updating-internal-docs

Updating Internal Documentation

Review internal documentation files against the actual codebase state and propose fixes for outdated, incorrect, or missing information.

When to use

• After significant codebase changes (new features, refactors, tooling updates)
• When documentation drift is suspected
• After updating make targets, folder structure, dependencies, skills, or workflows
• When a PR adds or modifies Streamlit features — check if bundled skills (lib/streamlit/.agents/skills/) need updates

Key files to check

Priority files (most likely to contain codebase-specific instructions):

• **/AGENTS.md - AI agent instructions
• **/README.md - Package/directory documentation
• .claude/skills/*/SKILL.md - Skill definitions for Streamlit library development
• .claude/agents/*.md - Subagent definitions
• wiki/**/*.md - Developer wiki
• CONTRIBUTING.md - Contributor guide
• lib/streamlit/.agents/skills/AGENTS.md - Authoring instructions for bundled skills
• lib/streamlit/.agents/skills/*/SKILL.md - Bundled skills for Streamlit app development (shipped with the library)
• lib/streamlit/.agents/skills/*/references/*.md - Reference docs for bundled skills

Files to skip (synced copies, updated separately):

• .github/copilot-instructions.md
• .github/instructions/*.md
• .cursor/rules/*.mdc
• .claude/agents/reviewing-local-changes.md from ## Review Checklist onward (generated from scripts/assets/code-review-instructions.md)

If you edit a source AGENTS.md or scripts/assets/code-review-instructions.md, run uv run python scripts/generate_agent_rules.py so generated copies stay in sync.

Verification checklist

• Make commands exist and work (make help)
• File and folder paths exist
• Tool/dependency references are valid
• Tool version numbers match config files (see below)
• Testing instructions are correct
• Code examples match actual patterns
• Links resolve (internal and external)
• Skill/agent cross-references use current names
• .github/workflows/AGENTS.md reflects actual workflow files
• CONTRIBUTING.md skill/agent overview matches .claude/skills/*/ and .claude/agents/
• Bundled skills (lib/streamlit/.agents/skills/) reflect current Streamlit API and features
• Conventions in docs are not already fully enforced by lint, format, type-check, Knip, or other CI checks (if they are, treat as REDUNDANT and omit or remove)

Bundled skills and feature changes

When a PR adds or changes a Streamlit feature (new widget, API change, deprecation, new capability), check if the bundled skills need updates:

• Read lib/streamlit/.agents/skills/AGENTS.md before editing bundled skills. It decides which features get prominent guidance and how to update references, examples, routing, and public API summaries.
• Reference docs in lib/streamlit/.agents/skills/developing-with-streamlit/references/ — update the relevant existing reference to document the new feature or API change

For periodic reviews, treat recently merged PRs as leads for documentation drift. Inspect those diffs, then verify the current code before updating docs. A merge does not by itself require a bundled-skill update; apply the prominence and scope rules in lib/streamlit/.agents/skills/AGENTS.md.

Common triggers for bundled skill updates:

• New st.* commands or widgets
• Parameter changes to existing commands
• Deprecated APIs or patterns (add warnings, remove outdated examples)
• New layout or theming capabilities
• Performance-related changes (caching, fragments)

Quick verification commands

# Check path exists: test -e path && echo ok || echo missing
# Check URL reachable: curl -sI -o /dev/null -w "%{http_code}" <url>

Tool version sources

Issue types

Workflow

1. Enumerate: Find all markdown documentation files
2. Verify: Cross-reference documented commands, paths, and examples against the codebase
3. Report: Present findings grouped by priority
4. Fix: Apply changes after user approval

Presenting findings

List all issues and let the user choose which to fix:

Documentation Review: {SCOPE}
═══════════════════════════════════════════════════════════════

Found {N} issues across {M} files:

1. [OUTDATED] AGENTS.md:42
   Current:  `make python-check`
   Actual:   Command renamed to `make python-lint`

2. [INCORRECT] wiki/testing.md:15
   Current:  Tests in `lib/tests/unit/`
   Actual:   Path is `lib/tests/streamlit/`

3. [BROKEN_LINK] CONTRIBUTING.md:88
   Current:  Link to `./docs/setup.md`
   Actual:   File does not exist

4. [REDUNDANT] frontend/AGENTS.md:20
   Current:  Documents a specific oxlint/eslint/ruff rule (e.g. type-only imports)
   Actual:   Already enforced by lint/CI; omit from docs

Which issues should I fix?
Recommended: "all"
Options: "1" | "1,2,3" | "all" | "skip 3"

Rules

• Verify before proposing: Always check the codebase before suggesting a fix
• Minimal changes: Only change what's actually wrong
• Keep all documentation selective and brief: Not every codebase detail needs to be documented. Add information only when it is relevant to developer decisions, correct usage, maintenance, or preventing likely mistakes; do not expand docs with minor details merely for completeness.
• Do not document conventions already enforced by CI or linting: If a formatter, linter (ruff, oxlint, eslint), type checker, Knip, or other CI check already fails or auto-fixes a convention, do not add it and remove it if it is already documented. Confirm the named rule exists and is enabled before treating a convention as redundant. Agents and developers will see the tool error anyway. Document only judgment calls, exceptions, and "what to use instead" that the tool message does not explain. How to run those tools (make targets, when to use them) remains useful.
• Prefer durable, high-level descriptions: Describe make commands and workflows briefly in terms of their purpose, trigger, and when to use them. Avoid documenting individual implementation steps, options, or mechanics unless they are important for correct use or maintenance.
• Test commands: Run commands before documenting them
• Keep style consistent: Match existing documentation style

After completing review

1. Present all findings to user
2. Get approval before making changes
3. Apply fixes incrementally
4. Run /checking-changes to validate

Example summary:

Fixed 3 of 4 issues:

- #1 [OUTDATED]: Updated make command in AGENTS.md
- #2 [INCORRECT]: Fixed test path in wiki/testing.md
- #3 [BROKEN_LINK]: Removed dead link in CONTRIBUTING.md
- #4 [INCONSISTENT]: Skipped - requires manual verification

Files modified:
  AGENTS.md         |  2 +-
  wiki/testing.md   |  4 ++--
  CONTRIBUTING.md   |  1 -