quarry copilot-instructions.md

Copilot Code Review Instructions

What to focus on

• Logic errors, off-by-one, null/None paths, unhandled edge cases
• Security: injection, credential exposure, path traversal, unsafe deserialization
• API contract violations: function returns wrong type, missing error case, changed signature without updating callers
• Concurrency issues: race conditions, shared mutable state, missing locks
• Resource leaks: unclosed files, sockets, database connections

What NOT to flag

• OO structure scores (method_ratio, encapsulation, module_size, etc.) — tools/oo_score.py enforces these via a ratchet mechanism. Do not duplicate its job.
• New files that don't meet absolute OO thresholds — this is expected. The ratchet applies on subsequent changes.
• Formatting — ruff format handles this. Do not comment on whitespace, line length, quote style, or trailing commas.
• Import order — ruff check --select I handles this.

Do NOT suggest

• Adding # noqa, # type: ignore, # pylint: disable, or @pytest.mark.xfail to suppress errors. The project policy is: fix the code or escalate to the operator. Suppressions require explicit operator approval.
• Lowering max-complexity, relaxing type checker strictness, or widening per-file-ignores.
• Replacing __new__ with __init__ — this project uses __new__ as the constructor by design.

Context

This project uses an OO ratchet: make check-oo compares OO quality scores against a committed baseline (.oo-baseline.json). Every commit must improve at least one metric on touched files and regress none. The baseline is updated via make update-oo and committed alongside the code change.