rsconstruct 0.9.142

Rust based fast build system
# TOFIX

Findings from a code scan on 2026-10-25, verified against the code on 2026-10-25.
Findings that did not hold up were removed; details that were inaccurate were corrected.
The High findings were fixed on 2026-10-26 and removed from this list.

## Medium

- `src/remote_cache.rs:116-123, 143-157` and `src/object_store/operations.rs:27, 100-102` - In `S3Backend::exists` and `S3Backend::download_bytes`, any failure of the `aws s3` subprocess (expired AWS credentials, `AccessDenied`, network failure, missing bucket) is treated as `Ok(false)` / `Ok(None)` instead of returning an error, silently degrading every build to full rebuilds with zero diagnostics (violating Rule 1 "Fail hard, never degrade gracefully" and contradicting `HttpBackend` at lines 191-200 and 250-255). The `exists` call sites in `operations.rs` additionally do `.unwrap_or(false)`, collapsing even `HttpBackend`'s errors to "absent". Additionally, `aws s3 ls s3://bucket/prefix/key` performs a prefix match rather than an exact key check (harmless today because keys are fixed-length checksums). Use `aws s3api head-object` (or check stderr for `404`/`NoSuchKey` vs transport/auth errors) and propagate non-404 errors.
- `src/tool_lock.rs:237-256` - In `collect_wanted`, `uv_owned` checks `deps.pip_source == crate::config::PipSource::UvLock && uv_lock.exists()`, but `npm_owned` checks only `if package_lock.exists()` without checking `deps.npm_source == crate::config::NpmSource::PackageLock`. When a project sets `npm_source = "package-json"` and a `package-lock.json` file also exists on disk, `rsconstruct tool lock` wrongly drops the packages named in that lockfile from `rsconstruct.lock` (they are still installed from `package.json`, but unpinned). Check `deps.npm_source == crate::config::NpmSource::PackageLock && package_lock.exists()` on line 247.
- `src/tools.rs:625-701` (`describe_binary`) vs `src/tools.rs:817-895` (`fetch_binary`) and `src/tools.rs:916-944` (`run_binary`) - `describe_binary` does not take `InstallCtx` / `binary_pins` (the ctx is built at `src/builder/tools.rs:779-783` right before the describe loop but never passed in), so the install plan that `tool install` always prints (and that the `-i` confirmation prompt shows) uses the unpinned `releases/latest` URL even when `rsconstruct.lock` pins a version. Additionally, for `ArchiveKind::Gunzip` (`taplo`, line 686), `describe_binary` prints `gunzip -f /tmp/taplo.dl` (which `gunzip` rejects because the filename lacks `.gz`), whereas `fetch_binary` renames `.dl` -> `.dl.gz`, runs `gunzip -f`, and renames `.dl` -> `taplo`; and for `ArchiveKind::TarGz` and `ArchiveKind::Zip` when `inner != dest` (`oxlint`), `describe_binary` omits the rename from `/tmp/<inner>` to `/tmp/<dest>`. Pass `binary_pins` into `describe_binary` and align its shell commands with `fetch_binary`.
- `src/builder/product.rs:31, 53-59` and `src/builder/analyzers.rs:450` - `Builder::product_show` calls `analyzer.scan_hash_pieces(ctx, product.primary_input())` on every active analyzer in the project without checking `analyzer.match_product(product)`; errors are swallowed by `if let Ok(Some(pieces))`, so with `[analyzer.tera]` enabled a Rust source containing `{{ glob(pattern="src/*.rs") }}` shows a bogus `hash_pieces: [tera] glob ...` section for a non-tera product, and a Tera syntax error in a `.rs` file yields `hash_pieces: (none)` — misleading output, not a failure. Also a leading `./` on user-supplied paths is not normalized in `src/builder/product.rs:31` and `src/builder/analyzers.rs:450` (unlike `normalize_query_paths` in `src/builder/graph.rs:386-394`): `product show ./x` reports "No product owns or consumes" and `analyzer show files ./x` reports "not in dependency cache" while the same path without `./` works. Guard the loop at `src/builder/product.rs:54` with `if !analyzer.match_product(product) { continue; }`, and normalize `./` in both places.
- `src/builder/tools.rs:410, 677, 723, 733, 1071, 1157-1165`, `src/tool_lock.rs:807-820`, `src/builder/doctor.rs:201, 228` - Several tool/dependency management bugs: (1) `src/builder/tools.rs:1071` probes Ruby gems with `gem list -i <pkg>` without `--exact` (unlike `src/tool_lock.rs:798`), so `gem list -i rack` succeeds when only `rack-test` is installed; (2) `src/tool_lock.rs:807-820` (`parse_gem_list`) does not strip platform suffixes inside parentheses (e.g. `ffi (1.17.4 x86_64-linux-gnu, 1.17.3)`), recording `"1.17.4 x86_64-linux-gnu"` as the version string, so the drift check at `tool_lock.rs:885-893` reports a mismatch against a `1.17.4` pin; (3) `src/builder/tools.rs:1157-1165` formats `install-deps` errors with `{e}` instead of `{e:#}`, hiding the underlying `.with_context(...)` cause; (4) `src/builder/tools.rs:410, 677, 723, 733` select `info.install_methods.first()` / `[0]` while the `--all` path at line 634 uses `.find(|m| m.method != "manual")` — currently no `ToolInfo` has a `"manual"` method, so this has no observable effect and the comment at lines 630-633 citing `clojure` is stale; (5) `src/builder/doctor.rs:201` checks Python packages via `pip show --quiet`, failing in `uv`-managed virtualenvs (the default installer) where `pip` is not installed (unlike `src/tool_lock.rs:747-766` which uses `python3 -c "import importlib.metadata..."`); (6) `src/builder/doctor.rs:228` checks `[dependencies].npm` via `npm list --depth=0 <pkg>` without `-g`, whereas `install-deps` (`src/builder/tools.rs:1018`) installs `[dependencies].npm` globally with `npm install -g` and `src/tool_lock.rs:770-773` queries `npm ls -g`.
