| Action Item |
A discrete task to be completed. (Example: “Request transcript delivery confirmation.”) |
| Action Note |
New information (email, chat, news, etc.) that may change sequencing or validation of Action Items. |
| Action Trigger |
A specific event/input (e.g., a support‑ticket closure) that forces an Action Schema update. |
| AGENTS.md |
A Markdown file containing one or more LLM-only prompt-executable directives. Each AGENTS.md file defines automatic behaviors for language models operating within a given directory scope. These directives are always executed by LLMs, not humans. Root-level AGENTS.md files govern global rules; folder-level files govern local behavior. |
| Backlog |
All Action Items not marked Done ✅. No separate backlog artifact exists. |
| Component |
A reusable building block recorded in the Component Catalog (e.g., library, service, concept). |
| Dependency |
A directional “depends‑on” relationship recorded in Component Dependencies. |
| Dependency Type |
One of White‑box, Black‑box, or Product, defining how much internal knowledge is documented. |
| Directive |
A directive is a structured instruction or behavioral guideline that governs how an AI agent or system component should act, respond, or interact. Directives encode role definitions, constraints, reasoning patterns, or communication protocols, and are used to ensure consistency, safety, and intentionality in LLM-driven workflows.
Directives may be stored in various forms, including inline prompts, file-embedded tags, or dedicated AGENTS.md files (LLM-only). Not all directives require an AGENTS.md. |
| Entry Point |
A dependency designated as a starting node that links to one or more Use Cases. |
| Fail Until It Works |
Mind‑set of iterating rapidly—accepting early failure and leveraging LLM creativity until validity emerges. |
| Goal / Sub‑goal |
An objective defined in GOALS.md, recursively decomposed until it is actionable and testable. |
| LLM Precision Expectation |
Reminder: LLM output may contain errors; treat versions/outputs as approximations, validate critical details manually. |
| Mission |
The top‑level purpose statement that all Goals roll up to. |
| Optional Task |
A non‑blocking Action Item that can enhance a Rotation but is not strictly required. |
| Problem Space |
The portion of reality relevant, visible, and solvable now. Plans beyond this horizon are ignored until re‑validated. |
| Rotation |
The LLM‑driven loop that reviews Notes & Triggers, (re)generates the Action Schema, and increments its version. |
| Snapshot |
Any immutable, versioned artifact (Action Schema, GOALS.md version, etc.) produced during a Rotation. |
| Status |
Allowed progress states: Not started, In progress ⏳, Done ✅, Pending, Optional. |
| Templates |
A template is a structured artifact that encodes the rules, language, and expected behaviors for both humans and language models within the Grit Labs system. Templates include prescriptive formats (such as action schemas), shared vocabularies (terminology), and behavioral protocols (directives), enabling aligned execution, traceable reasoning, and reproducible decision-making.
Templates may be human-authored, LLM-generated, or collaboratively maintained—and serve as versioned source-of-truth references for performing Rotations, tracking goals, managing dependencies, or updating actions. |
| Test Case / Validation Check |
Concrete proof (file, URL, automated test) that an Action Item or Goal is complete. |
| Use Case |
A description of what a Product‑type Component does; linked to an Entry Point in dependency modeling. |
| Version Line |
The single‑line PascalCase title followed by “— vX.Y” that prefixes every Action Schema snapshot. |
| We Only Know What We’re Doing Now |
Acceptance that only the next action is trusted; future plans are discarded unless re‑validated. |