Skill library

A skill is an instruction you write once and let every agent read. It lives in your organization's own file tree as a small bundle — a SKILL.md carrying the instruction in its body, plus any reference material that instruction leans on — and Settings > Skills is where you create, upload, and maintain those bundles. Every member can create skills; what you may edit is decided per bundle.

This page covers what a skill is, the file it is made of, who gets to see it, and how you add and retire one. Read the agent side on Skills on agents once you want a particular agent to reach for a particular bundle.

What a skill is, and what it is not

A skill is a knowledge pack. Its body is instruction a model reads when the work calls for it: a house writing voice, a checklist your team follows, the way your organization phrases a refusal. A model finds the bundle by its description, reads the body when that description matches the task at hand, and opens individual bundle files when the body points at them.

A skill is never something the platform executes. There is no entry point, no command, and no runtime in a bundle — a file under scripts/ is material a model may read and adapt, not a program Tale runs on your behalf. That boundary is what makes a bundle safe to accept from outside: importing someone else's skill hands your organization prose and reference files, and nothing that can act on its own.

The SKILL.md file

Every bundle has exactly one SKILL.md at its root — a YAML frontmatter block, then the instruction body in markdown.

---
name: release-notes
description: Turn a list of merged changes into release notes in our house voice. Use when someone asks for a changelog, release notes, or a summary of what shipped.
visibility: team
teams:
  - jx7d…
license: CC-BY-4.0
---

Write release notes as three sections — Added, Changed, Fixed — and lead each
line with the verb...

The keys follow the agentskills.io convention in kebab-case, and any key Tale does not recognise is kept verbatim, so a bundle authored for another tool survives an edit and a save unchanged.

Two ceilings apply: the frontmatter block may run to 16 KB, and the whole SKILL.md to 512 KB. Bundle assets sit outside that budget.

Who can see it

Sharing is one field rather than a table of permissions. visibility: team shares the bundle with the teams listed under teams — pick them in the library's Visibility section. visibility: org means every member sees it, and any project's agents can equip it. Any member may share a skill team- or organization-wide; editing or deleting someone else's shared skill takes an org admin. A bundle carrying no visibility at all — including one you upload — counts as an organization skill, and the upload preview says so before you confirm.

Narrowing a skill's sharing — org to team, or dropping a team — asks for confirmation first: whoever loses sight of the skill also loses it in every agent that equipped it through them.

Add a skill

Open Settings > Skills. The page is a table of every skill you may see — its name, description, visibility, and labels — with search over all three of name, description and labels, and filters for visibility and label. Clicking a row opens the bundle. Add skill offers three starting points.

1. 1

   Start blank

   Blank skill asks for a name — the slug, in lowercase letters, numbers, and single hyphens — plus the description, the sharing, and an instruction body you can write on the spot. New skills start shared with the organization; narrow the sharing to teams where the knowledge is theirs.

2. 2

   Or upload a bundle

   Upload zip takes a .zip with SKILL.md at its root, alongside any scripts/, references/, or assets/ folders; Upload folder takes the folder itself and zips it for you. Either way Tale reads the frontmatter before writing anything and shows you what it found — the description, the sharing it will land with, the license, and a full file list with sizes — so you approve a bundle you have actually read. A bundle whose slug already exists asks first whether to replace it.

3. 3

   Write the body

   Open the skill and write the instruction under Instructions (body). This is the text a model reads, so write it the way you would brief a colleague: what the skill is for, when it applies, and what good output looks like.

What sits in the bundle

A skill's detail view shows Bundle — the file tree as it exists on disk — with a viewer for any file you click. The smallest useful skill is a single file, and most grow one folder at a time.

release-notes/
├── SKILL.md
├── references/
│   └── voice-and-tone.md
└── scripts/
    └── group-changes.py

Keep the assets small and readable. Text a model can open cheaply gets used; a large binary sits there unread, and the viewer says outright that it cannot preview it.

Retire a skill

Delete skill on the detail view removes the bundle from disk; every agent equipped with it loses access, with nothing to fall back on. There is no version pinning — a skill is always read exactly as it stands right now, which is also what makes it worth extracting: one edit reaches everyone who holds it.

Where this fits

The skill library is the lightest reuse Tale offers: one file, one field for sharing, and nothing to keep in sync across the people who need it. It is where a phrasing you keep retyping stops being something you retype. Once a bundle is in the library, the remaining decision is which agents get it — that is Skills on agents, which covers equipping a project's agents and how a bundle reaches a sandbox.