Documentation

Claude Plans and Context

Genesis adapts scaffold complexity based on your available context window. Every Claude Code session includes a context window measuring tokens, encompassing the system prompt, CLAUDE.md, settings, agents, skills, memory files, conversation history, and file contents. When context fills, Claude Code automatically compacts (summarises) the conversation, preserving key points but losing detail.

Plan comparison

MetricProMaxProMax
Context200k200k1M
ProfileLeanStandardFull
Domain agents1–22–33–4
Dynamic skills1–22–33–4
CLAUDE.mdCondensedStandardDetailed
Memory1 file3 files3 files
Scaffold cost~5–8k (2.5–4%)~10–15k (5–7.5%)~15–25k (1.5–2.5%)
Working context~190k~185k~975k
SessionsModerateModerateExtended
Rate limitsStandardHigherHighest

API users have variable context windows depending on the model. Genesis requests your context budget during setup.

Pro: Lean scaffold

200k tokens context; the scaffold consumes roughly 2.5–4% of that budget, leaving approximately 190k tokens for your work.

What you get:

  • Full CLAUDE.md (condensed format) with project rules, standards, conventions
  • Three mandatory workflow agents: test-runner, code-reviewer, doc-writer
  • One or two domain agents (the single most critical for your project type)
  • Four base skills: /test, /lint, /review, /commit
  • One or two domain-specific skills
  • Consolidated memory (user profile and project context in a single MEMORY.md)
  • Statusline showing context usage percentage
  • Full application scaffold (entry points, package config, test setup, docs)

You have everything you need to build your project productively. Longer sessions may require the /compact command. The statusline helps you monitor usage.

Max: Standard scaffold

200k tokens context (same as Pro) with higher rate limits and more daily messages. The scaffold consumes 5–7.5% of the context budget.

Everything in the lean scaffold, plus:

  • Full CLAUDE.md template with clearly separated sections (language, error handling, code standards, testing, documentation)
  • Two or three domain agents covering primary and secondary concerns
  • Two or three domain-specific skills
  • Full memory files: separate MEMORY.md index, user-profile.md, project-context.md

ProMax: Full scaffold

1M tokens context; five times the Pro/Max window. The scaffold consumes just 1.5–2.5% of this budget, leaving approximately 975k tokens for your work.

Everything in the standard scaffold, plus:

  • Detailed CLAUDE.md with extended context sections and project-specific rules
  • Three or four domain agents covering primary, secondary, and supporting concerns
  • Three or four domain-specific skills
  • Widest selection from the agent catalogue
This is the plan the author uses and tests against. Extended sessions without compaction, simultaneous reading and analysis of large files, thorough code reviews with full codebase context, and complex multi-step workflows without context loss.

API key users

Not subscription-based; billed per token. Genesis asks for your context budget during setup:

  • 200k context or below: standard profile (default)
  • Above 200k: full profile
  • You can override the recommendation regardless of budget

Context budget breakdown

ComponentLoadedCostIn Lean?
CLAUDE.mdAlways~2–4kYes (~2k)
settings.jsonAlways~300–500Yes
MEMORY.mdAlways~200–400Yes (consolidated)
user-profile.mdAlways~200–400No (inline)
project-context.mdAlways~300–600No (inline)
Workflow agent (×3)On spawn~600–800 eachYes
Domain agent (×1–4)On spawn~800–1200 eachYes (1–2)
Skill (×5–8)On invoke~400–800 eachYes (5–6)
MCP schemasAlways (if configured)~500–2000If needed

"Always in context" components are permanently loaded. Agents and skills load on demand, consuming context only when used.

Monitoring context usage

Every generated project includes a statusline showing the current model and context usage percentage:

[Claude Sonnet 4.6] ctx: 34%

This updates in real time. Usage guidance:

  • 50–60%: Consider whether everything loaded is still needed
  • 60–75%: Run /compact to summarise the conversation and free space
  • 75%+: Claude Code auto-compacts soon; run /compact proactively for a cleaner summary

The /context command provides a detailed breakdown of what's consuming your context.

Changing your profile

After a plan upgrade

Open environment.md in the Genesis root directory and update the Claude Plan section:

## Claude Plan

- **Tier:** ProMax
- **Context window:** 1000000
- **Scaffold profile:** full

New projects will use the updated profile. Existing projects keep their original scaffold.

Per-project override

During the planning phase (Phase 2), request a different profile: "Keep this one lean, it's a simple script" or "Give me the full scaffold, I want all the agents."

Enriching existing projects

  1. Generate a new throwaway project with the same stack at the higher profile
  2. Copy the additional agent files from .claude/agents/
  3. Copy the additional skill directories from .claude/skills/
  4. Update the project's CLAUDE.md to list the new agents and skills

Alternatively, write new agent and skill files directly using the templates in Genesis's .claude/skills/genesis/templates/ as a format reference.

auto_awesome