fix(core+cli+loom): core-0.10.1 / cli-3.49.2 / loom-0.7.1 — every project-wide walker skips nested checkouts (#434)

.straymark/07-ai-audit/agent-logs/AILOG-2026-09-18-001-core-nested-checkout-rule-434-follow-up.md

@@ -0,0 +1,85 @@
+---
+id: AILOG-2026-09-18-001
+title: core nested-checkout rule — every project-wide walker skips other checkouts (#434 follow-up)
+status: accepted
+created: 2026-09-18
+agent: claude-opus-5-1m
+confidence: high
+review_required: false
+risk_level: low
+eu_ai_act_risk: not_applicable
+nist_genai_risks: [information_integrity]
+iso_42001_clause: []
+files_modified:
+  - core/src/walk.rs
+  - core/src/lib.rs
+  - core/src/architecture/gather.rs
+  - cli/src/tree_grep.rs
+  - cli/src/analysis_engine.rs
+  - experiment-baton/src/scan.rs
+  - core/Cargo.toml, cli/Cargo.toml, experiment-loom/Cargo.toml, experiment-baton/Cargo.toml, Cargo.lock
+  - CHANGELOG.md, experiment-loom/CHANGELOG.md, version tables x6
+observability_scope: none
+tags: [core, architecture, loom, baton, adopter-feedback]
+related:
+  - 07-ai-audit/agent-logs/AILOG-2026-09-17-002-baton-inventory-hygiene.md
+---
+
+# AILOG: core nested-checkout rule (#434 follow-up)
+
+## Summary
+
+#435 taught Baton's walkers to skip another checkout nested in the project. The same blind spot
+existed in `straymark-core`'s source scanner and in two CLI walkers. The rule now lives once, in
+`core::walk::is_nested_checkout`, and all of them use it. Released as core 0.10.1, CLI 3.49.2 and
+Loom 0.7.1.
+
+## Context
+
+Demonstrated on an adopter repository with one linked worktree under `.worktrees/`. `straymark
+architecture generate`, writing to a scratch `--out`, produced 15 components, one of them
+`.worktrees` (glob `.worktrees/**`): a full copy of the repo, 5606 source files against 5235 in the
+real tree. With this change the same command produces 14 components and no `.worktrees`. The
+repository was read-only in both runs (git status unchanged).
+
+## Actions Performed
+
+1. `core/src/walk.rs`: `pub fn is_nested_checkout(dir)`. A sub-directory with its own `.git` (a
+   file for a linked worktree or submodule, a directory for a nested clone) is another checkout.
+2. `core::architecture::collect_source_files_with` skips such directories. This covers
+   `architecture generate | sync | validate`, `status --where`, Loom's architecture view and
+   Intent plane, and Baton's overlay.
+3. CLI: `tree_grep::walk` (used by `followups verify --claims` and `analyze declared-vs-wired`) and
+   `analysis_engine::walk_recursive` (`analyze`) use the same rule.
+4. Baton: `scan::is_nested_checkout` is now a re-export of the `core` rule. Its behavior is
+   unchanged.
+5. Tests: `core::walk` itself, and one per walker. Each walker test uses a directory name that is
+   not already excluded (the first `analyze` fixture used `vendor/`, which `analyze` already
+   excludes, so it passed without the fix; it was moved to `third_party/`). All three walker tests
+   fail with the guard removed.
+6. Versions move atomically: core 0.10.0 → 0.10.1, and the `straymark-core` bound moves in the
+   CLI, Loom **and** Baton Cargo.toml, per the lesson from #282. CLI 3.49.2, Loom 0.7.1, CHANGELOG
+   and Loom CHANGELOG.
+
+## Decisions Made
+
+- **One rule, in `core`.** Walkers keep their own name lists, because those differ by purpose, but
+  the "another checkout" rule is shared so that no tool can disagree with another about what the
+  project contains.
+- **Patch versions.** The change adds an API to core and fixes behavior; no public API breaks.
+- Baton is not re-released: its behavior is identical, and the next Baton release will pick up the
+  new dependency bound.
+
+## Impact
+
+- **Functionality**: architecture models, `status --where`, Loom and `analyze` / `verify --claims`
+  no longer read another checkout as the project's code.
+- **Performance**: one `stat` per visited directory. Negligible.
+- **Security / Privacy / Environmental**: N/A.
+
+## Verification
+
+- [x] `cargo test --workspace`: 1041 passed
+- [x] Clippy: core and Baton clean with `-D warnings`; Loom clean; CLI unchanged from the baseline
+      (18/24)
+- [x] Before/after `architecture generate` on the adopter repository, read-only

