Add structural dependency analysis artifacts

- graph.json: Canonical dependency graph (10 crates, 16 edges, 76 files)
- graph.summary.md: One-page overview with fan-in/fan-out rankings
- sccs.md: Strongly Connected Components analysis (no cycles)
- layers.observed.md: 5-layer architecture diagram
- hotspots.md: Coupling hotspots (g3-config, g3-cli)
- limitations.md: Extraction limitations and validity conditions

analysis/deps/limitations.md

@@ -0,0 +1,102 @@
+# Extraction Limitations
+
+## Scope of Analysis
+
+This structural analysis was performed using static extraction methods only.
+
+## What Was Observed
+
+| Source | Method | Confidence |
+|--------|--------|------------|
+| Crate dependencies | Cargo.toml `[dependencies]` parsing | High |
+| Module structure | `pub mod` declarations in lib.rs files | High |
+| File inventory | Filesystem traversal of `src/` directories | High |
+| Use statements | `grep` for `^use ` patterns | Medium |
+| External dependencies | Cargo.toml `[dependencies]` sections | High |
+
+## What Was NOT Observed
+
+### 1. Conditional Compilation
+
+`#[cfg(...)]` attributes were not parsed. Dependencies that only exist under certain feature flags or target platforms are included unconditionally.
+
+**Example:** `g3-computer-control` has platform-specific dependencies:
+- `core-graphics`, `cocoa`, `objc` (macOS only)
+- `x11` (Linux only)
+- `windows` (Windows only)
+
+These are all listed as dependencies but only one set compiles per platform.
+
+### 2. Test Dependencies
+
+`[dev-dependencies]` were excluded from the graph. Test-only coupling is not represented.
+
+### 3. Build Dependencies
+
+`[build-dependencies]` were excluded. Build script dependencies are not represented.
+
+### 4. Dynamic/Runtime Dependencies
+
+The following cannot be detected statically:
+- Process spawning (e.g., `g3-console` spawns `g3` processes)
+- Plugin loading
+- Configuration-driven provider selection
+- WebDriver browser automation targets
+
+### 5. Trait Implementation Coupling
+
+Trait implementations create implicit coupling not captured by `use` statements:
+- `UiWriter` implementations in `g3-cli`
+- `LLMProvider` implementations in `g3-providers`
+- `ComputerController` implementations in `g3-computer-control`
+
+### 6. Re-exports
+
+`pub use` re-exports create transitive dependencies not fully traced:
+- `g3-core` re-exports types from `g3-providers`
+- `g3-computer-control` re-exports WebDriver types
+
+### 7. Macro Expansion
+
+Macro-generated code (e.g., `#[derive(...)]`, `async_trait`) creates dependencies not visible in source.
+
+### 8. Workspace Inheritance
+
+`workspace = true` dependencies inherit versions from root `Cargo.toml`. The actual version constraints are not captured in crate-level analysis.
+
+## Inference Notes
+
+### Inferred: g3-console Isolation
+
+`g3-console` has no internal crate dependencies in its `Cargo.toml`. It interacts with `g3` via:
+- Process spawning (`std::process::Command`)
+- Log file parsing
+- Filesystem monitoring
+
+This runtime coupling is not represented in the dependency graph.
+
+### Inferred: g3 → g3-providers Direct Dependency
+
+The root `g3` crate depends directly on `g3-providers` despite `g3-cli` also depending on it. This may be for:
+- Type re-exports
+- Direct provider access in main.rs
+- Historical reasons
+
+The actual usage was not verified.
+
+## Recommendations for Future Analysis
+
+1. **cargo-depgraph**: Use `cargo depgraph` for authoritative Cargo-resolved dependencies
+2. **cargo-tree**: Use `cargo tree` for transitive dependency analysis
+3. **rust-analyzer**: Use LSP for precise type-level dependency tracking
+4. **cargo-udeps**: Identify unused dependencies
+5. **cargo-machete**: Detect potentially unused dependencies
+
+## Validity Conditions
+
+This analysis is valid as of the extraction date. It may be invalidated by:
+
+- Adding/removing crates from the workspace
+- Changing `[dependencies]` in any Cargo.toml
+- Restructuring module hierarchies
+- Adding conditional compilation features