Glossary

Glossary

Quick definitions for terms used throughout the wiki. Linked to the pages that go deeper.

A

ADR (Architecture Decision Record) — Immutable doc capturing a single architectural choice and its rationale. Superseding requires a new ADR. See Specs.

Agent — A defined role with model, tools, and operating principles, invokable by an agent CLI runtime. Autonomous (runs to completion without user input). See Agents.

Agentic coding — Software engineering with AI agents as collaborators that produce reviewable software when given a spec. See Agentic Coding Model.

AGENTS.md — The canonical, vendor-neutral root context file. The single source of truth that every agent CLI working in a SpecRoute-driven repo should read first. See Multi-Vendor Context Files.

Archetype — A role concept ("what does a security agent do?"), not a file the runtime loads. Used when designing your project's roster. See Agents.

Authoring agent — The agent in the implementation team that owns a particular artifact type (e.g. prd-author owns PRDs, spec-author owns specs). See Implementation Team.

C

Command — A simple slash-invoked operation. No parameters (or one argument). Always does the same thing. See Commands.

Coverage table — Mapping in tasks.md of every requirement and NFR to the tasks that satisfy it. Every cell populated; no TODO rows. See Specs.

D

Delegation shim — A short (~30–50 line) per-vendor root context file (CLAUDE.md, GEMINI.md) that points at the canonical AGENTS.md and adds vendor-specific overrides. See Multi-Vendor Context Files.

F

Folder-per-skill convention — Every skill is a directory containing SKILL.md. Never a flat file. See Skills.

Frontmatter — YAML block at the top of a markdown file, fenced by ---. Required fields vary by artifact type. See Frontmatter Contracts.

G

Global master prompt — 000_GLOBAL_MASTER.md, the single entry-point for a phased initiative. Names the role, mission, source-of-truth table, and phase order. See Prompts.

H

Hook — Event-triggered automation. Runs automatically on file edit, session start, pre-tool, etc. Sub-second target. Fails closed. See Hooks.

I

Implementation team — The .claude/ directory at the repo root. Twelve agents, six skills, four commands, three hooks that build SpecRoute itself. Not the consumer template. See Implementation Team.

M

MCP (Model Context Protocol) — A protocol for exposing tools and resources to agent CLIs. All six vendors consume MCP server configs, in two emit shapes: a mcpServers JSON object for everyone except Codex, which uses [mcp_servers] TOML. (The Claude Code CLI reads .mcp.json / ~/.claude.json, not the Claude Desktop app's claude_desktop_config.json.) See MCP Integration.

N

NFR (Non-Functional Requirement) — A requirement that constrains how the system performs rather than what it does (performance, security, observability, etc.). Stable ID format: NFR-<N.M>. See Specs.

P

Phase master prompt — 000_MASTER_<phase>.md, the second tier in the phased master-prompt pattern. Summarizes a phase, lists prerequisites, names agent assignments. See Prompts.

PRD (Product Requirements Document) — The business-intent layer. Full 23-section template or lightweight single-page alternative. See PRDs.

Primitive (automation primitive) — One of the four ways SpecRoute automates work: skill, agent, command, or hook. See Automation Decision Framework.

Prompt — A reusable instruction. Three layers: global master, phase master, task prompt. See Prompts.

R

Requirement ID — Stable, unique identifier (R1.1, R2.3, NFR-1.1). Once published, never reused. Tasks, PRs, and tests back-reference these. See Specs.

Roster (agent roster) — Cross-vendor inventory of a project's agents, grouped by department. See Agents.

Rule — A standing engineering constraint that's always true. Lives in rules/. See Rules.

Runtime — A copy-pasteable per-vendor layout consumers drop into their own repos. See Vendor Matrix and Agent CLI Integrations.

S

Sanitization — Keeping proprietary content (private project names, customer data, secrets, internal endpoints) out of tracked content. Enforced via wordlist + hook + commands + agent review. See Sanitization.

Skill — An interactive, parameterized workflow. User invokes by name; skill guides through decisions and validation checkpoints. See Skills.

Spec triplet — requirements.md + design.md + tasks.md. The technical contract for a feature with stable IDs and back-references. See Specs.

Stable ID — See "Requirement ID."

Steering — Kiro's term for rule files loaded at session start (.kiro/steering/*.md). Two modes: inclusion: always and inclusion: fileMatch. See Rules.

T

Task prompt — A production-grade prompt instantiating the shape: Objective / Context / Agent Assignment / Prerequisites / Task Details with current→target diff blocks / Acceptance Criteria. See Prompts.

Trigger phrase — A literal user utterance (in quotes within an agent's description frontmatter) that triggers automatic agent selection. See Agents.

Two-tier docs — Short root context files (loaded into every conversation) + deep references in agentic-docs/ (read on demand). See Two-Tier Docs Pattern.

V

Vendor matrix — The table of supported agent CLIs and what each consumes (runtime dir, root context file, skills, agents, commands, hooks, MCP). Adding a new tool means a new column. See Vendor Matrix.

Vendor neutrality — A contract: don't fold one vendor into another, don't treat any as default, every artifact uses each vendor's native shape. See Philosophy.

W

Workflow — An end-to-end engineering execution model. A verb (procedure), not a noun (standard). See Workflows.

Worked example — examples/sample-project/. Drop-in runnable demonstration that exercises every artifact shape. See Worked Example.

See also

• Home — start here
• Repository Structure — what lives where
• Frontmatter Contracts — the required-field reference