[workspace-split] Phase 6: Extract core crates (engine, query, tools, lsp-service)

[yaohaowei0914]

Part of #68. Target branch: V1. Depends on #74.

Scope

With cycles broken, extract the four core crates.

• cc-engine ← src/engine/ + src/engine/agent/ (15+3 files, ~5,200 LOC)
• cc-query ← src/query/ (7 files, 2,273 LOC)
• cc-tools ← src/tools/ minus agent (44 files, ~16,800 LOC) — biggest single crate
• cc-lsp-service ← src/lsp_service/ (5 files, 2,419 LOC)

Special considerations

• cc-tools is the largest crate. If its incremental build is still slow post-extraction, sub-split into:
  • cc-tools-fs (FileRead/Write/Edit/Glob/Grep)
  • cc-tools-exec (Bash)
  • cc-tools-web (WebSearch/WebFetch)
  • cc-tools-misc (everything else)
• Defer the sub-split unless Phase 7 measurements show it's a bottleneck.

Acceptance

• Four new crates; all hub modules now live outside root
• Touching cc-tools/src/tools/file_read.rs rebuilds cc-tools + claude-code-rs only (check with -Zbuild-plan or cargo build -v)
• Incremental build time recorded

[yaohaowei0914]

Progress update (PR #86)

PR #86 lands the Phase 6 scaffold plus the first three real source moves. Roughly the first half of Phase 6.

Done in #86

• All four Phase 6 crates scaffolded and wired into the workspace: cc-engine, cc-query, cc-tools, cc-lsp-service. Each has a real Cargo.toml with its proper dep envelope.
• Five cycle-breaking moves that unblock the remaining source moves:
  • cc-types::background_agents — breaks query → tools via types
  • cc-types::{agent_channel, agent_events, agent_types} — breaks engine → ipc via agent events; also breaks types::tool → ipc via ToolUseContext::bg_agent_tx
  • cc-types::teams::{TeamContext, TeammateInfo} — breaks types::app_state → teams::types
  • HookRunner::run_stop_hooks added to the trait in cc-types + QueryDeps::hook_runner() accessor — the query loop now drives all hook events via the trait instead of direct crate::tools::hooks::* calls
  • lsp_service/types.rs now owns HoverInfo / SourceLocation / SymbolInfo — breaks lsp_service → tools::lsp
• Source moves (real code relocations, not just type hoists):
  • src/ui/status_line/ (~1k LOC) → cc-engine/src/status_line/
  • src/types/{app_state, tool, config}.rs → cc-engine/src/types/
  • With this, the Tool trait and ToolUseContext have left the root bin crate, which is the architectural unblocker for every remaining Phase 6/7 crate.

Incremental build measurement

Baseline (design doc): 7.5s for a main.rs edit.

Touch point	Rebuilds	Time
main.rs	root bin	5.96s (~20% off baseline)
Root tool file	root bin	7.21s
cc-engine/src/types/tool.rs	cc-engine + root bin	6.76s
cc-types leaf	cc-permissions + cc-engine + cc-session + root	8.38s
No-op	—	0.5–1.5s

The target (<3s) arrives once src/tools/ physically moves into cc-tools — that's the biggest single-step win still ahead.

Remaining for #75

• Move src/engine/ lifecycle code into cc-engine (~5k LOC)
• Move src/query/ into cc-engine (or independent cc-query depending on cc-engine; ~2.3k LOC)
• Move src/tools/ minus the four cycle-causing files into cc-tools (~16.8k LOC — biggest single-step win for build speed)
• Move src/lsp_service/ + src/tools/lsp.rs into cc-lsp-service
• Move the three cycle-causing tool files to their natural homes:
  • tools/lsp.rs → cc-lsp-service
  • tools/send_message.rs + tools/team_spawn.rs → cc-teams (Phase 7)
  • tools/system_status.rs → cc-ipc (Phase 7)

Design doc updated: docs/superpowers/specs/2026-04-20-workspace-split-design.md.