# Dg Skill Authoring > Author a new Agent Skill that adds a feature to an Obsidian Digital Garden (oleeskild digital-garden plugin + Eleventy) site. Use when a user wants to package a Digital Garden customization - a component, widget, layout tweak, or style - as a reusable, upstream-safe SKILL.md that another DG user's agent can install. Produces a spec-compliant skill (frontmatter, progressive disclosure, references,… - **Type:** Skill - **Install:** `agentstack add skill-palol-skills-zoo-dg-skill-authoring` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [palol](https://agentstack.voostack.com/s/palol) - **Installs:** 0 - **Category:** [Productivity](https://agentstack.voostack.com/c/productivity) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [palol](https://github.com/palol) - **Source:** https://github.com/palol/skills-zoo/tree/main/skills/dg-skill-authoring - **Website:** http://skillz.paologabriel.com/ ## Install ```sh agentstack add skill-palol-skills-zoo-dg-skill-authoring ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Digital Garden Skill Authoring A meta-skill: it teaches an agent how to **build another Agent Skill** that adds a feature to an Obsidian [Digital Garden](https://github.com/oleeskild/digital-garden) (Eleventy) site - following the upstream-safe, progressively-disclosed pattern proven by `dg-floating-tray`. The output is a spec-compliant skill folder (`SKILL.md` + `assets/` + `references/`), validated with `skills-ref validate`, ready to drop into a skills repo or serve over MCP. ## When to Use This Skill Use when a Digital Garden user wants to **package a DG customization as a reusable skill** - e.g. "turn this floating dock into a skill", "make a skill for my custom callout", "write a skill another DG user can install to add a guestbook widget". Do **not** use to *install* an existing DG skill (that's the target skill's job), or to author skills for non-DG stacks (Quartz, Astro, plain Eleventy) - the DG conventions below won't apply. ## The DG Skill Pattern (what "good" looks like) Every DG feature skill this meta-skill produces must satisfy seven invariants. These are the lessons baked into `dg-floating-tray` (plus inspiration citations from later skills): 1. **User-owned paths only.** Write into paths the DG plugin treats as user territory, so upstream `git pull` never overwrites them: - Components: `src/site/_includes/components/user/common//*.njk` - Styles: append to `src/site/styles/custom-style.scss` - Helpers: `src/site/helpers/userUtils.js` - Notes/content: `src/site/notes/**` Never edit `layouts/*.njk`, plugin-core `components/*.njk`, or `_data/` internals. 2. **Autoload, don't wire.** DG's `src/site/_data/dynamics.js` globs `components/user/common//*.njk`, sorts **alphabetically**, and injects into layout slots (`{% for imp in dynamics.common. %}`). To control render order use a filename prefix (`zzz-` = last, `aaa-` = first). No layout edit needed. 3. **Theme via tokens, not hardcoded color.** Style with DG theme variables (`--background-secondary`, `--background-modifier-border`, `--text-muted`, `--text-accent`, `--link-color`, `--input-shadow`) so light/dark works automatically. Colors are never a knob. 4. **Accessibility is not optional.** Real text in `aria-label`, `aria-hidden` on icons, `role="button"` + `tabindex="0"` + Enter/Space handlers for non-`` controls, visible `:focus-visible`, and `prefers-reduced-motion` respected. 5. **Consolidate tunables.** Put every safe size/shape value in one `>>> TUNING KNOBS >> TUNING KNOBS / ├── SKILL.md ├── assets/ ← drop-in files the target agent copies verbatim └── references/ ← annotated docs, loaded on demand ``` Copy `assets/SKILL.template.md` to `dg-/SKILL.md` and fill it in. ### 3. Write the frontmatter (spec-compliant) Only these top-level keys validate: `name`, `description`, `license`, `compatibility`, `metadata`, `allowed-tools`. **Nest `author`, `version`, `risk-level`, `tags` under `metadata:`**; top-level `version`/`author` fail `skills-ref validate`. See `references/dg-mechanics.md`. `description` is the **only field the agent reads to decide activation** - write it as a trigger: first clause = what it does, then "Use when [X]" with concrete trigger phrases, plus a "Trigger on: …" list of literal keywords. It's discovery metadata, not marketing. ### 4. Write the body (progressive disclosure) Keep `SKILL.md` under ~500 lines. Structure: `When to Use` → `Inspirations` (when any prior work informed the feature) → `Prerequisites (verify first)` → `Quick Install (agent-driven)` → step-by-step `Instructions` → `Key Design Points` → `Risk Level` → `Reference Files` / `Assets`. Push verbose annotated source into `references/` and link each with a read-me-when note. Assets are drop-in files, never auto-loaded. The **Inspirations** section is required whenever the skill is inspired by, tributes, or packages prior public work. Ground every claim in a live public URL; name what each source contributed; exclude private provenance. See invariant 6 and `dg-dungeon-map` / `dg-summon-cats` / `dg-obsifetch` for models. Include a **Prerequisites** section that verifies the target repo is actually DG+Eleventy with the autoloader present before touching anything (see `dg-floating-tray` for the model). ### 5. Populate assets + references - `assets/` - the exact files the target agent will copy (`.njk` component, `.scss` styles) with generic placeholder links/URLs the installer customizes. Put the TUNING KNOBS block on top of any SCSS. - `references/` - one file per concern: annotated markup/scripts, annotated styles, a `tuning.md` of every knob, and a `verify.md` post-build QA checklist. Mirror `dg-floating-tray`'s set. Before moving on, make a de-slop pass over everything you wrote (invariant 7): strip narrating comments, defensive scaffolding, and filler prose. Assets ship verbatim into user repos, so any slop you leave is slop every installer inherits. ### 6. Label the risk level State an L0–L3 risk level in both the description and a `Risk Level` section. DG feature skills that only write user-owned files with no shell/secrets/network are **L1** - say so explicitly so installers can trust it. ### 7. Validate & package ```bash npx skills-ref validate skills/dg-/ # pass the directory, not a file cd .. && zip -r dg-.zip dg- # bundled resources => zip the whole dir ``` SKILL.md-only skills can ship as the bare `.md`; anything with `assets/`/`references/` ships as a zip of the whole directory. Keep it ≤100 files / ≤70 MB. Don't ship secrets or `node_modules`. ### 8. Add to the zoo + docs (this repo's pattern) If publishing to `skills-zoo`: 1. Drop the folder in `skills/dg-/`. 2. Add a `` block to `docs/index.html` and a `docs/skills/dg-/index.html` detail page (copy an existing one as a template). 3. Build the target Digital Garden and capture the feature's real rendered output in both themes. Save 900×560 screenshots as `docs/skills/dg-/preview-dark.png` and `preview-light.png`; reference them from both pages. Never substitute generated art or a hand-built mockup for product output. 4. Add a row to the README skills table. 5. Commit + push - the Pages workflow auto-deploys `docs/**`. ## Key Design Points - **Meta, not installer** - this skill *produces* skills; it doesn't install features itself. - **Invariants over templates** - the seven invariants above are the real product; the template is a convenience. A skill that violates invariant 1 or 2 is broken regardless of how it looks. A skill that adapts prior work without invariant-6 citations is incomplete, and one shipped without an invariant-7 de-slop pass carries its slop into every repo that installs it. - **Verified spec facts** - frontmatter rules and progressive disclosure come from the [agentskills.io](https://agentskills.io) spec; see `references/dg-mechanics.md` for citations. ## Risk Level **L1 (low).** Scaffolds and writes skill files into a skills directory and runs the local `skills-ref validate` check. No other code execution, no secrets, no network calls. Reversible via git. The skills it *produces* carry their own risk labels - enforce step 6. ## Reference Files - `references/authoring-workflow.md` - full annotated walkthrough of steps 1–8 with examples. - `references/dg-mechanics.md` - DG plugin internals (autoloader, slots, theme tokens, page modes) and the verified agentskills.io frontmatter/disclosure rules with sources. ## Assets - `assets/SKILL.template.md` - starter frontmatter + body skeleton for a new DG feature skill. ## Validate & Package ```bash npx skills-ref validate skills/dg-skill-authoring/ zip -rq dg-skill-authoring.zip dg-skill-authoring ``` ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [palol](https://github.com/palol) - **Source:** [palol/skills-zoo](https://github.com/palol/skills-zoo) - **License:** MIT - **Homepage:** http://skillz.paologabriel.com/ 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-palol-skills-zoo-dg-skill-authoring - Seller: https://agentstack.voostack.com/s/palol - 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%.