.cursorrules

# microsoft-security-agent-toolkit — agent rules

Read `HANDOFF.md` first. It is the single source of truth for design decisions, conventions, tool surface, and open work.

## Hard rules

- Never log or commit tenant secrets. `.env` stays gitignored.
- Every write tool requires `confirm: true` + separate write-app registration (see `docs/permissions-required.md`).
- Skills only declare `tools_required`; they do not bundle this toolkit.
- Defensive / governance content only — no offensive (red-team) tooling.
- No live-tenant calls in tests, demos, samples — use synthetic data.

## Conventions

- Branch off `main`, PR into `main`. One server / snippet pack / demo per PR.
- Conventional-commit-ish messages with `Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>` when AI-assisted.
- KQL files require the header in `kql-snippets/README.md`.
- MCP server READMEs follow the contract in `HANDOFF.md`.

## Pairs with

[`microsoft-security-skills`](https://github.com/vinayaklatthe/microsoft-security-skills) v0.2.0, 88 skills. Tool ID renames must update `INTEGRATIONS.md` in both repos simultaneously.

## Where to start

Topmost unchecked item from `HANDOFF.md` "Open work". Phase 1 = `sentinel-mcp` implementation.