Microsoft Foundry Foundry Agent Webapp Reviewing Documentation

Documentation Review Guidance

Audit, improve, and maintain documentation quality across the repository. Search for doc files — don't rely on hardcoded paths.

Architecture Maintenance

For ARCHITECTURE-FLOW.md updates, load .github/skills/understanding-architecture/SKILL.md first — it has validation commands and source-of-truth mappings.

SKILL.md Quality Gates

Naming rules (agentskills.io spec):

• ✅ writing-csharp-code (lowercase, hyphens, max 64 chars)
• ❌ WritingCSharpCode | writing--csharp | -writing-code

Required frontmatter: name (must match directory name), description (must explain WHAT and WHEN, max 1024 chars)

Body must include: goal statement, practical code examples, common mistakes, related skill cross-references.

.agent.md Quality Gates

Required frontmatter: name, description, argument-hint, tools, model

Verify: handoffs reference valid agent name values from other agent files.

copilot-instructions.md

Loads on every request — keep it lean. Must have: architecture quick reference, dev commands, agents table, skills table.

Audit Checklists

ARCHITECTURE-FLOW.md

• Mermaid diagrams render without syntax errors
• State machines match appState.ts type definitions
• SSE events match Program.cs Write*Event methods
• Actions match AppAction type union

README.md

• Quick start works for new users
• Commands table is accurate
• Links to sub-READMEs resolve

Skills & Agents

• Valid YAML frontmatter on all files
• SDK versions in examples match *.csproj / package.json
• Cross-references between skills are valid

Cross-Document Consistency

• Architecture tables match across all docs
• Port numbers consistent (5173, 8080)
• SDK versions match actual dependencies

Review Output Format

For each document reviewed:

Document: path/to/file.md — ✅ Good | ⚠️ Needs Improvement | ❌ Broken

Issues: numbered list with suggested fixes

Cross-Reference Issues: broken links, version mismatches

Content Quality Rules

All documentation must state how things work now, not history or rationale for past changes.

Must NOT contain

• Version numbers in prose — reference *.csproj or package.json instead. Hardcoded versions go stale silently.
• Issue/PR tracker links — these are PR artifacts, not reference docs. Exception: links to external tracking if explicitly requested.
• Duplicate information — if a fact is stated in one section, don't restate it in another. Cross-reference instead.
• Narrative or history — "we changed X because Y happened" belongs in commit messages, not docs. State the current behavior.
• "Fallback" or "automatic" language for mutually exclusive paths — if two modes are mutually exclusive, say so explicitly. Don't imply graceful degradation that doesn't exist.
• Stale comments — // TODO (if added) when the thing exists, // backward compat without explaining what it's compatible with.

Must contain (for AI agent effectiveness)

• Non-obvious constraints — things an AI agent would get wrong by reading only the code (e.g., "consent to Azure ML Services, not Cognitive Services")
• Mutually exclusive choices — state which modes exist and that they don't mix
• Error signatures — what error an AI agent will see if it does the wrong thing (e.g., AADSTS65001, AADSTS500131)
• File-to-concept mapping — which file controls which behavior, so agents don't search blindly

Package/SDK version references

• Tables with versions: remove the version column, point to source of truth
• Prose references: say "pre-release" or "beta" without specific version numbers
• If a version matters for API compatibility, state the minimum version and why

Style Rules

• Code blocks: always include language identifier (```typescript not ```)
• Links: relative paths for internal, absolute for external
• Tables: consistent column widths

Constraints

• ❌ Don't change code (only docs)
• ❌ Don't invent features not in codebase

Source: microsoft-foundry/foundry-agent-webapp — distributed by TomeVault.