crow

General Behaviour Guidelines

• When asked to explain a choice, don't apologize. Stick to facts; the goal is to learn and improve, not to assign blame.
• Before invoking task, create_session, a custom agent, or any skill that may spawn an agent:

1. Determine the current model's canonical ID from the session log and the associated pricing tier in the list below. If you cannot determine the current model, ask the user for clarification.
2. Explicitly select a delegated model from the same or a lower pricing tier. if there is any doubt about the calling model or pricing tier, let the user know why there's doubt and default to the model named as "default" in the list for the delegated model.
3. Use canonical model IDs only; do not infer equivalence from display names. Never rely on an agent or skill default model.
4. This applies equally to built-in agents, custom agents, and skill-invoked agents.
5. If the requested model is unavailable or the runtime reports a different model, stop the delegated work and select a different model. Do not continue on a higher-cost model. Model cost limits are mandatory, not optimization guidance. If a requested specialist cannot run with an allowed model, stop and ask the user rather than selecting a more expensive model. { "default": "gpt-6-luna", "0 - low": [ "claude-haiku-4.5", "claude-haiku-5.5", "gemini-3.8-flash", "gpt-5.4-nano", "gpt-5.6-luna", "gpt-6-luna", "mai-code-1.1-flash", ], "1 - medium": [ "claude-sonnet-4.6", "claude-sonnet-5", "claude-sonnet-5.5", "gemini-4-argon", "gpt-5.3-codex", "gpt-5.6-terra", "gpt-6-sol", "gpt-6.1-sol", "grok-4.5", "grok-4.6", "grok-4.7", "kimi-k3" ], "2 - high": [ "claude-opus-4.8", "claude-opus-5", "claude-opus-5.5", "gpt-5.6-sol" ], "3 - premium": [ "gpt-6-astra", "claude-fable-5", "claude-fable-5.1" ] }

Automated Tool & Skill Enforcement

• codebase-memory-mcp — Provides the code intelligence graph used for indexing, architecture discovery, symbol search, and cross-file tracing. If it is not available, warn the user that analysis coverage may be reduced. Install it globally in the VS Code user profile.
• If the codebase-memory-mcp server is available, use it and update the index as needed. Its scans are usually faster and cheaper than command-line file searches.
• When performing code analysis or SonarQube scans, ALWAYS invoke the sonar-scan skill tool BEFORE calling any underlying Sonar MCP tools.
• If Sonar or ADO MCP tools fail with an Unauthorized or 403 response, stop and check in with the user. They may need to connect to the VPN to reach those tools.

Development Guidelines

• Once you've completed a code change, run the linter for the language you're working in and fix any issues before running any other tests.
• After verifying your changes via linting and automated tests, run a sonar scan and fix any issues reported in your new code before committing your changes.