refactor(cache): move the task fs story into crates/vt_fs_fingerprint

The task_fs module already mirrors a standalone crate: lift it out of the
engine wholesale so the unsettled read-write-overlap policy has a home that is
tested with synthetic traces instead of whole-engine runs. The engine keeps
storage, archiving, env tracking, and spawning; rayon, twox-hash, and wax
leave its dependency list with the moved code.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: wan9chi

CLAUDE.md

@@ -9,6 +9,7 @@ A monorepo task runner (like Nx/Turbo) with intelligent caching and dependency r
 - `crates/vt_graph` — Task dependency graph construction and config loading
 - `crates/vt_plan` — Execution planning (resolves env vars, working dirs, commands)
 - `crates/vt_workspace` — Workspace detection and package dependency graph
+- `crates/vt_fs_fingerprint` — Filesystem fingerprinting for task caching (pre-run snapshot, traced-access judgment, cache-entry validation)
 - `crates/fspy*` — File system access tracing (9 crates: supervisor, preload libs, platform backends)
 - `crates/pty_terminal*` — Cross-platform headless terminal emulator (3 crates)
 - `crates/vt_path` — Type-safe absolute/relative path system

Cargo.lock

@@ -149,6 +149,7 @@ tui-term = "0.3.1"
 twox-hash = "2.1.1"
 uuid = "1.18.1"
 vec1 = "1.12.1"
+vt_fs_fingerprint = { path = "crates/vt_fs_fingerprint" }
 vt_glob = { path = "crates/vt_glob" }
 vt_graph_ser = { path = "crates/vt_graph_ser" }
 vt_path = { path = "crates/vt_path" }

Cargo.toml

@@ -25,7 +25,6 @@ once_cell = { workspace = true }
 owo-colors = { workspace = true }
 petgraph = { workspace = true }
 pty_terminal_test_client = { workspace = true }
-rayon = { workspace = true }
 rusqlite = { workspace = true, features = ["bundled"] }
 rustc-hash = { workspace = true }
 serde = { workspace = true, features = ["derive", "rc"] }
@@ -43,9 +42,9 @@ tokio = { workspace = true, features = [
 ] }
 tokio-util = { workspace = true }
 tracing = { workspace = true }
-twox-hash = { workspace = true }
 materialized_artifact = { workspace = true }
 uuid = { workspace = true, features = ["v4"] }
+vt_fs_fingerprint = { workspace = true }
 vt_glob = { workspace = true }
 vt_path = { workspace = true }
 vt_select = { workspace = true }
@@ -55,7 +54,6 @@ vt_ipc_shared = { workspace = true }
 vt_plan = { workspace = true }
 vt_server = { workspace = true }
 vt_workspace = { workspace = true }
-wax = { workspace = true }
 zstd = { workspace = true }
 
