For coding agents
Agent-ready recipes
Copy focused implementation prompts for common Noir tasks with the right documentation context and verification gates.
These prompts are intentionally specific about outcome and authority. Attach the linked Markdown pages rather than asking a coding agent to infer Noir from package names.
Add a Slack agent
Implement a Noir Slack agent using the provider-native noir() API and our existing
OpenRouter model. Keep credentials in environment variables. Use the included Slack
adapter, a durable RuntimeStore, explicit limits, and structured runtime events.
Configure only mention events and required scopes. Verify type checking, a signed
webhook fixture, duplicate delivery, one thread reply, and a failed model call.Context: Quickstart, Production Slack agent, Configuration, Execution model, ChannelAdapter, Security.
Add a provider tool
Add a namespaced connector around the provider's official SDK. Validate input,
project output, declare effect/approval/retry, forward the abort signal, and use the
Noir idempotency key for writes. Keep the client and token in application code.
Add contract tests for authorization, malformed input, retry classification, and
projection. Do not expose the raw SDK response to the model.Context: Create a connector, Tools and effects, Tool definition, Testing agents.
Add a database adapter
Implement RuntimeStore using the existing ORM and migration system. Preserve atomic
insert-once behavior, lease ownership, checkpoint first-writer-wins, requester-bound
approval transitions, bounded indexed history, and outbox settlement. Add competing
worker and duplicate-event tests. Do not replace the project's database client.Context: Bring your own database, Storage and durability, RuntimeStore.
Add a sandbox provider
Implement SandboxDriver over the selected provider. Map verified Noir owner scope to
a workspace, contain paths, enforce file/output/time limits, forward cancellation,
restrict environment and network access, persist leases, and verify file writes.Context: Sandbox, Sandbox drivers, Security model.