AI agent pipeline for continuous test backfill

[jiaenren]

Summary

Build an automated pipeline where AI coding agents continuously identify test coverage gaps, generate tests, and open PRs — closing the coverage gap without dedicated SWE time.

Motivation

• Coverage reporting (Add test coverage reporting to CI #720) makes gaps visible, but someone still has to write the tests
• Manually backfilling tests for ~23 untested CLI subcommands and 5 zero-coverage services is tedious
• AI agents (Claude Code, Cursor) can generate pure-logic unit tests rapidly, but need human review for correctness
• This pipeline turns coverage data into a continuous improvement loop

Acceptance Criteria

• Pipeline can identify top N uncovered functions from coverage report
• AI-generated tests follow existing test patterns (fixtures from tests/common/)
• Generated tests pass bazel test before PR is opened
• PRs include coverage diff showing improvement
• Human review required before merge

Dependencies

• Add test coverage reporting to CI #720 — Coverage reporting must be in place first

Parent Issue

Part of #719 — Improve test coverage and release confidence