Chatmodes — quick reference¶
Chatmodes are slash commands you run inside GitHub Copilot Chat in VS Code, against the cloned customer repo. Each one is a structured prompt that drives Copilot through a specific delivery task — discover, scaffold, configure infra, add a tool, etc. They keep you on the supported path and prevent accidental architecture drift.
How to invoke
- Open the customer repo in VS Code.
- Open the Copilot Chat sidebar (
Ctrl+Alt+Ior the 💬 icon). - Type
/— the picker shows every chatmode shipped in.github/chatmodes/. - Pick one and follow the prompts.
The full prompt for each chatmode is published in the navigation under Reference → Chatmodes — open one if you want to see exactly what Copilot will be told.
When to run which — by walkthrough step¶
| Walkthrough step | Chatmode | What it does | Inputs you provide | What it writes |
|---|---|---|---|---|
| 5. Discover with the customer | /ingest-prd (optional) |
Pre-drafts the brief from a customer PRD / BRD / spec | Path to PRD (.md / .txt / .docx / .pdf) |
Draft docs/discovery/solution-brief.md with STATUS: AI-extracted draft banner + per-section evidence comments |
| 5. Discover | /discover-scenario |
Runs the structured discovery interview; fills solution-brief.md and accelerator.yaml |
Live workshop answers, notes, or PRD-draft TBDs | docs/discovery/solution-brief.md (canonical), accelerator.yaml (solution.*, acceptance.*, kpis[]) |
| 6. Scaffold from the brief | /scaffold-from-brief |
Materialises the brief into code, infra, evals, telemetry | Filled solution-brief.md (zero TBDs) |
New src/scenarios/<id>/ package, accelerator.yaml scenario: block, eval golden cases, redteam cases |
| 7. Provision the customer's Azure | /configure-landing-zone |
Picks the landing-zone tier and aligns infra/ accordingly |
Customer environment shape (pilot / mid-market / regulated) | accelerator.yaml -> landing_zone.mode; infra/ shape selection |
| 7. Provision | /deploy-to-env |
Registers a new Azure environment, wires OIDC, dispatches first deploy | Env name (dev / uat / prod), customer Entra app reg, target subscription |
deploy/environments.yaml entry, GitHub Environment, OIDC federated credential |
| 8. Iterate & evaluate | /add-tool |
Scaffolds a side-effect tool with HITL + redteam baked in | Tool name, external system, side-effect category | src/tools/<tool>.py, unit test, redteam case, registration on the right worker |
| 8. Iterate | /add-worker-agent |
Runs scripts/scaffold-agent.py and the manual follow-ups |
Agent id, scenario id, one-sentence capability | New src/scenarios/<scenario>/agents/<agent_name>/ (3-layer module), WORKERS registry entry, Foundry agent spec stub |
| 8. Iterate | /explain-change |
Preflight: maps your current diff to lint rules, evals, deploy steps that will fire | (Reads current git diff) | Read-only readout — does not modify files |
| 8. Iterate | /switch-to-variant |
Walks through re-authoring the scenario as single-agent or chat-with-actioning |
Target variant + scenario id | New src/scenarios/<id>/ package shaped for the variant; manual follow-ups for prompts/tests |
| Any (engagement-wide) | /delivery-guide |
End-to-end engagement co-pilot — answers "what's next?" across the whole motion | Free-form question | No file writes; conversational guidance |
When you can skip a chatmode¶
| Situation | Why you can skip |
|---|---|
| Customer handed you a clean PRD already | Skip /ingest-prd — you can pre-fill the brief by hand if the doc is short |
| Returning engineer, no scenario change | Skip /scaffold-from-brief — the existing scaffold is fine if accelerator.yaml matches the brief |
| Pilot in your own dev sub | Skip /configure-landing-zone — Tier 1 (standalone) is the default and works for a sandbox |
| Tool is read-only (a retriever) | Skip /add-tool — add a module under src/retrieval/ instead; HITL is for side-effects |
| Adding a single agent in the existing scenario | Skip the chatmode and run python scripts/scaffold-agent.py directly — same code path |
Where the prompts live¶
Every chatmode is a Markdown file under .github/chatmodes/ in the repo. The header description: field is what Copilot Chat shows in the picker; the body is the prompt Copilot follows. You can read or fork them — they're plain Markdown, version-controlled, and reviewable.
If you want to add a custom chatmode for your partner practice, follow the existing files as templates and PR it back; the accelerator team welcomes new patterns.