CHANGELOG.md

@@ -7,6 +7,21 @@ and this project uses [independent versioning](README.md#versioning) for Framewo
 
 ---
 
+## CLI 3.49.2 / core 0.10.1 / Loom 0.7.1 — 2026-09-18
+
+Follow-up to #434, which fixed the Baton half. The shared source scanner in `straymark-core` and two CLI walkers also descended into **another checkout nested in the project**: a linked git worktree such as `.worktrees/<name>/`, a submodule or a nested clone. Each read that copy as the project's own code. On an adopter's repository with one worktree inside, `straymark architecture generate` invented a `.worktrees` component holding a full copy of the repo.
+
+### Fixed (CLI / core / Loom)
+
+- **`straymark-core` 0.10.1**: new `walk::is_nested_checkout`, the one rule every project-wide walker shares. A sub-directory with its own `.git` (file or directory) is another checkout and is never walked. It is structural, so no gitignore parsing and nothing to configure. `architecture::collect_source_files` uses it, which covers:
+  - `straymark architecture generate | sync | validate` and `status --where`;
+  - Loom's architecture view and Intent plane (**Loom 0.7.1**, rebuilt);
+  - Baton's overlay.
+- **CLI 3.49.2**: `tree_grep` (behind `followups verify --claims` and `analyze declared-vs-wired`) and the `analyze` complexity walker use the same rule. A worktree copy no longer counts as "callers" of a symbol, and its functions are no longer analyzed twice.
+- Baton now reuses the `core` rule instead of its own copy. Its behavior is unchanged.
+
+Until upgraded, the workaround is `architecture: { excluded_dirs: [".worktrees"] }` in `.straymark/config.yml`, which covers only the architecture scanner and only that directory name.
+
 ## Baton 0.3.0 — 2026-09-17
 
 First Baton release driven by an adopter's Track C work (Estoa, Discussion #426). Before it, the declared-verb router (#332) classified only units with their own slot: every SpecKit task and every batch was `undeclared`, and the inventory counted work that was not live. Experimental; distributed as `baton-*` GitHub releases (`--latest=false`).

README.md

@@ -279,7 +279,7 @@ StrayMark uses independent version tags for each component:
 | Component | Tag prefix | Example | Includes |
 | --- | --- | --- | --- |
 | Framework | `fw-` | `fw-4.45.0` | Templates (12 types), governance, directives, Charter template + schema |
-| CLI | `cli-` | `cli-3.49.1` | The `straymark` binary |
+| CLI | `cli-` | `cli-3.49.2` | The `straymark` binary |
 | Loom (EXPERIMENTAL) | `loom-` | `loom-0.4.2` | The `straymark-loom` visualization server, downloaded on demand by `straymark loom serve` |
 
 Check installed versions with `straymark status` or `straymark about`.

cli/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "straymark-cli"
-version = "3.49.1"
+version = "3.49.2"
 edition = "2021"
 description = "CLI for StrayMark — the cognitive discipline your AI-assisted projects need"
 license = "MIT"
@@ -25,7 +25,7 @@ path = "src/bin/gen_minimal_skills.rs"
 required-features = ["dev-tools"]
 
 [dependencies]
-straymark-core = { version = "0.10.0", path = "../core" }
+straymark-core = { version = "0.10.1", path = "../core" }
 clap = { version = "4", features = ["derive"] }
 reqwest = { version = "0.12", features = ["blocking", "rustls-tls", "json"] }
 serde = { version = "1", features = ["derive"] }

cli/src/analysis_engine.rs

@@ -73,7 +73,9 @@ fn walk_recursive(dir: &Path, files: &mut Vec<PathBuf>) {
         let path = entry.path();
         if path.is_dir() {
             if let Some(name) = path.file_name().and_then(|n| n.to_str()) {
-                if !EXCLUDED_DIRS.contains(&name) {
+                if !EXCLUDED_DIRS.contains(&name)
+                    && !straymark_core::walk::is_nested_checkout(&path)
+                {
                     walk_recursive(&path, files);
                 }
             }
@@ -206,6 +208,20 @@ mod tests {
         assert!(files[0].ends_with("main.rs"));
     }
 
