feat: add Kilo agent (kilo CLI)

[mariusvniekerk]

Summary

• Implements the Agent interface for Kilo CLI (kilo run)
• Prompt piped via stdin, --model for model selection, --auto for agentic mode, --variant for reasoning levels
• Reuses filterOpencodeToolCallLines for output cleanup (kilo shares opencode's output format)
• Implementation intentionally cloned from opencode.go rather than abstracted, since kilo may diverge in the future
• Adds kilo to auto-discovery fallback order and help/error strings

Test plan

• go test ./internal/agent/ -run TestKilo passes (13 tests)
• go test ./... green (2973 tests)
• roborev review --local --agent kilo --wait HEAD produces a review

🤖 Generated with Claude Code

[roborev-ci]

roborev: Combined Review (40b3bf4f)

Verdict: One medium-severity test reliability issue; no security vulnerabilities identified.

Medium

1. Parallel test mutates shared/global test state (flaky/racy)
   • Locations: internal/agent/kilo_test.go:111, internal/agent/kilo_test.go:113, internal/agent/kilo_test.go:116
   • Finding: TestKiloAgenticAutoFlag is marked t.Parallel() while using withUnsafeAgents(...), which mutates shared/global agent safety state. This can cause nondeterministic failures (and potentially env/state conflicts) when tests run concurrently.
   • Suggested fix: Remove t.Parallel() from tests that call withUnsafeAgents(...), or refactor test setup to avoid global mutation in parallel execution.

---

Synthesized from 4 reviews (agents: codex, gemini | types: security, default)