Help us improve
Share bugs, ideas, or general feedback.
From agent-skills
Use when defining AI agents declaratively with Agent Definition Language (ADL). Covers agent identity, LLM configuration, tools, permissions, RAG inputs, and governance metadata. USE FOR: declarative agent blueprints, agent identity and permissions, LLM configuration, governance metadata DO NOT USE FOR: agent runtime orchestration (use cagent), tool integration (use mcp), agent communication (use a2a)
npx claudepluginhub tyler-r-kendrick/agent-skills --plugin agent-skillsHow this skill is triggered — by the user, by Claude, or both
Slash command
/agent-skills:adlThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
ADL is a vendor-neutral, declarative specification for defining AI agents — their identity, capabilities, tools, permissions, and governance metadata. It acts as a portable blueprint (like OpenAPI for APIs) that is independent of any runtime, framework, or vendor. Open-sourced by Next Moca under Apache 2.0, also adopted by Eclipse LMOS.
AGENTS.mdREADME.mdmetadata.jsonrules/_sections.mdrules/_template.mdrules/adl-declare-dependencies-on-other-agents-explicitly-so-the.mdrules/adl-define-permissions-explicitly.mdrules/adl-include-governance-metadata-owner-review-status-for.mdrules/adl-keep-system-prompts-in-adl-rather-than-hardcoded-in.mdrules/adl-use-adl-alongside-mcp-tool-runtime-and-a2a.mdrules/adl-use-semantic-versioning-so-dependent-systems-can-track.mdCreates and validates production-grade agent .md files for Anthropic 2026 16-field spec. Use for custom subagents, agent quality review, or orchestrator architectures. Triggers: /agent-creator, 'create an agent'.
Creates new Claude Code agent definition files using agent-almanac templates and registry conventions. Covers persona design, tool/skill selection, model choice, frontmatter schema, and symlink verification. Use for specialized subagents or library additions.
Write and configure custom agent definitions in Claude Code's agents/ directory, including system prompts, tools, context forking, delegations, and restrictions for specialized agents.
Share bugs, ideas, or general feedback.
ADL is a vendor-neutral, declarative specification for defining AI agents — their identity, capabilities, tools, permissions, and governance metadata. It acts as a portable blueprint (like OpenAPI for APIs) that is independent of any runtime, framework, or vendor. Open-sourced by Next Moca under Apache 2.0, also adopted by Eclipse LMOS.
ADL defines what an agent is and what it can do, not how it runs. It complements:
adl: "1.0"
agent:
name: research-assistant
version: "1.0.0"
description: "Researches topics and produces structured summaries"
author: "team-name"
license: MIT
llm:
provider: anthropic
model: claude-sonnet-4-5-20250929
temperature: 0.3
max_tokens: 4096
system_prompt: |
You are a research assistant. Produce well-structured,
factual summaries with cited sources.
tools:
- name: web-search
type: mcp
server: "search-server"
description: "Search the web for information"
- name: read-document
type: mcp
server: "doc-server"
description: "Read and parse documents"
rag:
- name: knowledge-base
source: "vector-store://company-docs"
description: "Internal documentation and policies"
permissions:
allowed_tools:
- web-search
- read-document
denied_actions:
- file_write
- code_execution
boundaries:
max_tokens_per_request: 8192
max_requests_per_minute: 30
dependencies:
- name: fact-checker
type: agent
description: "Validates factual claims before including them"
governance:
owner: "research-team@company.com"
review_status: approved
last_reviewed: "2026-01-15"
tags:
- research
- internal
| Field | Description |
|---|---|
name | Unique agent identifier |
version | Semantic version |
description | What the agent does |
author | Creator or team |
license | SPDX identifier |
| Field | Description |
|---|---|
provider | Model provider (anthropic, openai, etc.) |
model | Specific model ID |
temperature | Sampling temperature |
max_tokens | Maximum output tokens |
system_prompt | System-level instructions |
Declares which tools the agent can use, with type and connection info.
Declares knowledge sources the agent can access for retrieval-augmented generation.
Defines what the agent is allowed and forbidden to do, plus rate limits and resource caps.
Other agents or services this agent relies on.
Ownership, review status, and audit metadata for enterprise compliance.
| Aspect | ADL | AGENTS.md | Agent Skills |
|---|---|---|---|
| Focus | Agent definition (identity + config) | Project-level coding guidance | Reusable capability packaging |
| Format | Structured YAML schema | Freeform Markdown | YAML frontmatter + Markdown |
| Scope | Per-agent blueprint | Per-project instructions | Per-skill instructions |
| Runtime | Framework-agnostic | N/A | Platform-agnostic |