Specification
Look System
kanonak.org
agent-skills@1.2.0
agent-skills
Activation Method
The mechanism by which an agent activates a skill.
Agent File Format
On-disk format a client expects for agent/subagent definition files. Different AI agent products have converged on different representations for the same…
Agent File Format
The on-disk file format this client reads for agent definitions. Installers must write each Agent in this format when deploying to this client. Named…
Agents (cross-client default)
The cross-client interoperable default. Any AI agent product that reads from the .agents/ tree can consume skills and agents deployed under this client. Pick…
Allowed Tools
Optional, experimental. Tools that are pre-approved to run without user confirmation when this skill is active. Serialized to SKILL.md as a space-separated…
Asset Resource
Static resources in assets/ such as templates, images, or data files.
Bash
Execute shell commands.
Body
Full markdown content for this section, including prose, code blocks, tables, and lists. This is rendered directly in the SKILL.md output.
Catalog
At session start, agents load only the name and description of each available skill - just enough to know when it might be relevant.
Claude Code
Anthropic's Claude Code CLI and desktop agent. Reads skills from .claude/skills/ and subagent definitions from .claude/agents/ at both project and user scopes.…
Client
An AI agent runtime or product that discovers and uses skills (and, from 1.2.0, agent/subagent definitions) from known on-disk locations. Each Client instance…
Client ID
Short identifier used on the command line to select a client when installing a skill or agent (e.g., "agents", "claude-code", "codex", "gemini"). Lowercase…
Command
A shell command to execute this script.
Compatibility
Optional. Max 500 characters. Indicates environment requirements such as intended agent product, required system packages, or network access.
Content
The text content of this resource.
Description
Required. 1-1024 characters. Describes what the skill does and when to use it. Should include keywords that help agents identify relevant tasks.
Disclosure Tier
A stage in the progressive disclosure architecture that controls how much of a skill is loaded into agent context and when.
Discovery Path
A filesystem location where agents scan for installed skills. Superseded in 1.1.0 by the Client class. Retained for backward compatibility.
Discovery Scope
Whether a discovery path is project-level or user-level.
Edit
Make targeted edits to existing files.
File Read Activation
Model-driven activation where the agent uses its standard file-read tool to read the SKILL.md file.
Gemini CLI
Google's Gemini CLI. Reads subagent definitions from .gemini/agents/ as Markdown files with YAML frontmatter, the same format Claude Code uses. Skill discovery…
Glob
Find files by pattern.
Grep
Search file contents by pattern.
Has Asset
Embedded static assets within this section. Provides structured metadata for programmatic extraction alongside the body content.
Has Reference
Embedded reference material within this section. Provides structured metadata for programmatic extraction alongside the body content.
Has Resource
A supporting file bundled with this skill. Use for standalone resources not tied to a specific section. For section-local resources, use…
Has Script
Embedded executable scripts within this section. Provides structured metadata for programmatic extraction alongside the body content.
Has Section
Ordered list of content sections that form the skill's documentation. Each section becomes a headed section in the generated SKILL.md. Sections are rendered in…
Instructions
When a task matches a skill's description, the agent reads the full SKILL.md body into context.
Language
The programming language of the script (e.g., bash, python, javascript).
License
Optional. The license applied to the skill. May be referenced by name (e.g., "license: MIT") or embedded inline for custom licenses.
Markdown with YAML Frontmatter
A plain Markdown file beginning with a YAML frontmatter block (delimited by triple-dash lines) that carries the agent's name, description, tools, model, and…
Metadata
Optional. Arbitrary key/value metadata on the skill, modeled as a list of MetadataEntry instances. Clients can use this to store additional properties that the…
Metadata Entry
A single key/value pair in a skill's metadata map. The Agent Skills spec allows arbitrary string-to-string metadata on a skill; each entry in that map is…
Metadata Key
The key portion of a metadata entry.
Metadata Value
The value portion of a metadata entry.
Name
Required. 1-64 characters, lowercase letters, numbers, and hyphens only. Must not start or end with a hyphen, must not contain consecutive hyphens, and must…
OpenAI Codex
OpenAI's Codex CLI. Reads skills from the cross-client .agents/skills/ tree and reads subagent definitions from .codex/agents/ as TOML files. The TOML schema…
Path Pattern
The filesystem path pattern for this discovery location.
Path Scope
Whether this path is project-level or user-level.
Project Agent Directory
Relative directory path, from the project root, where this client expects project-scoped agent (subagent) definitions to live. The installer writes one file…
Project Agents Skills
Cross-client interoperable skills in the project directory.
Project Client Skills
Client-specific skills in the project directory.
Project Scope
Skills specific to a project or repository.
Project Skill Directory
Relative directory path, from the project root, where this client expects project-scoped skills to live. The installer appends the skill's name to this path to…
Read
Read file contents.
Reference Resource
Additional documentation in references/ that agents can read on demand.
Resource Title
A short title describing this resource.
Resources
The agent follows the instructions, loading referenced scripts, documentation, and assets on demand as needed.
Script Resource
Executable code in scripts/ that agents can run.
Skill
A directory containing a SKILL.md file that provides an AI agent with instructions, scripts, and resources to perform a specific task. Follows the open Agent…
Skill File Name
File name the client reads to discover a skill's frontmatter and instructions inside each skill directory. Defaults to "SKILL.md" for clients that follow the…
Skill Resource
A supporting file bundled with a skill.
Skill Section
A section of skill content with a title, markdown body, and optional embedded resources. Sections are ordered within a skill and each becomes a headed section…
Tier Order
The sequence number of this tier in the progressive disclosure architecture.
Title
The section heading. Rendered as a level-2 markdown heading (##) in the generated SKILL.md.
Token Budget
Recommended token cost for this tier.
TOML
A TOML file where all agent configuration - including the developer instructions that serve as the system prompt - lives under named keys, not separated into a…
Tool
A capability that an agent can invoke, such as reading files, running shell commands, or searching code. Skills declare which tools they are pre-approved to…
Tool Activation
Model-driven activation via a dedicated activate_skill tool registered by the agent runtime.
User Agent Directory
Directory path, rooted at the user home directory, where this client expects user-scoped agent definitions to live. The literal "~" at the start is expanded to…
User Agents Skills
Cross-client interoperable skills in the user home directory.
User Client Skills
Client-specific skills in the user home directory.
User Explicit Activation
User-initiated activation via slash command (/skill-name) or mention syntax ($skill-name).
User Scope
Skills available across all projects for a user.
User Skill Directory
Absolute-ish directory path, rooted at the user home directory, where this client expects user-scoped skills to live. The literal "~" at the start is expanded…
Web Fetch
Fetch content from a URL.
Web Search
Search the web.
Write
Write or create files.