config/skills/core/skill-builder/SKILL.md
This skill guides intentional skill design. Use when creating, improving, or reviewing Claude Code skills. Requires a zone assessment to clarify what kind of skill is being built before writing content.
npx skillsauth add gavinmcfall/agentic-config skillWriterInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
A skill encodes what cannot be derived from first principles. Before writing, understand what you're encoding.
Use your todo tool to track these checkpoints:
writing-documents skill invoked before writing each documentskill-name/
├── SKILL.md # Gestalt and cross-cutting concepts
├── meta/
│ ├── DISCOVERY.md # Research notes, domain understanding
│ └── CREATION_NOTES.md # Feedback on authoring experience
└── references/
└── ... # Detailed guidance by topic
Invariant You cannot write an effective skill without understanding the subject in excruciating detail first.
Example Before writing a mermaid skill: read all mermaid documentation, create dozens of diagrams, discover what breaks, learn which features are experimental, understand why certain patterns fail. //BOUNDARY: Zone assessment comes after discovery, not before. You can't assess what you don't understand.
Depth
meta/DISCOVERY.md—this is your working memorySkills contain different kinds of content. Most skills blend these, but the blend matters.
Invariant Inject context that cannot be derived from first principles.
Example API schemas defining valid request shapes. Company naming conventions. Business rules for edge cases. //BOUNDARY: If Claude could figure it out by reasoning, it belongs elsewhere.
Depth
references/knowledge.mdInvariant Coordinate sequences where order matters and steps cannot be safely skipped.
Example Deployment: build → test → stage → approve → deploy. Skipping "test" or reordering causes harm. //BOUNDARY: If steps can be reordered freely, use a checklist instead.
Depth
references/process.mdInvariant Enforce rules where deviation is the failure mode.
Example Security policy: Never log credentials. Never commit secrets. Always validate external input. //BOUNDARY: If deviation is acceptable in some contexts, it's guidance.
Depth
references/constraint.mdInvariant Introduce ways of thinking that generalize across contexts.
Example "Write for intelligent agents who need structure, not scripts." This reshapes approach to documentation, skills, communication—everything. //BOUNDARY: If it only applies to one domain, it's knowledge.
Depth
references/wisdom.mdAfter discovery, before writing, distribute 100 points across the zones.
This is not optional. The distribution clarifies what you're building and guides every subsequent decision.
| Zone | Points | What You're Encoding | |------|--------|---------------------| | Knowledge | ___ | Facts and context Claude lacks | | Process | ___ | Sequences where order matters | | Constraint | ___ | Rules that must be followed | | Wisdom | ___ | Ways of thinking | | Total | 100 | |
| Skill | K | P | C | W | Why | |-------|---|---|---|---|-----| | API integration | 70 | 15 | 10 | 5 | Mostly injecting facts Claude doesn't have | | Deployment workflow | 25 | 50 | 20 | 5 | Sequence matters, some rules | | Security review | 15 | 20 | 55 | 10 | Constraints dominate | | Writing for agency | 5 | 5 | 5 | 85 | Pure philosophy | | Code review | 30 | 15 | 25 | 30 | Balanced blend |
Where did most points go? That zone's reference file is your primary guide.
| Highest Zone | Primary Reference |
|--------------|-------------------|
| Knowledge | references/knowledge.md |
| Process | references/process.md |
| Constraint | references/constraint.md |
| Wisdom | references/wisdom.md |
Your distribution suggests how much to trust the agent versus prescribe behavior.
| Pattern | Agency Implication | |---------|-------------------| | High Knowledge, low Constraint | High agency — inject context, trust application | | High Process | Medium agency — flexibility within structure | | High Constraint | Variable — rules are firm, but application may flex | | High Wisdom | High agency — ideas reshape thinking, not behavior |
Note: High constraint doesn't automatically mean low agency. "These outcomes must be achieved" (high agency) differs from "Follow these exact steps" (low agency). Both enforce constraints.
| Zone | Content Feels Like | |------|-------------------| | Knowledge | Reference material, structured for retrieval | | Process | Sequenced steps, checkpoints, validation | | Constraint | Rules simply stated, boundaries explicit | | Wisdom | Ideas introduced, not templates provided |
Invoke the writing-documents skill now. Skills are documents. Their effectiveness depends on writing quality. Know how to write well before writing skill files.
With your distribution clear, ask:
On purpose:
On scope:
On form:
Consult based on your distribution:
references/knowledge.md — Structuring facts for retrieval, when high agency fitsreferences/process.md — When sequences matter, balancing flexibility with structurereferences/constraint.md — Rules without justification, when prescription is appropriatereferences/wisdom.md — Introducing ideas that change thinking, not templatesSkills with balanced distributions: consult multiple references, weight by allocation.
references/scripts.md — When to use scripts, best practices, platform considerations. Scripts serve Knowledge (fetch current truth), Process (automate steps), and Constraint (verify compliance).Before shipping, verify:
Launch a subagent to use the skill for real work:
The subagent catches rule violations (linear sequences in a mermaid diagram, missing error handling in a process). The human catches qualitative issues (is this actually useful? does it feel right?). The feedback captures what the skill got wrong.
Invariant A polymorphic skill presents shared concepts upfront, then branches to specific variations via progressive disclosure.
Example This skill: SKILL.md defines four zones and assessment. Each zone has a reference file with specific guidance. You learn the shared model first, then dive into the variant you need. //BOUNDARY: If variations share nothing, they're separate skills, not a polymorphic one.
Depth
Invariant Omit enough that the reader cannot succeed without engaging the referenced material.
Example
The zone capsules above tell you what each zone IS, not how to write one well. You cannot write a good wisdom skill from this page alone—you must read references/wisdom.md.
//BOUNDARY: Absence is productive only when the path forward is clear. Missing information with no pointer is just incomplete.
Depth
Invariant
Each skill includes meta/CREATION_NOTES.md documenting the authoring experience, rewritten by each maintainer.
Example
1. Create or update the skill using skillWriter
2. Read existing meta/CREATION_NOTES.md (if exists)
3. Rewrite with your experience synthesized (replacement, not append)
//BOUNDARY: Notes are about creating/maintaining the skill, not using it. Read only when rewriting.
Depth
You can't teach what you don't understand.
development
Deeply personal mentor and guide. Use when struggling, wanting to quit, feeling overwhelmed, or doubting yourself. Empathy-first. Build this skill around YOUR psychology.
tools
Build automation workflows with n8n for game dev tasks. Use when automating repetitive processes, setting up notifications, scheduling backups, or connecting services. Reduces manual overhead that ADHD brains find hardest to maintain.
testing
Query and diagnose the home Kubernetes cluster. Use when checking cluster health, troubleshooting pods/services/routes, inspecting storage, or understanding what's deployed. Covers Talos node management, Ceph storage, Cilium networking.
devops
Deploy and manage applications in the home-ops Kubernetes cluster via GitOps. Use when deploying new apps, modifying existing ones, adding routing, managing secrets, or working with the home-ops repo structure.