Documentation Lifecycle#
Documentation should move with product and protocol changes. Use this workflow for new features, changed behavior, releases, and larger content updates.
1. Intake#
Capture docs needs as soon as engineering work starts:
- What changed?
- Who is affected?
- What does the reader need to do differently?
- Which pages, examples, screenshots, or API references need updates?
- What source of truth verifies the change?
For engineering changes, link docs work to the issue, pull request, release note, EIP, or repository discussion.
2. Draft#
Write from tested behavior. Prefer small, reviewable pull requests:
- Add or update the main task page.
- Add concept context only when needed.
- Update navigation and related links.
- Add troubleshooting notes for expected failure modes.
- Include verification steps.
3. Technical Review#
Ask a subject matter expert to check:
- Technical accuracy.
- Missing edge cases.
- Version or network assumptions.
- Security, privacy, or funds-at-risk warnings.
- Whether examples work as written.
4. Content Review#
Check the page against Content Standards:
- Reader goal is clear.
- Structure fits the page type.
- Terminology is consistent.
- Links and navigation help discovery.
- The page is concise enough to use during a task.
5. Publish#
Before merge:
mkdocs build
python tools/nav_audit.py
If the page changes navigation, inspect the local site with:
mkdocs serve
6. Maintain#
Docs need owners after launch:
- Review high-traffic and high-risk pages regularly.
- Update pages when linked repositories, APIs, wallets, or services change.
- Remove stale workarounds after fixes ship.
- Convert repeated support questions into troubleshooting entries.
- Use search and analytics signals to find pages readers cannot discover.
- Use Source Watch metadata on pages tied to source repositories.
AI-Assisted Drafting#
AI can speed up docs work, but human review remains required.
Good uses:
- Summarizing source issues or pull requests into a draft outline.
- Finding stale terms across the docs.
- Suggesting missing prerequisites, warnings, or troubleshooting entries.
- Producing first-pass examples for review.
Use the AI Docs Review Prompt for a structured first-pass review of pull request diffs.
Do not publish AI output without checking it against source code, command output, release notes, or subject matter review.
AI review checklist:
- Claims trace back to a source.
- Commands were tested or clearly marked as examples.
- No invented flags, endpoints, APIs, versions, or product names.
- No secrets, private data, or copied third-party text were introduced.