+    #[test]
+    fn test_walk_skips_nested_checkouts() {
+        // #434: functions in a worktree copy are not this project's functions.
+        let dir = TempDir::new().unwrap();
+        fs::write(dir.path().join("main.rs"), "fn included() {}").unwrap();
+        let clone = dir.path().join("third_party/clone");
+        fs::create_dir_all(clone.join(".git")).unwrap();
+        fs::write(clone.join("lib.rs"), "fn duplicated() {}").unwrap();
+
+        let files = walk_source_files(dir.path());
+        assert_eq!(files.len(), 1);
+        assert!(files[0].ends_with("main.rs"));
+    }
+
     #[test]
     fn test_single_rust_file() {
         let dir = TempDir::new().unwrap();

cli/src/tree_grep.rs

@@ -80,7 +80,9 @@ fn walk(base: &Path, dir: &Path, out: &mut Vec<TextFile>) {
         let path = entry.path();
         if file_type.is_dir() {
             let name = entry.file_name();
-            if EXCLUDED_DIRS.contains(&name.to_string_lossy().as_ref()) {
+            if EXCLUDED_DIRS.contains(&name.to_string_lossy().as_ref())
+                || straymark_core::walk::is_nested_checkout(&path)
+            {
                 continue;
             }
             walk(base, &path, out);
@@ -156,6 +158,20 @@ mod tests {
         assert_eq!(tree[0].rel_path, "main.rs");
     }
 
+    #[test]
+    fn read_text_tree_skips_nested_checkouts() {
+        // #434: a worktree copy would count its symbols as the project's own.
+        let tmp = tempfile::TempDir::new().unwrap();
+        std::fs::write(tmp.path().join("main.rs"), "fn main() {}\n").unwrap();
+        let worktree = tmp.path().join(".worktrees/feature");
+        std::fs::create_dir_all(&worktree).unwrap();
+        std::fs::write(worktree.join(".git"), "gitdir: /elsewhere\n").unwrap();
+        std::fs::write(worktree.join("main.rs"), "fn main() {}\n").unwrap();
+        let tree = read_text_tree(tmp.path());
+        assert_eq!(tree.len(), 1);
+        assert_eq!(tree[0].rel_path, "main.rs");
+    }
+
     #[test]
     fn symbol_occurrences_respects_word_boundaries() {
         let tmp = tempfile::TempDir::new().unwrap();

core/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "straymark-core"
-version = "0.10.0"
+version = "0.10.1"
 edition = "2021"
 description = "Shared document model and knowledge graph for StrayMark — parses governance documents and builds their typed traceability graph"
 license = "MIT"

core/src/architecture/gather.rs

@@ -77,7 +77,8 @@ pub fn build_governance_state(root: &Path) -> GovernanceState {
 
 /// Source files (paths relative to `root`), filtered by the project's
 /// [`ScanConfig`] (source extensions + excluded dirs, defaults ∪ the
-/// `architecture:` config section, #279). Used for the on-disk inventory (the
+/// `architecture:` config section, #279). Never descends into another
+/// checkout nested in the project ([`crate::walk::is_nested_checkout`], #434). Used for the on-disk inventory (the
 /// `uncharted` signal) and by the CLI generator to discover component dirs.
 pub fn collect_source_files(root: &Path) -> Vec<PathBuf> {
     collect_source_files_with(root, &resolve_scan_config(root))
@@ -94,7 +95,7 @@ pub fn collect_source_files_with(root: &Path, cfg: &ScanConfig) -> Vec<PathBuf>
             let path = entry.path();
             if path.is_dir() {
                 if let Some(name) = path.file_name().and_then(|n| n.to_str()) {
-                    if !cfg.is_excluded_dir(name) {
+                    if !cfg.is_excluded_dir(name) && !crate::walk::is_nested_checkout(&path) {
                         walk(&path, root, cfg, out);
                     }
                 }
@@ -309,6 +310,26 @@ mod tests {
         assert_eq!(files, vec!["src/main.rs"]); // README.md (not source), target/, node_modules/ excluded
     }
 
