Skills

Skills let you package repeatable expertise once and assign it to agents in Admin → Agents. Each skill is a governed playbook: when to use it, the steps to follow, optional reference files, and which tools may run while the skill is active. Agents see a lightweight catalog on every turn and load the full procedure only when the task requires it, so chat stays focused without sacrificing depth.

Fontana implements the open agentskills.io format, so skills remain portable, versionable, and auditable inside and outside your workspace.

What are Skills?

Modular instruction packages in the agentskills.io format (SKILL.md plus optional bundled files). Assignment is per agent; activation is explicit. Skills are playbooks for how to perform a task, distinct from Knowledge Graph articles that hold what your organisation knows.

Why use skills

Without skills, long procedures either bloat every prompt or live only in an agent’s system message where they are hard to reuse across personas. Skills give you:

• One authored source of truth for a procedure (canvas edits, reconciliations, approval checklists, domain playbooks)
• Per-agent assignment so only the right personas may activate a skill
• Governed tool expansion while a skill session is open, without permanently widening an agent’s allowlist
• Audit-friendly sessions with activation, resource fetches, and close events on the chat transcript

Fontana ships default skills for common canvas and workflow tasks; you can author additional packages in the Skills admin UI or import validated SKILL.md packages from your repository.

Progressive disclosure

Skills use progressive disclosure so context stays lean until the model commits to a playbook:

1. Catalog on each turn - assigned skill names and descriptions appear in agent context so the model can decide whether a skill applies.
2. Activation - when a skill matches the work, the agent calls activate_skill with the skill’s stable id. Only then is the full procedure body injected.
3. Bundled references - while a session is open, the agent can fetch manifest-bound files (patterns, examples, checklists) with fetch_skill_resource instead of inlining every attachment up front.
4. Close - when the playbook is complete, the agent calls close_skill. Provider context redacts closed skill internals on later turns; the transcript retains audit rows for review.

Assignment in Admin → Agents means a skill may be used in that agent’s threads. Unassigned skills are invisible to the model. Activation is always explicit; vector search never auto-activates a skill.

Author a skill package

Each skill is a versioned SKILL.md file with YAML frontmatter and a Markdown procedure body. Fontana validates packages at create, edit, and import (frontmatter keys, allowed tools, resource manifest, and file hashes).

Frontmatter key	Required	What it controls
name	Yes	Human-readable title in the catalog and admin UI
description	Yes	When the model should consider this skill; shown in the catalog before activation
allowed-tools	No	Tool ids from the audited catalog; unioned into the agent’s effective tools while the skill session is active

The Markdown body after the frontmatter holds the full procedure: steps, guardrails, and examples. It is not sent on every turn.

Manage skills under Skills in the Flow sidebar (same admin surface family as Agents and Knowledge Graph). The skill editor shows which agents currently use each package.

Tools during a skill session

An agent’s baseline tool allowlist is configured in Admin → Agents. When a skill session is active, Fontana temporarily unions that agent’s manual tools with the skill’s allowed-tools list, so specialised playbooks can unlock canvas, workflow, or data tools only for the duration of the work. See Tools and MCP for the full catalog.

Multiple skill sessions can be active concurrently; effective tools are the union of the agent baseline and all open skill sessions. When a session closes, its extra tools drop out of the next turn.

Skills and the Knowledge Graph

Skills and the Knowledge Graph solve different problems and stay separate in Fontana:

	Skills	Knowledge Graph
Purpose	Repeatable procedures and playbooks	Governed operational knowledge and reference articles
Content	SKILL.md packages with activation lifecycle	Namespace-scoped documents with version history
Retrieval	Catalog plus explicit activation	Automatic pre-turn injection plus agentic search
Typical use	”How to build this canvas pattern"	"What our NAV approval rules say”

Assign both to the same agent when it needs firm knowledge and structured runbooks for complex tasks.