README.md

BEMYAGENT.md

Mission: Save tokens for the machine. Save orientation for the human.

📖 Website: bemyagent.md

BEMYAGENT.md is a lightweight, self-bootstrapping protocol that bridges the gap between humans and AI agents. Instead of forcing alignment through code reviews or rigid procedures, it creates a shared workspace where the machine thinks in structured files and the human validates at the right level of abstraction.

The Problem

When working with AI agents on complex projects, three things break down:

Context bloat — The agent reads thousands of irrelevant lines, inflating costs and slowing down.
Silent drift — The agent executes a task but drifts from the original intent. Nobody catches it until it's too late.
Validation fatigue — The human must review every line of output because there's no structured checkpoint between "done" and "delivered".

The Solution: TTEV Workflow

BEMYAGENT.md provides a single markdown file (BEMYAGENT.md) that acts as a bootstrap prompt. When fed to an AI assistant, it generates a structured .bemyagent/ workspace:

.bemyagent/docs/ — Permanent project memory (architecture, code map, tech stack, decisions).
.bemyagent/work/ — Tactical, volatile memory organized as a Hierarchical Task Network (HTN).

Core Concepts

Concept	What it does
TTEV Workflow	Think → Task → Execute → Verify. A four-phase cycle where the agent strategizes, plans atomic steps, executes, and self-validates before notifying the human.
Lazy Loading	The agent never reads specs, drafts, or decisions during context restoration unless the current task explicitly requires them. Saves tokens by default.
Fractal Decomposition (HTN)	If a task is too large, the agent decomposes it into sub-tasks (e.g., `work/1/1.1/`, `work/1/1.2/`). Each leaf node gets its own TTEV cycle.
Context Saturation Check	Before executing, the agent verifies it has enough context (target files, expected behavior, constraints, dependencies). If too much is unclear, it asks instead of guessing.
Contextual DNA Mapping (CDM)	During planning, the agent embeds validation criteria directly into each task — scaled by complexity. Simple tasks get none; complex tasks get Drift sensors, Validation criteria, and Pivot triggers.
Symbiotic Validation	After execution, the agent evaluates its own output against the CDM criteria and produces a verdict (PASS / PASS_WITH_CAVEATS / FAIL) before presenting results. The human validates the sense, the agent has already validated the form.
Self-Registration	The agent configures the project's native rule files (`.cursorrules`, `AGENTS.md`, etc.) to read `00-ai-rules.md` at every session start.

Pacing Modes

The human controls how much autonomy the agent has:

SEAMLESS — The agent runs TTEV automatically. It only stops if verification finds issues.
INTERACTIVE — The agent pauses after THINK (plan approval) and after VERIFY (result approval). Two human gates.
AUTO-CLI — The agent switches AI models per phase (e.g., large model for THINK, fast model for EXECUTE).

Usage

Drop BEMYAGENT.md into the root of your project.
Ask your AI assistant to read the file and execute its instructions.
The AI generates the .bemyagent/ directory structure and templates.
Delete BEMYAGENT.md and start a fresh chat session (the bootstrap context is no longer needed).

That's it. From this point on, the agent reads .bemyagent/docs/00-ai-rules.md at the start of every session and knows how to operate.

How It Works (The Files)

.bemyagent/
├── docs/                          # Permanent project memory
│   ├── 00-ai-rules.md             # The protocol itself (agent reads this first)
│   ├── 01-overview.md             # What the project does, quick start
│   ├── 02-architecture.md         # System diagram, component roles
│   ├── 03

bemyagent