Plan & Act Dual-Mode

Most AI coding errors occur when an assistant attempts complex multi-file edits without first understanding the codebase topology and agreeing on a plan. Agento enforces a strict boundary between planning and execution.

The Dual-Mode Model

Agento operates in two distinct modes: Plan Mode and Act Mode.

┌────────────────────────────────────────────────────────┐
│                      PLAN MODE                         │
│  - Codebase exploration & research                     │
│  - Architecture formulation & dependency mapping       │
│  - Todo list & milestone definition                    │
│  - MUTATIONS BLOCKED (Write/Edit tools disabled)       │
└──────────────────────────┬─────────────────────────────┘
                           │ User approves plan
                           ▼
┌────────────────────────────────────────────────────────┐
│                      ACT MODE                          │
│  - Governed task execution                             │
│  - In-memory pre-write rule checks (LINTOs)            │
│  - Code modification & file updates                    │
│  - Automated post-write verification (Checks)          │
└────────────────────────────────────────────────────────┘

Plan Mode: Research and Design

In Plan Mode, the agent focuses entirely on discovery, research, and formulating an actionable plan:

Read-Only Safety: The agent cannot create, edit, or delete files, nor can it execute destructive shell commands.
Context Synthesis: The agent reads relevant files, analyzes dependencies, and formulates a step-by-step plan document with clear milestones and acceptance criteria.
Review Gate: The generated plan is presented in the Plans Hub (F2). You can review, refine, or request adjustments before approving it.

Act Mode: Governed Execution

Once a plan is approved, switching to Act Mode (Tab or Ctrl+X m) unlocks code modification tools under strict guardrails:

Task-Driven Work: The agent works methodically through approved todos and milestones.
Pre-Write Verification: Every proposed edit is checked against active LINTO rules before touching disk.
Post-Write Diagnostics: Compilers and test runners verify each change immediately, providing instant feedback if a syntax error or regression is introduced.
Milestone Completion: When all tasks in a milestone are finished, Agento runs Tier-2 verification checks to confirm the codebase is healthy.