/teams-app-developer

researcher

purpose

Step-by-step workflow for fleshing out any stub micro-expert. Use this when the user says "research expert X" or when you encounter a stub that needs content.

how to identify a stub

A file is a stub if its ## instructions section contains only a web-search placeholder and its ## rules section is missing or empty.

workflow

step 1 — read the stub

Read the target expert file. Locate its ## research section and extract the Deep Research prompt.

step 2 — execute research

Run the Deep Research prompt as a series of web searches:

1. Break the prompt into discrete topics (each SDK concept, API surface, or pattern mentioned).
2. Search for each topic individually. Prefer:
   • Official documentation (learn.microsoft.com, api.slack.com, SDK GitHub repos)
   • SDK source code and type definitions
   • Recent blog posts or guides (add current year to query if results are stale)
3. For each search, capture: API signatures, parameter types, return types, default values, and gotchas.

step 3 — synthesize into canonical sections

Replace the stub content with real content using these sections:

• ## rules — Numbered list of do/don't rules derived from the research. Each rule should be actionable (e.g., "Always call ack() before async work" not "ack is important").
• ## patterns — Code snippets (TypeScript) showing canonical usage. Use fenced code blocks. Keep each snippet minimal and focused on one concept.
• ## pitfalls — Common mistakes, breaking changes, or version-specific gotchas.
• ## references — URLs to official docs, SDK source, or authoritative blog posts used during research.

step 4 — update instructions section

Replace the ## instructions placeholder content with a concise summary of what the expert covers and when to use it. This is the "quick reference" an agent reads first.

step 5 — preserve the research prompt

Keep the ## research section intact with the original Deep Research prompt. This allows future re-research if the SDK changes.

step 6 — rollup to domain index

Open the domain's index.md and verify:

1. The expert file appears in the correct task cluster's Read: list.
2. The When: description for that cluster still accurately reflects the expert's content (update if the scope changed during research).
3. The expert file appears in the ## file inventory list.

quality checks

• Every ## rules entry must cite a source (doc link or SDK behavior).
• Every ## patterns code snippet must be valid TypeScript that compiles in isolation (imports included).
• No fabricated API signatures — if you cannot confirm a signature, note it as unverified.
• Keep the total file under 300 lines. Split into multiple experts if it grows larger.