Problem / Motivation

Several pieces of code are declared but never called outside tests, accumulating dead code with #[allow(dead_code)] annotations. These should either be wired up or deleted.

Items

1. Evaluation module unused APIs

File: src/evaluation/success.rs

• RuleBasedEvaluator::with_min_success_rate() (line 85) — never called
• RuleBasedEvaluator::with_max_failures() (line 92) — never called
• LlmEvaluator::new() (line 208) — never called

Decision needed: Are these planned features or dead weight? If planned, add a tracking TODO. If dead, delete them.

2. chunk_by_paragraphs() unused alternative

File: src/workspace/chunker.rs (line 120)

Alternative chunking strategy. Has tests (lines 259, 287) but zero callers in production code. Only chunk_document() is used.

Decision needed: Delete if the token-window chunker is the permanent choice. Keep if paragraph chunking is planned for a future indexing mode.

3. install_bundled_channel_from_artifacts() dead code

File: src/extensions/manager.rs (line 1173)

Marked "Used by upcoming hot-activation flow" but not called anywhere. If hot-activation is no longer planned, delete it.

4. Reasoning::safety field unused

File: src/llm/reasoning.rs (line 206)

Marked "Will be used for sanitizing tool outputs" but the safety field is never read. Tool output sanitization happens in the worker (src/agent/worker.rs), not in the Reasoning struct. If sanitization belongs in the worker, remove this field. If it should move to Reasoning, wire it.

Acceptance Criteria

• Each item is either deleted (if dead) or wired up (if needed), with #[allow(dead_code)] removed
• No new #[allow(dead_code)] annotations added
• All existing tests pass (delete the chunk_by_paragraphs tests if the function is deleted)
• Zero clippy warnings

Non-Goals

• Does NOT refactor the evaluation or chunking systems
• Does NOT change behavior of any live code paths