AI coding tools
Continue Operating Card
Configure Continue agents with explicit models, rules, tools, secrets, and a reproducible workspace-level verification loop.
- Format
- Reference
- Level
- Beginner
- Audience
- Practitioner, Developer, Operator
- Owner
- Project42 Editorial
- Review cadence
- Every 30 days
- Prerequisites
- A supported Continue client; An approved model configuration; A version-controlled working copy
Use configuration as a reviewed agent contract
Continue agents combine models, rules, and tools through local or shared configuration. Choose it when you want an IDE or CLI workflow whose model and behavior can be configured independently of a single provider.
Treat imported blocks, documentation indexes, and MCP tools as dependencies. Review their source, permissions, data path, and update policy before enabling them.
Keep rules versioned and credentials external
Place project rules in the documented workspace location and version them with the repository. Resolve the effective configuration before work so global, saved, and command-specific values do not silently change the task.
Reference secrets through the supported secret mechanism. Test the agent on a small representative change and verify the resulting diff and checks independently.
Task: [one measurable outcome]
Scope: [workspace and file patterns]
Permissions: [models, tools, MCP, shell, and network]
Configuration: [reviewed config and rules]
Secrets: [approved external references only]
Verification: [diff, tests, and effective config]
Recovery: [disable tool/config and restore checkpoint]Expected evidence and verification
Expected evidence includes the effective config, rule files, enabled tools, chosen model, reviewed diff, and test output. Verify that shared blocks and MCP servers stayed within their stated authority.
If behavior changes unexpectedly, return to the last reviewed config, disable newly introduced blocks or tools, restore the affected files, and repeat the same case.