Skip to content

CONTEXT.md

Latest commit

 

History

History
48 lines (35 loc) · 1.69 KB

CONTEXT.md

File metadata and controls

48 lines (35 loc) · 1.69 KB
Raw
type context
title AI Agent Skills
description Vocabulary and system map for the skills set used to shape AI agent behavior.
tags
vocabulary
architecture
timestamp 2026-06-27

AI Agent Skills

Vocabulary and system map for the skills set used to shape AI agent behavior.

Language

Skill: A discrete capability or discipline an agent can perform, defined by a SKILL.md contract. Avoid: ability, tool (a tool is a technical primitive; a skill is a behavioral discipline)

Trigger: A natural language phrase or condition that causes an agent to invoke a specific skill. Avoid: hook, command

Workflow: A canonical sequence of skills composed to solve a specific type of task (e.g., greenfield feature, bug fix). Avoid: process, path

Substrate: The shared foundation of vocabulary (LANGUAGE.md) and formats (formats/) that all skills consume. Avoid: base, common

Lens: A discipline applied during other skills to maintain quality, rather than being invoked as a standalone phase. May be a skill (e.g., caveman) or a shared reference (e.g., the code-hygiene lens in formats/). Avoid: principle, guideline

Gate: A skill that verifies conditions before allowing a workflow to proceed (e.g., prod-ready, security-review). Avoid: check, blocker

Relationships

  • A Workflow is composed of multiple Skills.
  • Skills consume the Substrate.
  • Lenses are applied during Skills.
  • Gates protect the transition between Workflows and the final release.

Flagged ambiguities

  • "state" was used for both agent session state and project-specific feature state — resolved: "feature state" is scoped to docs/features/<name>/state/.