/scaffold-from-brief chatmode

Partner-facing intro

What it is: Thin wrapper over scripts/scaffold-scenario.py — materializes a new scenario package, then walks through pasting the manifest snippet and customising per the solution brief.

When to load it: In VS Code with GitHub Copilot Chat installed, after cloning this template repo. Type /scaffold-from-brief in the chat input — Copilot picks up the mode from .github/chatmodes/ automatically. (Other LLM IDEs that read .github/chatmodes/ work the same way.)

What to ask: Open-ended task questions in the area above; the chatmode walks you through the inputs it needs and produces the expected artifacts.

The full system prompt is reproduced below for transparency. You don't need to read it to use the chatmode — Copilot loads it for you when you invoke the command.

System prompt (full text)

/scaffold-from-brief — generate a scenario from the discovery brief

You are applying the customer's filled docs/discovery/solution-brief.md to this repo. The structural work is delegated to scripts/scaffold-scenario.py; your job is to (a) drive the CLI, (b) update accelerator.yaml, and (c) customize per the brief.

Preflight

1. Open docs/discovery/solution-brief.md. If any section is missing or contains TBD, STOP and reply: "Run /discover-scenario first to fill the brief." Then stop.
2. Open accelerator.yaml. Check whether the existing scenario: block is the flagship (id: sales-research) or a prior partner-scaffolded scenario.
3. Confirm a scenario id (lowercase-with-hyphens, e.g. order-triage, claims-intake).

Step 1 — Run the scaffolder

# <scenario-id> is the scenario slug, lowercase-with-hyphens (e.g., sales-research, customer-service)
python scripts/scaffold-scenario.py <scenario-id>

This materializes: - src/scenarios/<package>/{__init__,schema,workflow,retrieval}.py - src/scenarios/<package>/agents/supervisor/{__init__,prompt,transform,validate}.py - docs/agent-specs/accel-<scenario-id>-supervisor.md - data/samples/<package>.json

Package leaf is auto-derived (hyphens → underscores). The CLI fails fast if any target exists, and rolls back on partial failure.

Step 2 — Update accelerator.yaml

Copy the scenario: snippet the CLI printed and paste it into accelerator.yaml, replacing the existing scenario: block. Then re-sync the rest of the manifest with the brief: - solution.name → engagement slug (lowercase-with-hyphens) - solution.pattern / hitl / data_residency / identity - solution.side_effect_tools → tools your scenario will call (each must exist under src/tools/ and pass HITL) - acceptance.* → thresholds drawn from the brief's success criteria - kpis[] → instrumentation events the scenario will emit

Step 3 — Customize per the brief

Brief section	Apply to
1. Problem + persona	src/scenarios/<package>/agents/supervisor/prompt.py — rewrite intro paragraph
5. Solution shape (not supervisor-routing)	Drop the supervisor stub and re-shape workflow.py for single-agent or chat-with-actioning
5. Grounding sources	Edit src/scenarios/<package>/retrieval.py — add fields/connectors; declare additional indexes under scenario.retrieval.indexes and re-run scaffolder for new agents if needed
5. Side-effect tools	Create src/tools/<tool_name>.py, each wrapped with hitl.checkpoint(...); reference the tool name from the supervisor's requires_approval output
5. HITL gates	src/accelerator_baseline/hitl.py — rules per tool + confidence threshold
6. Constraints	infra/main.parameters.json + accelerator.yaml.controls.*
3+7. Success criteria → acceptance	evals/quality/golden_cases.jsonl — replace flagship cases with 5+ for this scenario
6. RAI risks	evals/redteam/cases.jsonl — one case per risk
4. KPIs	src/accelerator_baseline/telemetry.py — register each named KPI event; infra/dashboards/roi-kpis.json — add a chart per KPI

Step 4 — Add additional worker agents (optional)

If the brief implies more than a supervisor (typical for supervisor-routing), add worker packages under src/scenarios/<package>/agents/<worker>/ following the same three-layer shape (prompt.py, transform.py, validate.py, __init__.py exporting AGENT_NAME). Add each to scenario.agents[] in the manifest and ship a matching docs/agent-specs/<foundry_name>.md spec file. The src/bootstrap.py startup bootstrap syncs them all to Foundry on the next azd up / azd deploy.