+    #[test]
+    fn collect_source_files_skips_nested_checkouts() {
+        // #434: a linked worktree (`.git` file) and a nested clone (`.git`
+        // dir) are other checkouts, not this project's source.
+        let tmp = tempfile::TempDir::new().unwrap();
+        let root = tmp.path();
+        for d in ["src", ".worktrees/feature/src", "vendor/clone/src"] {
+            std::fs::create_dir_all(root.join(d)).unwrap();
+            std::fs::write(root.join(d).join("main.rs"), "fn main() {}\n").unwrap();
+        }
+        std::fs::write(root.join(".worktrees/feature/.git"), "gitdir: /elsewhere\n").unwrap();
+        std::fs::create_dir(root.join("vendor/clone/.git")).unwrap();
+
+        let files: Vec<String> = collect_source_files(root)
+            .iter()
+            .map(|p| p.to_string_lossy().replace('\\', "/"))
+            .collect();
+        assert_eq!(files, vec!["src/main.rs"]);
+    }
+
     #[test]
     fn build_state_empty_on_bare_dir() {
         // No .straymark, no git, no charters → every set empty, no panic.

core/src/lib.rs

@@ -15,3 +15,4 @@ pub mod drift;
 pub mod entities;
 pub mod graph;
 pub mod utils;
+pub mod walk;

core/src/walk.rs

@@ -0,0 +1,38 @@
+//! The one rule every project-wide directory walker shares.
+//!
+//! Walkers skip a fixed list of names (`target`, `node_modules`, …), but a
+//! project can also hold *another checkout of itself*: a linked git worktree
+//! under `.worktrees/`, a submodule, a nested clone. Reading it duplicates the
+//! project — `architecture generate` invented a `.worktrees` component holding
+//! a full copy of an adopter's repo (#434). The marker is structural, so no
+//! gitignore parsing and nothing for an adopter to configure.
+
+use std::path::Path;
+
+/// True when `dir` is the root of another checkout: a linked worktree or a
+/// submodule (a `.git` *file*) or a nested clone (a `.git` directory). Call it
+/// on sub-directories only — the project root itself has a `.git`.
+pub fn is_nested_checkout(dir: &Path) -> bool {
+    dir.join(".git").exists()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn git_file_or_dir_marks_a_nested_checkout() {
+        let tmp = tempfile::tempdir().unwrap();
+        let worktree = tmp.path().join(".worktrees/feature");
+        let clone = tmp.path().join("vendor/clone");
+        let plain = tmp.path().join("src");
+        for d in [&worktree, &clone, &plain] {
+            std::fs::create_dir_all(d).unwrap();
+        }
+        std::fs::write(worktree.join(".git"), "gitdir: /elsewhere\n").unwrap();
+        std::fs::create_dir(clone.join(".git")).unwrap();
+        assert!(is_nested_checkout(&worktree));
+        assert!(is_nested_checkout(&clone));
+        assert!(!is_nested_checkout(&plain));
+    }
+}

docs/adopters/CLI-REFERENCE.md

@@ -46,7 +46,7 @@ StrayMark uses **independent version tags** for each component:
 | Component | Tag prefix | Example | What it includes |
 |-----------|-----------|---------|------------------|
 | Framework | `fw-` | `fw-4.45.0` | Templates (12 types), governance docs, directives, Charter template + schema |
-| CLI | `cli-` | `cli-3.49.1` | The `straymark` binary |
+| CLI | `cli-` | `cli-3.49.2` | The `straymark` binary |
 | Loom (EXPERIMENTAL) | `loom-` | `loom-0.4.2` | The `straymark-loom` visualization server, downloaded on demand by `straymark loom serve` |
 
 Framework and CLI are released independently. A framework update does not require a CLI update, and vice versa.

docs/i18n/es/README.md

@@ -242,7 +242,7 @@ StrayMark usa tags de versión independientes para cada componente:
 | Componente | Prefijo de tag | Ejemplo | Incluye |
 |------------|---------------|---------|---------|
 | Framework | `fw-` | `fw-4.45.0` | Plantillas (12 tipos), gobernanza, directivas, plantilla + schema de Charter |
-| CLI | `cli-` | `cli-3.49.1` | El binario `straymark` |
+| CLI | `cli-` | `cli-3.49.2` | El binario `straymark` |
 | Loom (EXPERIMENTAL) | `loom-` | `loom-0.4.2` | El servidor de visualización `straymark-loom`, descargado bajo demanda por `straymark loom serve` |
 
