[workspace-split] Epic: monolithic crate → Cargo workspace

[yaohaowei0914]

Workspace Split — Epic

Integration branch: V1
Design doc: docs/superpowers/specs/2026-04-20-workspace-split-design.md

Why

Today the Rust source is one crate (~92k LOC, 356 files). Any edit triggers full-crate re-codegen. After switching to rust-lld (commit 620e7ad), incremental cargo build is ~7.5s — still dominated by per-crate codegen.

Target: after splitting into a workspace, editing a single tool touches ~3s.

Dependency analysis (source: grep use crate::X;)

• Leaves (zero internal deps): types, auth, observability, skills, keybindings, bootstrap
• Cycles (must be broken): tools::agent↔engine::lifecycle, engine↔commands, via narrow call sites (1–3 files each)

Phased execution

Each child issue is an independently mergeable PR on V1. Build and test must pass at every phase.

• [workspace-split] Phase 0: Workspace skeleton (no code moves) #69 — Phase 0: Workspace skeleton (no code moves)
• [workspace-split] Phase 1: Extract leaf crates (types, observability, keybindings) #70 — Phase 1: Extract leaf crates (types, observability, keybindings)
• [workspace-split] Phase 2: Extract leaf crates (auth, bootstrap, skills) #71 — Phase 2: Extract leaf crates (auth, bootstrap, skills)
• [workspace-split] Phase 3: Extract level-1 crates (config, utils, mcp, services, computer-use) #72 — Phase 3: Extract level-1 crates (config, utils, mcp, services, computer-use)
• [workspace-split] Phase 4: Extract level-2 crates (compact, sandbox, permissions, browser, session) #73 — Phase 4: Extract level-2 crates (compact, sandbox, permissions, browser, session)
• [workspace-split] Phase 5: Break hub cycles (tools↔engine, engine↔commands) #74 — Phase 5: Break hub cycles (agent→engine, Hook trait, commands inversion)
• [workspace-split] Phase 6: Extract core crates (engine, query, tools, lsp-service) #75 — Phase 6: Extract core crates (engine, query, tools, lsp-service)
• [workspace-split] Phase 7: Extract high-level crates (plugins, teams, daemon, commands, ipc) #76 — Phase 7: Extract high-level crates (plugins, teams, daemon, commands, ipc)
• [workspace-split] Phase 8: Thin root bin crate + final measurements #77 — Phase 8: Thin root bin crate + final measurements

Acceptance

• Incremental build after editing one tool: <3s (baseline 7.5s)
• Clean release build within 10% of current (~2m)
• cargo test green at every phase boundary
• No new public API, no new runtime deps