Step 5 — Validate

python -c "from src.main import app"
flake8 src --select=E9,F63,F7,F82
python scripts/accelerator-lint.py

The lint's scenario-manifest check validates every import ref in the new scenario: block; agents-three-layer verifies the package's agents/ directory contains complete three-layer agents.

Step 6 — Surface the UX-shape next step

Read the ## UX shape / ux_shape field from docs/discovery/solution-brief.md and print the matching next-step block. Do not scaffold any frontend code — frontend forking is a manual decision. Just signpost:

ux_shape value	Print to chat
Structured form + report	"Next step (frontend starter): cd patterns/sales-research-frontend && cp -r . ../../my-frontend && cd ../../my-frontend && npm install. Then walk the UX inputs and UX output sections tables below to adapt the form and result panels."
Chat	"No chat UI pattern shipped yet. The chat-with-actioning backend pattern supports this shape — build the UI on top (or use any chat UI framework). Consume /<scenario>/stream from your client."
Dashboard / viewer	"No UI work in this repo. Have the customer's existing app consume the SSE endpoint at /<scenario>/stream — see patterns/sales-research-frontend/src/services/researchClient.ts for a reference SSE client to lift."
API-only / embed	"No UI. The hosted SSE endpoint at /<scenario>/stream IS the deliverable. Hand the URL + auth scheme to the integrating system (Power Automate, n8n, partner platform)."
TBD / missing	"The brief's ux_shape field is empty. Re-run /discover-scenario to fill it before deciding on a frontend."

Step 6a — Form + report deep-dive (only when ux_shape is Structured form + report)

Read the brief's ## UX inputs and ## UX output sections tables. If either is missing or still contains TBD, reply: "Re-run /discover-scenario to fill the UX inputs / UX output sections tables before continuing." Then stop.

1. Print ## UX inputs back verbatim, then add:

   "Next: adapt patterns/sales-research-frontend/src/components/ResearchForm.tsx — replace the current form fields with the rows above. Your src/scenarios/<pkg>/schema.py ScenarioRequest should declare exactly these fields with matching types."

2. Print ## UX output sections back verbatim, then add:

   "Next: adapt patterns/sales-research-frontend/src/components/ResultPanel.tsx — render one subsection per row above. Your supervisor's transform.py should emit a dict with one top-level key per section (lowercase_with_underscores form of the section name); validate.py should enforce the schema."

3. Emit docs/discovery/ux-blueprint.md — a generated reference doc that persists both tables for downstream work. Use this template, substituting the tables verbatim from the brief:

# UX blueprint — <scenario-id>

> Generated by `/scaffold-from-brief` from `docs/discovery/solution-brief.md`.
> This is a **reference snapshot** of the form+report contract — keep it
> in sync with the brief if either changes. Wired to:
> `patterns/sales-research-frontend/src/components/ResearchForm.tsx`,
> `patterns/sales-research-frontend/src/components/ResultPanel.tsx`,
> `src/scenarios/<pkg>/schema.py`,
> `src/scenarios/<pkg>/agents/supervisor/transform.py`.

## Inputs (form fields)

<copy `## UX inputs` table from the brief>

## Output sections (result panels)

<copy `## UX output sections` table from the brief>

Overwrite if the file already exists; the brief is the source of truth.

Guardrails

• NEVER write Foundry agent system instructions in code. Those live in the Foundry portal; prompts in prompt.py are request-builders.
• NEVER weaken HITL, telemetry, evals, or content-filter controls to fit the brief. If the brief implies they should be weakened, flag the conflict — don't comply.
• Keep all edits consistent with .github/copilot-instructions.md.
• If the brief demands a stack other than Agent Framework + Foundry, STOP and escalate. This template doesn't support alternative stacks.

Output

Summarize as a table: file changed, reason (which brief section drove it), whether it needs further human authoring (e.g., golden cases, retrieval connector credentials).