Generate Copilot Instructions

Concrete, factual list of the detected stack — no guesses, only what's in .discovery/context.json:

• Language(s) and runtime
• Backend framework
• Frontend framework (if present)
• Database(s)
• Package manager, linter, formatter, test runner

3. Architecture

From context.json → arch: style (monolith/layered/microservices), monorepo yes/no, service count. Include the main entrypoints if detected.

4. Domain model

Key entities from docs/discovery/domain-model.md and their owning capabilities. List the top 5–10 entities with a one-line description. This anchors naming conventions.

5. Capability map

List of L1 capabilities from docs/discovery/l1-capabilities.md. One line each: name + one-sentence description. This tells the AI what the codebase does.

6. Coding conventions

Derive from the detected stack:

• File naming: follow language conventions (snake_case for Python, camelCase/PascalCase for TS/Java, etc.)
• Test file placement: follow detected test runner conventions
• Import style: infer from language
• API naming: infer from framework (REST resource names from domain entities, etc.)

7. What NOT to do

Three to five hard constraints for this specific stack:

• e.g., "Do not add a second ORM — the project uses {detected ORM}"
• e.g., "Do not introduce a new package manager — use {detected PM} exclusively"
• e.g., "Do not bypass {detected linter} — all code must pass lint before commit"

Output format

Write valid Markdown. Be specific — use the actual project name, actual entity names, actual tool names throughout. Do not write generic placeholder text.

After writing

Print:

  ✔ .github/copilot-instructions.md written ({N} sections)