Writing skills and prompt templates

SLICC follows the Agent Skills specification and pi's prompt templates. Skills written for pi or for other Agent Skills tools work here unchanged, as long as the commands they use exist in SLICC.

A skill is a folder with SKILL.md

~/.pi/agent/skills/pdf-tools/
├── SKILL.md
├── scripts/extract.sh
└── references/formats.md

---
name: pdf-tools
description: Extract text and tables from PDF files. Use when reading, converting, or inspecting PDFs.
---

# PDF tools

Read `references/formats.md` before converting a document. Run scripts relative to this skill directory.

• name: lowercase letters, digits and hyphens, at most 64 characters, no leading, trailing or double hyphens. Without it, the folder name is used. Match the folder name to stay portable.
• description (required, at most 1024 characters): this is all you see of a skill until you load it, so say what it does and when to use it. "Helps with PDFs" is too vague to route on.
• disable-model-invocation: true keeps a skill out of your list; only /skill:name loads it.
• license, compatibility, metadata and allowed-tools are accepted. allowed-tools doesn't restrict anything here.

Refer to bundled files with paths relative to the skill folder; the skill's location tells you where that is.

Where SLICC looks

In this order; when two skills share a name, the first one found wins and the other is reported:

1. ~/.pi/agent/skills/
2. ~/.agents/skills/
3. /usr/share/skills/, global skills every cone and scoop sees; it's root's, so write there as root
4. packages under /node_modules whose package.json lists folders in pi.skills
5. SLICC's built-in skills in /var/lib/slicc/agent/skills/ (rewritten on every start; don't edit them there)

Folders containing SKILL.md are found at any depth; a plain .md file with a description directly in one of these folders counts as a skill too. Folders starting with . and node_modules are skipped. A project's .pi/skills or .agents/skills folder is read only once the folder is trusted, which SLICC doesn't support yet.

Changes are picked up at once. A skill with a problem (no description, a bad name, a name collision) is reported in a lick that names the file.

Prompt templates

A prompt template is a Markdown file that becomes a / command the user types:

---
description: Review staged git changes
argument-hint: "[focus]"
---
Review the staged changes. Focus on ${1:-correctness, security, and error handling}.

Saved as ~/.pi/agent/prompts/review.md, it's /review. Only .md files directly in that folder count. Packages can add templates with pi.prompts, and SLICC ships /parallel-review and /review-loop.

Substitutions: $1, $2, …; $@ or $ARGUMENTS for all arguments; ${1:-default}; ${@:-default}; ${@:N} and ${@:N:L} for slices. Arguments are split like a shell does, so /review "API compatibility" is one argument.

Commands

• /skill:name [request] loads the skill and adds the request after it.
• /template [arguments] expands the template.

Both only work when the user types them in a chat; text that arrives in licks or from agent send is never expanded.

What v6 skills need

v6 skills keep working where their commands exist. Many refer to v6 paths (/shared/…, /workspace/…) or v6-only commands (ipk, .jsh, playwright-cli, upskill, sprinkle, mount, …); rewrite those parts before relying on them.