autoscaling is relevant
• templates/pdb.yaml — for HA deployments (replicaCount > 1)
• templates/configmap.yaml — when the app has config files
• templates/secret.yaml — when the app has sensitive config
• templates/serviceaccount.yaml — when RBAC is needed
• templates/networkpolicy.yaml — when network isolation is needed
• templates/NOTES.txt — always recommended

Step 3. Write SKILL.md

Write to /skills/helm-operator/{app}-chart-generator/SKILL.md with this structure:

---
name: {app}-chart-generator
description: >-
  Generates production-grade Helm chart for {App}. Use when asked to
  create, scaffold, or deploy {App} on Kubernetes. [Add specific keywords].
compatibility: >-
  Requires write_file, read_file, edit_file, ls, grep tools.
  Virtual filesystem rooted at /workspace/.
metadata:
  author: helm-skill-builder
  version: "1.0"
allowed-tools: write_file read_file edit_file ls grep
---

Body MUST include:

• When to Use section
• Reference Files table (progressive disclosure)
• Chart Generation Workflow with step checklist
• Safety Rules (≥5 rules)
• Gotchas (≥3 items)
• Response Format

Follow the patterns from existing skills under /skills/helm-operator/.

Step 4. Write Reference Files

Create these reference files under /skills/helm-operator/{app}-chart-generator/references/:

File	Contents
execution-blueprint.md	Kubernetes resource specifications (ports, probes, env vars, volumes)
values-schema.md	Complete values.yaml structure with defaults for all parameters
scaling-and-resources.md	Resource requests/limits, HPA targets, PDB config
security-blueprint.md	SecurityContext, NetworkPolicy rules, RBAC

Step 5. Return Summary

Skill written at /skills/helm-operator/{app}-chart-generator/.
Declared file set: [list of template files the generator should create]

Safety Rules — MUST Follow

1. SKILL.md must follow agentskills.io specification. Name must be lowercase a-z and hyphens only, 1-64 chars, matching the directory name.

2. Description must include "Use when" triggers. The description field is used for skill routing — vague descriptions cause misrouting.

3. Never hardcode app-specific values in SKILL.md body. Put data in reference files. SKILL.md is the procedure, references are the data.

4. Reference files must have concrete defaults. Every value in values-schema.md must have a sensible production default, not TODO placeholders.

Gotchas

• The skill directory name MUST match the name field in SKILL.md frontmatter exactly. Mismatches cause the skill loader to skip the skill silently.

• Don't over-specify templates for simple apps. A basic web app needs: deployment, service, ingress, _helpers.tpl. Adding HPA/PDB/NetworkPolicy for a dev-only tool wastes tokens.

• The helm-generator sub-agent reads the SKILL.md first, then loads references on demand. Keep SKILL.md under 5000 tokens and put detailed specs in references.

Response Format

Skill written at /skills/helm-operator/{app}-chart-generator/.
Declared file set: [Chart.yaml, values.yaml, templates/_helpers.tpl, templates/deployment.yaml, ...]
Reference files: [execution-blueprint.md, values-schema.md, ...]