Docs Update Or Create

Create or Update Docs

Create minimal docs, avoid duplication and make content discoverable. Treat docs for both human and AI agents together. Follow our principles of progressive disclosure, locality and canonicalization.

📝 Process to follow

1. Understand the topic of knowledge and its purpose
2. Gather existing knowledge on that topic
3. Agree on the best format and location for it to live
4. Write/update the files, inline with what's agreed
5. Validate the results and refine as necessary

📜 Guidelines

These guidelines define some boundaries within which to operate. They do not dictate everything. Follow them, but also use judgement and creativity to find the best way to document knowledge in each case.

❌ BAD:

• Avoid long files containing multiple levels of details
• Avoid overloading context with rarely used knowledge
• Avoid duplicating knowledge in multiple files
• Avoid documenting community-obvious knowledge
• Avoid preserving old structure for its own sake

✅ GOOD:

• Add docs that measurably improve agent performance
• Do document repeatable workflows, commands and behaviours
• Capture what is peculiar to this repo, domain, or team
• Only add docs that solve real problems
• Delete files that have no unique value after changes
• Use emojis, checkboxes and formatting sparingly to make docs more appealing

Keep it short

• Keep it short and direct. Treat 100 lines as a target (not a hard limit)
• Keep each file at one level of detail.
• Link to narrower docs when detail must branch.
• Use short inline examples when they replace explanation.
• Link longer examples instead of embedding them.

Use frontmatter

• Skill and agent definitions should use frontmatter to be discoverable by tools.
• Frontmatter for READMEs is optional. If it's useful for you, include it
• Tool-discoverable docs and skills should have name and description
• Add allowed-tools, or compatibility only when they help.
• Use disable-model-invocation: true for user-invoked skills (commands)
• Use user-invocable: false for agent-only skills

Frontmatter Example for a skill

---
name: desktop-swap-guidance
description: Read when editing swap flows in Ledger Live Desktop.
---

💡 Principles

• Apply principle of progressive disclosure
• Follow our rules on docs locality
• Avoid duplication by applying rules on canonicalization

🔎 Validation

• Make sure links point to the canonical source
• Ensure it is discoverable by agents and humans
• Search for duplicated key phrases with rg
• Check long docs with wc -l.
• Verify AGENTS.md stays minimal.