Guide

Environment

Environment is the project's descriptive knowledge — what exists now. It captures tech stack, source structure, build configuration, domain knowledge, and code dependencies. Unlike Genome (prescriptive — how to build), Environment describes the current state.

2-Tier Structure

text

.reap/environment/
├── summary.md      # Always loaded (~250 lines) — tech stack, source structure, build, tests
├── domain/         # Domain knowledge (on-demand)
├── resources/      # External reference docs — API docs, SDK specs (on-demand)
├── docs/           # Project reference docs — design docs, specs (on-demand)
└── source-map.md   # Code structure + dependencies (on-demand)

Tiers

Tier	Loading	Content	Limit
summary.md	Always loaded at session start	Tech stack, source structure, build config, test setup. The AI's baseline understanding.	~250 lines
domain/	On-demand (loaded when needed)	Domain knowledge — business rules, API specs, infrastructure details.	No limit
resources/	On-demand (loaded when needed)	External reference documents — API docs, SDK specs, third-party documentation.	No limit
docs/	On-demand (loaded when needed)	Project reference documents — design docs, specs, architecture decisions.	No limit
source-map.md	On-demand (loaded when needed)	Current code structure and dependency map.	No limit

Environment Immutability

Like the Genome, the Environment is never modified directly during a generation. Changes are recorded as environment-change backlog items and applied during the Completion reflect phase.

By capturing changes in the backlog rather than rewriting Environment mid-generation, the generation completes on a stable map. The update happens once, deliberately, during the reflect phase with full context of what was built.

Loading Strategy

summary.md is always loaded at session start. domain/ and source-map.md are loaded on-demand when the AI needs deeper context for a specific task.

Knowledge Management

Use /reap.knowledge to review and manage the Environment. During the Completion reflect phase, the AI automatically refreshes the environment to reflect changes made during the generation.

summary.md

Always loaded

Compact overview of the project's technical state. Loaded into every session so the AI has baseline context.

domain/ + resources/ + docs/ + source-map.md

On-demand

Deeper knowledge loaded when the AI needs specific domain, external reference, or structural context for the current task.

The 2-tier strategy balances context window efficiency (summary.md is small) with depth (domain/ and source-map.md are available when needed).