 # Artifact build-deps must be unconditional: cargo's resolver panics when

crates/vt/Cargo.toml

@@ -567,18 +567,19 @@ Each `&&` separated command is cached independently. If only terser config chang
 ### Core Cache Components
 
 ```
+crates/vt_fs_fingerprint/src/  # The run's filesystem story (own crate)
+├── task_run.rs           # TaskFs (pre_run/post_run), Conclusion
+├── fingerprint.rs        # InputFingerprints, PathFingerprint, InputChange
+├── tracked_accesses.rs   # fspy access normalization
+├── glob.rs               # Glob walking + input hashing
+└── hash.rs               # Content hashing
+
 crates/vt/src/session/
 ├── cache/
 │   ├── mod.rs            # ExecutionCache, CacheEntryKey/Value, FingerprintMismatch
 │   └── display.rs        # Cache status display formatting
 ├── execute/
 │   ├── mod.rs            # execute_spawn, SpawnOutcome
-│   ├── task_fs/          # The run's filesystem story
-│   │   ├── task_run.rs   # TaskFs (pre_run/post_run), Conclusion
-│   │   ├── fingerprint.rs # InputFingerprints, PathFingerprint, InputChange
-│   │   ├── tracked_accesses.rs # fspy access normalization
-│   │   ├── glob.rs       # Glob walking + input hashing
-│   │   └── hash.rs       # Content hashing
 │   ├── cache_update.rs   # Post-run cache update decision
 │   ├── post_run.rs       # TrackedEnvFingerprints (tracked env validation)
 │   └── spawn.rs          # spawn_with_tracking, fspy integration

crates/vt/docs/task-cache.md

@@ -14,6 +14,8 @@ pub use display::{
 use rusqlite::{Connection, OptionalExtension as _};
 use serde::{Deserialize, Serialize};
 use tokio::sync::Mutex;
+pub use vt_fs_fingerprint::InputChangeKind;
+use vt_fs_fingerprint::{InputChange, InputFingerprints};
 use vt_graph::config::ResolvedGlobConfig;
 use vt_path::{AbsolutePath, RelativePathBuf};
 use vt_plan::cache_metadata::{CacheMetadata, ExecutionCacheKey, SpawnFingerprint};
@@ -25,11 +27,9 @@ use wincode::{
     io::{Reader, Writer},
 };
 
-pub use super::execute::task_fs::InputChangeKind;
 use super::execute::{
     pipe::StdOutput,
     post_run::{PostRunMismatch, TrackedEnvFingerprints, TrackedEnvQuery},
-    task_fs::{InputChange, InputFingerprints},
 };
 
 const TASK_CACHE_PREALLOCATION_SIZE_LIMIT: usize = 256 * 1024 * 1024;

crates/vt/src/session/cache/mod.rs

@@ -4,6 +4,7 @@
 use std::{collections::BTreeMap, sync::Arc, time::Duration};
 
 use rustc_hash::FxHashSet;
+use vt_fs_fingerprint::{Conclusion, PostRunError};
 use vt_path::{AbsolutePath, RelativePathBuf};
 use vt_plan::cache_metadata::{CacheMetadata, EnvValueHash};
 use vt_server::Reports;
@@ -13,7 +14,6 @@ use super::{
     CacheState,
     post_run::{TrackedEnvFingerprints, TrackedEnvQuery},
     spawn::ChildOutcome,
-    task_fs::{Conclusion, PostRunError},
 };
 use crate::session::{
     cache::{CacheEntryValue, ExecutionCache, archive},

crates/vt/src/session/execute/cache_update.rs

@@ -3,7 +3,6 @@ pub mod pipe;
 pub mod post_run;
 mod scheduler;
 pub mod spawn;
-pub mod task_fs;
 #[cfg(windows)]
 mod win_job;
 
@@ -15,6 +14,7 @@ use std::{
 
 use futures_util::future::LocalBoxFuture;
 use tokio_util::sync::CancellationToken;
+use vt_fs_fingerprint::TaskFs;
 use vt_ipc_shared::NODE_CLIENT_PATH_ENV_NAME;
 use vt_path::AbsolutePath;
 use vt_plan::{SpawnExecution, cache_metadata::CacheMetadata};
@@ -23,7 +23,6 @@ use vt_server::{Recorder, Reports, ServerHandle, StopAccepting, serve};
 use self::{
     pipe::{PipeSinks, StdOutput, pipe_stdio},
     spawn::{ChildHandle, ChildOutcome, SpawnStdio, spawn},
-    task_fs::TaskFs,
 };
 use super::{
     cache::{CacheEntryValue, CacheMiss, ExecutionCache, archive},

crates/vt/src/session/execute/mod.rs

@@ -1,7 +1,7 @@
 //! Post-run environment fingerprinting: env values and bulk env queries
 //! observed by runner-aware tools during execution, validated again at cache
 //! lookup. The filesystem half of post-run fingerprinting lives in
-//! [`super::task_fs`].
+//! [`vt_fs_fingerprint`].
 
 use std::{collections::BTreeMap, ffi::OsStr, sync::Arc};
 

crates/vt/src/session/execute/post_run.rs

@@ -2,7 +2,7 @@
 //!
 //! [`spawn`] does one thing: hand back the child's stdio pipes plus a
 //! cancellation-aware `wait` future. Draining the pipes is [`super::pipe`]'s
-//! job; the raw path accesses are judged by [`super::task_fs`] during the
+//! job; the raw path accesses are judged by [`vt_fs_fingerprint`] during the
 //! cache update.
 
 use std::{ffi::OsStr, io, process::Stdio};