Glossary

Short definitions of the terms used across the documentation. Use the search box (top) to jump to one.

Term	What it means
ADR (Architecture Decision Record)	A short document recording a technical decision, its context and consequences. In Tramalia they live in docs/ai/05.
AGENTS.md	Standard file (Linux Foundation) with the project rules that all AI agents read. Single source of truth.
AI agent	A tool that reads context, reasons, edits files and/or runs commands (Claude Code, Codex, Cursor, Antigravity…).
Bootstrap	The base tools you install manually first (mise, git, uv), because they can't install themselves. Once present, mise install brings everything else.
CLI (Command-Line Interface)	A command-line interface: used by typing commands in the terminal.
Copyleft	A license type that requires derived code to stay open (GPL, LGPL, MPL). It doesn't affect Tramalia because the tools are invoked, not linked.
Enforcement	Tramalia blocking a task's close if a gate fails (unless a documented exception with --allow-fail).
Evidence pack	A dated folder with the verifiable proof of a close: commands, raw outputs, risks, rollback, next steps and metadata.json.
Façade	A thin layer that puts a single interface in front of a complex subsystem. The CLI and tramalia mcp are façades.
Fan-out	Propagating a single source (AGENTS.md) to the formats of several agents (Cursor, Copilot…), with rulesync.
Gate (quality gate)	A mandatory validation before closing a task: build, test, lint, security, database, UX/UI.
Handoff	A structured handover between agents/sessions: task → files → commands → result → risks → pending → next step.
Idempotent	Running twice produces the same result without harm. tramalia init is idempotent: it never overwrites existing files.
Interop (interoperability)	Optional external tools that Tramalia orchestrates but does not require; if missing, it still governs and records the exception.
LSP (Language Server Protocol)	A protocol that provides code intelligence (definitions, references). Serena uses it to navigate symbols without reading whole files.
MCP (Model Context Protocol)	A standard protocol to connect AI agents with tools and data. Tramalia exposes an optional MCP façade.
metadata.json	A structured summary of each close (task, agent, timestamps, exit codes, status) that makes the audit queryable.
Moat	A product's defensible differentiator. In Tramalia: evidence pack, handoff, gates and repo-first audit.
N0 / N1 / N2 (memory tiers)	N0 = files + CLI · N1 = MCP façade · N2 = real persistent memory (Engram / basic-memory / mem0).
Ponytail / YAGNI	Minimalism principles: do the minimum correct thing, don't rebuild what already exists, don't over-abstract.
Repo-first	Using the repository as the source of truth: everything important is versioned in it, not hidden in global configs.
SAST (Static Application Security Testing)	Static security analysis of code (done by Semgrep).
Shell-out	Tramalia running an external command (subprocess) and passing its output through as-is, without reimplementing it.
Snapshot	A packaged picture of the repo for AI consumption (done by Repomix).
Stack	The set of technologies in a project (Angular, .NET, PostgreSQL…). tramalia detect identifies it.
Standalone	Works on its own, without depending on anything external. Tramalia's core is standalone (Python only).
Token	The minimal unit of text an AI model consumes. "Saving tokens" = sending less context to cut cost and latency.
Wheel	Python's installable package format (.whl); it's what pip install uses.

Missing a term? It's a good first contribution.