 Verifica las versiones instaladas con `straymark status` o `straymark about`.

docs/i18n/es/adopters/CLI-REFERENCE.md

@@ -46,7 +46,7 @@ StrayMark usa **tags de versión independientes** para cada componente:
 | Componente | Prefijo de tag | Ejemplo | Qué incluye |
 |------------|---------------|---------|-------------|
 | Framework | `fw-` | `fw-4.45.0` | Plantillas (12 tipos), docs de gobernanza, directivas |
-| CLI | `cli-` | `cli-3.49.1` | El binario `straymark` |
+| CLI | `cli-` | `cli-3.49.2` | El binario `straymark` |
 | Loom (EXPERIMENTAL) | `loom-` | `loom-0.4.2` | El servidor de visualización `straymark-loom`, descargado bajo demanda por `straymark loom serve` |
 
 Framework y CLI se publican de forma independiente. Una actualización del framework no requiere actualización del CLI, y viceversa.

docs/i18n/zh-CN/README.md

@@ -260,7 +260,7 @@ StrayMark 为每个组件使用独立的版本标签：
 | 组件 | 标签前缀 | 示例 | 包含内容 |
 |------|----------|------|----------|
 | Framework | `fw-` | `fw-4.45.0` | 模板（12 种类型）、治理文档、指令、Charter 模板 + schema |
-| CLI | `cli-` | `cli-3.49.1` | `straymark` 二进制文件 |
+| CLI | `cli-` | `cli-3.49.2` | `straymark` 二进制文件 |
 | Loom（实验性） | `loom-` | `loom-0.4.2` | `straymark-loom` 可视化服务器，由 `straymark loom serve` 按需下载 |
 
 使用 `straymark status` 或 `straymark about` 查看已安装的版本。

docs/i18n/zh-CN/adopters/CLI-REFERENCE.md

@@ -46,7 +46,7 @@ StrayMark 为每个组件使用**独立的版本标签**：
 | 组件 | 标签前缀 | 示例 | 包含内容 |
 |------|----------|------|----------|
 | Framework | `fw-` | `fw-4.45.0` | 模板（12 种类型）、治理文档、指令 |
-| CLI | `cli-` | `cli-3.49.1` | `straymark` 二进制文件 |
+| CLI | `cli-` | `cli-3.49.2` | `straymark` 二进制文件 |
 | Loom（实验性） | `loom-` | `loom-0.4.2` | `straymark-loom` 可视化服务器，由 `straymark loom serve` 按需下载 |
 
 Framework 和 CLI 独立发布。Framework 更新不需要 CLI 更新，反之亦然。

experiment-baton/Cargo.toml

@@ -18,7 +18,7 @@ name = "straymark-baton"
 path = "src/main.rs"
 
 [dependencies]
-straymark-core = { version = "0.10.0", path = "../core" }
+straymark-core = { version = "0.10.1", path = "../core" }
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 serde_yaml = "0.9"

experiment-baton/src/scan.rs

@@ -5,16 +5,10 @@
 //! Also the one rule every directory walker shares: [`is_nested_checkout`].
 
 use std::collections::HashSet;
-use std::path::Path;
 
-/// True when `dir` is the root of *another* checkout — a linked git worktree
-/// or a submodule (a `.git` file) or a nested clone (a `.git` directory).
-/// Walkers never descend into one: its artifacts belong to another working
-/// copy, and reading them duplicates units and contracts (#434). Structural on
-/// purpose — no gitignore parsing, nothing for an adopter to configure.
-pub(crate) fn is_nested_checkout(dir: &Path) -> bool {
-    dir.join(".git").exists()
-}
+/// Never descend into another checkout nested in the project (#434) — the
+/// rule lives in `core` so Baton, the CLI and Loom's scanner agree on it.
+pub(crate) use straymark_core::walk::is_nested_checkout;
 
 /// Scan `text` for all identifiers shaped `<prefix><body>` where `body` is made
 /// of ASCII alphanumerics / hyphens and contains at least one digit. Returns