# Authoring Skills

> Generates a new, spec-compliant, portable Agent Skill from a capability description, frontmatter, focused body, and the supporting file set, following progressive-disclosure and single-responsibility rules. Use when asked to create, scaffold, or package a skill. Sizes/splits via composing-skills and self-checks the output with verifying-reasoning, then hands to auditing-skills.

- **Type:** Skill
- **Install:** `agentstack add skill-adrianparedez-capability-fabric-authoring-skills`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [AdrianParedez](https://agentstack.voostack.com/s/adrianparedez)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** Apache-2.0
- **Upstream author:** [AdrianParedez](https://github.com/AdrianParedez)
- **Source:** https://github.com/AdrianParedez/capability-fabric/tree/main/skills/authoring-skills
- **Website:** https://paredez.dev

## Install

```sh
agentstack add skill-adrianparedez-capability-fabric-authoring-skills
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Authoring skills

This is the **skill factory**: it turns a capability description into a production-ready
skill directory that any runtime can use. It encodes the open spec and the authoring
best-practices so the output is correct by construction.

## Use this when
- Asked to create / scaffold / package a new skill.
- You notice a repeated procedure worth capturing (the "I keep re-explaining this" signal).
- **Skip** if an existing skill already covers the capability, extend it, don't fork.

## Evaluation-driven authoring (do this order)

```
- [ ] 1 GAP     run the task WITHOUT a skill; note what context/steps were missing
- [ ] 2 EVALS   write >=3 concrete eval scenarios that capture those gaps (before docs!)
- [ ] 3 FRAME   one capability, one cell of the taxonomy; name it (gerund, [a-z0-9-])
- [ ] 4 DESCRIBE write the what+when description (third person, trigger keywords, 
description: 
license:             # optional but recommended
metadata:                             # optional
  author: 
  version: "1.0"
  tier: ""
  layer: /
compatibility:    # optional
# allowed-tools: 
---
```

## Body rules the factory enforces
1. **One responsibility.** If the description needs "and also," split into two skills.
2. **Concise; assume intelligence.** Only context the model lacks. Cut anything it knows.
3. **Open with "use this when / not when."** Drives correct activation (`docs/07`).
4. **Degrees of freedom matched to fragility:** high (text) for judgment tasks, low (exact
   script/command) for fragile/irreversible ones.
5. **Progressive disclosure:** body = table of contents; detail in references, one level deep.
6. **No time-sensitive info, consistent terms, concrete examples, forward-slash paths.**
7. **Scripts solve, don't punt; no magic constants; declare dependencies.**

## Portability contract (every emitted skill obeys)
- Reference **capabilities** ("read a file", "run the tests"), never specific tool names in
  required steps.
- Forward-slash paths only. Declare real requirements in `compatibility`; assume nothing else.
- Include a short **"Runtime adaptation"** section: the minimum mechanisms needed and the
  graceful fallback when richer ones (subagents, hooks, compaction) are absent.
- `allowed-tools` only when genuinely needed; treat as experimental.

## Output layout the factory produces
```
/
├── SKILL.md
├── references.md        (depth, loaded on demand)
├── examples.md          (concrete input/output pairs)
├── templates/           (fill-in artifacts the skill uses)
├── checklists/          (gates the skill references)
└── benchmarks/          (>=3 eval scenarios + metrics, from step 2)
```
Use `templates/skill-skeleton/` as the starting structure and `templates/SKILL.md.tpl` for
the body.

## Runtime adaptation
- **Minimum:** filesystem write. Validate names with the spec rules (and `skills-ref
  validate` if available).
- The factory is itself portable: it emits plain files, no runtime-specific assumptions.

## Files
- `references.md`, the spec + best-practices distilled to authoring rules.
- `examples.md`, a capability turned into a full skill, start to finish.
- `templates/SKILL.md.tpl`, body template.
- `templates/skill-skeleton/`, the directory skeleton to copy.
- `checklists/authoring.md`, the pre-ship gate (mirrors Anthropic's checklist + portability).
- `benchmarks/`, does the generated skill pass its own evals & the audit?

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [AdrianParedez](https://github.com/AdrianParedez)
- **Source:** [AdrianParedez/capability-fabric](https://github.com/AdrianParedez/capability-fabric)
- **License:** Apache-2.0
- **Homepage:** https://paredez.dev

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-adrianparedez-capability-fabric-authoring-skills
- Seller: https://agentstack.voostack.com/s/adrianparedez
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.