Update Docs

Ensures documentation stays in sync with code changes. Run before or during every PR that touches library code.

Before anything else

1. Read CLAUDE.md — understand boundaries (core vs server), dropped features, and session model
2. Read .docs-style-guide.md — binding terminology reference

Step 1: Detect what changed

bash
1git diff main...HEAD --name-only

Map changes to affected docs:

If no library code changed (docs-only PR), skip to Step 4.

Step 2: Read the changed code

For each changed file:

1. Read the file and the diff (git diff main...HEAD -- <file>)
2. Identify: new public APIs, changed signatures, new classes, removed exports, changed behavior

Step 3: Update affected docs

For each affected page:

1. Read the current doc
2. Skip if already correct
3. Update code examples, descriptions, YAML examples
4. Verify "When to use this" section exists (see .docs-style-guide.md page structure pattern step 3):
   • Every page MUST have a ## When to use this section after the opening/example and before the main content
   • If missing, add one with: 2-4 concrete scenarios (real situations, not abstract descriptions), user personas who benefit, and how this feature relates to other Edictum features
   • If new code adds a feature, update the scenarios to cover the new capability
   • Read the ACTUAL SOURCE CODE for the feature before writing scenarios — reference real method names, real classes, real behavior
5. Verify terminology against .docs-style-guide.md:
   • "contracts" not "policies" or "rules"
   • Use denied (see .docs-style-guide.md for banned alternatives)
   • Use enforces (not "governs")
   • Use pipeline (not "engine")
   • Use tool call (not "function call")
   • Use adapter (not "integration" or "plugin")
   • Use observe mode (see .docs-style-guide.md for banned alternatives)
   • Use finding / findings (not "alert")
6. Verify core vs server boundaries against CLAUDE.md:
   • All contract evaluation (pre, post, session, sandbox) is core
   • StdoutAuditSink, FileAuditSink, OTel are core
   • Production approval workflows (ServerApprovalBackend) require the server
   • Centralized audit dashboards require the server
   • Multi-process session tracking requires the server
   • MemoryBackend is the only local StorageBackend (no Redis/DB)
   • No references to dropped features or ee/ tier

Step 3.5: Update repo-level markdown files

These files live outside docs/ but track code changes:

1. CHANGELOG.md — if the PR introduces user-visible changes (fixes, features, breaking changes), add an entry under the current version heading. Use the existing entry format. Keep descriptions neutral (no exploit details for security fixes).
2. CLAUDE.md "What's Shipped" section — if this is a new version, add a one-line entry to the version history list matching the existing format.

Step 4: Update README if needed

If public API, install extras, framework support, or version changed:

1. Read README.md
2. Update affected sections
3. Ensure README matches docs/index.md positioning

Step 5: Verify

bash
1pytest tests/test_docs_sync.py -v

Step 6: Report

Summarize:

• Which code files changed
• Which doc pages were updated (and what changed)
• Which doc pages were checked but needed no changes
• Build verification result

Rules

• Don't rewrite for the sake of rewriting. Only update what the code change actually affects.
• Don't add features that don't exist. If code was added but not released, note it as unreleased.
• Don't reference dropped features. No Redis/DB StorageBackend, no reset_session().
• Preserve the voice. Match existing style — problem first, short paragraphs, code examples.
• Check cross-links. Verify links in updated pages still work.
• README and homepage must stay aligned. If you update one, check the other.