llms-keeper

name

llms-keeper

description

This skill should be used when the user asks to "generate llms.txt", "create llms-full.txt", "update llms.txt", "regenerate llms-full.txt", "create AI context files for this project", "обновить llms.txt", "сгенерировать llms-full.txt", "создать контекст проекта для AI", or mentions "llms.txt", "llms-full.txt", "llmstxt", "llmstxt.org standard", "project context for AI agents", "контекст проекта для AI". Also applicable when user asks about the llmstxt.org format, wants to know how to structure llms.txt, or asks "what should llms-full.txt contain".

llms-keeper — Project Context for AI Agents

Generate and maintain llms.txt and llms-full.txt files following the llmstxt.org standard to give AI agents quick project understanding.

Core Philosophy

"Find the smallest set of high-signal tokens that maximize the likelihood of desired outcome."

Every sentence must answer: "Will an AI agent produce worse results without this information?" If no — delete without hesitation.

Two Files, Two Purposes

File	Purpose	Audience	Length
llms.txt	Navigation index	AI needing quick orientation	50-100 lines
llms-full.txt	Complete context	AI needing full understanding	200-600 lines

Signal vs Noise

Always Include (High-Signal)

• Project purpose and tech stack
• Architecture: module hierarchy, dependency direction
• Data flow patterns (ASCII diagrams for complex flows)
• Key patterns used across 3+ files
• Configuration approach and entry points
• Common pitfalls with solutions
• Essential development commands
• Testing approach and conventions

Never Include (Noise)

• Step-by-step tutorials or implementation guides
• Temporary workarounds, TODOs, FIXMEs
• Patterns used in only 1 file
• Verbose explanations ("This module is responsible for...")
• Obvious information for experienced developers
• Credentials, API keys, secrets

Workflow

Delegate analysis and generation to the documentation-keeper agent.

The agent handles the full process: framework auto-detection, codebase scanning, pattern extraction, file generation, and quality verification.

For manual or informational use, the workflow is:

1. Scan project structure and detect tech stack from manifest files
2. Read key sources: entry points, core modules, configs, tests, docs
3. Extract high-signal information (architecture, patterns, pitfalls)
4. Generate llms-full.txt first (self-contained content)
5. Derive llms.txt from it (navigation index linking to sections)
6. Verify quality checklist from references/llmstxt-spec.md

Update vs Generate

• No llms.txt exists → full generation from scratch
• llms.txt exists → analyze git changes since last update, update only affected sections, verify consistency between both files

Reference

For format details, structure requirements, and quality checklist, see references/llmstxt-spec.md.