Terminology

One-sentence definitions of the vocabulary used throughout the Canopy docs. Each entry links to the section that covers it in depth — start here when a term shows up in Concepts or Reference and you want a quick anchor before diving in.

Skills & spec

• Agent Skill — a SKILL.md package distributed per the agentskills.io standard. → Reference — Framework Skills
• Canopy-flavored skill — an Agent Skill that uses Canopy’s ## Tree section; requires canopy-runtime to execute. → Concepts — The tree as source of truth
• agentskills.io — the cross-vendor Agent Skill specification Canopy is layered on top of. Same SKILL.md, same install tooling. → Concepts — agentskills.io alignment
• SKILL.md — the required entry-point file at the root of a skill directory; case-sensitive, uppercase. → Reference — Framework Skills
• Frontmatter — YAML metadata block at the top of SKILL.md (name, description, compatibility, metadata, etc.). → Concepts — Skill anatomy
• Compatibility field — frontmatter free-text declaration (≤500 chars) naming runtime dependencies and install tools. → Reference — Compatibility Field
• Safety preamble — runtime-required guard block at the top of a ## Tree skill’s body that halts execution on agents without canopy-runtime. → Concepts — Skill anatomy
• metadata — frontmatter sub-object that holds non-spec fields like argument-hint and user-invocable. → Reference — Framework Skills

Tree & ops

• Tree — the sequential execution pipeline under ## Tree; nodes run top-to-bottom with explicit branching. → Reference — Tree Execution Model
• Tree notation — the << (input), >> (output), | (separator) symbols used in tree nodes. → Reference — Notation
• Op — a named, reusable step identified by an ALL_CAPS name (DEPLOY, VERIFY, ROLLBACK). → Concepts — Ops
• Op call — a tree node that invokes an op by name with optional << inputs and >> outputs. → Reference — Tree Execution Model
• Op definition — the body that implements an op, written as prose or as a sub-tree, in references/ops.md or references/ops/<name>.md. → Reference — Skill-Local references/ops.md
• Primitive — a built-in op (IF, SWITCH, FOR_EACH, ASK, SHOW_PLAN, VERIFY_EXPECTED, etc.) provided by canopy-runtime; never overridable. → Reference — Primitives
• Op lookup chain — the resolution order for an ALL_CAPS identifier: skill-local → consumer-defined → framework primitives. → Reference — Op Lookup Order

Subagents

• Subagent — an isolated context window invoked from the parent skill; reads files, distills results, and returns a schema-shaped summary. → Concepts — Subagents
• ## Agent — the declaration block in SKILL.md that describes the explore subagent’s task. → Concepts — Subagents
• EXPLORE — the runtime-injected primitive that invokes the declared ## Agent subagent; required as the first tree node when ## Agent is present. → Concepts — Subagents
• Schema — JSON contract at assets/schemas/<name>.json defining the shape of a subagent’s output. → Reference — Category Resource Subdirectories

Framework skills

• canopy-runtime — execution engine; interprets canopy-flavored skills, owns the primitives spec, ships the per-platform runtime rules. Always required. → Reference — Framework Skills
• canopy — authoring agent invoked as /canopy; create / modify / scaffold / validate / improve / advise / refactor / convert. Depends on canopy-runtime. → Reference — Framework Skills
• canopy-debug — optional trace wrapper invoked as /canopy-debug <skill>; adds phase banners and per-node tracing. → Reference — Debug Mode

Activation & runtime

• Activation — the one-time process that registers canopy-runtime with the host platform by writing the marker block to its ambient instructions file. → Reference — Activation
• Marker block — the canopy-runtime activation block written into CLAUDE.md (Claude Code) or .github/copilot-instructions.md (Copilot). → Reference — Activation
• Skills root — the directory holding installed skills; resolved at runtime as .agents/skills/ → .claude/skills/ → .github/skills/. → Reference — Skills Root Resolution
• Runtime spec — the per-platform execution rules loaded by canopy-runtime based on the active host (Claude Code or GitHub Copilot). → Reference — Runtimes
• Interpreter model — Canopy’s cross-platform approach: SKILL.md is the single source of truth, the runtime walks the tree at execution time and adapts to the active host. → Reference — Runtime Model

Resources & layout

• Standard layout — the agentskills.io directory layout: only SKILL.md at root, with scripts/, references/, and assets/ as the three top-level subdirectories. → Reference — Framework Skills
• Category resource — a structured file inside a category subdir (assets/templates/, assets/constants/, assets/schemas/, etc.); referenced from the tree by Read \/\``. → [Reference — Category Resource Subdirectories](/claude-canopy/reference/framework-spec/#category-resource-subdirectories)
• Legacy flat layout — older skills with category dirs at the skill root (schemas/, templates/, commands/, …) instead of nested under assets/; still fully supported. → Reference — Framework Skills