- `src/config/mod.rs:957` (`Config::apply_overrides`) - Formats `let section_label = format!("processor.{}", inst.instance_name);` even though `inst.instance_name` already starts with `"processor."` (built at `src/config/processors.rs:463-487`), producing `[processor.processor.<type>.<name>]` in `--iset` / `--pset` override validation errors (reproduced: `build --dry-run --iset 'processor.checker.rumdl.max_jobs=0'` prints `[processor.processor.checker.rumdl]: 'max_jobs' must be greater than 0`). Use `&inst.instance_name` directly.
- `src/config/analyzer_configs.rs:205-289` and `src/analyzers/mod.rs:646-650` - `KnownFields::checksum_fields()` on all 6 analyzer config structs is dead code (never called anywhere; every caller goes through the processor-only `checksum_fields_for` / `checksum_fields_of`). `config_fingerprint` (`src/analyzers/mod.rs:646-650`) serializes the entire config struct with `toml::to_string(config)`, so changing non-output-affecting fields like `src_exclude_dirs` changes the fingerprint and invalidates all cached entries in `deps.redb` (`enabled` is in the fingerprint too, though toggling it off removes the analyzer anyway). Filter the serialized analyzer config by `KnownFields::checksum_fields()` in `config_fingerprint`.
- `src/executor/execution.rs:1177-1186, 1268-1283` and `src/executor/handlers.rs:37-51, 149-164` - When multiple products fail concurrently in the same parallel level without `--keep-going`, `record_failure` (called from `handle_error`) only prints to stderr when `keep_going` is set, otherwise pushing the error to `shared.errors` silently, and `collect_build_stats` returns only `errs.into_iter().next().unwrap()`, silently discarding all other concurrent product errors from that level (nothing else prints them; `main.rs:138` prints just the returned one). Also, `prepare_level_work` (`src/executor/execution.rs:1268-1283`) omits the `[processor] product:` context prefix on error when `!self.keep_going`. Combine all errors from the level (or print each before returning) and include the product prefix in `prepare_level_work`.
- `src/analyzers/tera.rs:170, 185, 199-248, 291-300, 341-345` - Two bugs in `TeraDepAnalyzer`: (1) `GREP_COUNT_RE` (`r"grep_count\s*\(([^)]*)\)"`, line 170) and `SHELL_OUTPUT_RE` (`r"shell_output\s*\(([^)]*)\)"`, line 185) stop at the first `)` even when that `)` is inside a quoted string argument (e.g. `shell_output(command="echo $(uname -r)", depends_on=["foo"])`), truncating `body` before `depends_on` or `glob` and causing a false missing-argument error; (2) `scan_dependencies` (lines 199, 209, 234, 248) resolves `include`, `load_*`, and `version_str` paths relative to `source_dir` first and then the project root, whereas `TeraProcessor` (`src/processor/generator/tera.rs:88-104, 634-641, 686-697, 807-813`) registers includes under their project-relative name and resolves loaders against the current directory only. Match arguments across quoted strings and resolve paths relative to the project root to match `TeraProcessor`.
- `src/processor/checker/aspell.rs:149-150, 199-200, 211` - `AspellProcessor::discover` pushes only `words_file` onto `dep_auto` and never `self.config.conf`. The plugin's default `dep_auto` (`[".aspell.conf", ".aspell.en.pws", ".aspell.en.prepl"]`, line 211) tracks the default conf path, but a non-default `conf = "other.conf"` is not tracked as a dependency (its path is in the config hash via `affects_output: true`, its content is not), unless the user also lists it in `dep_auto`. Add a non-empty `self.config.conf` to `dep_auto` in `discover`.
- `src/processor/checker/cpplint.rs:25`, `src/processor/checker/hadolint.rs:25`, `src/processor/checker/clang_tidy.rs:111`, `src/processor/checker/cmake.rs:25`, `src/processor/checker/htmllint.rs:25`, `src/processor/checker/markdownlint.rs:89`, `src/processor/checker/mdl.rs:153`, `src/processor/checker/pylint.rs:25`, `src/processor/checker/rumdl.rs:31`, `src/processor/checker/luacheck.rs:25` - External checkers missing default config files in `dep_auto` (or `output_depends_on_input_name`); the plugin's `ProcessorDefaults` is the only source of such defaults (`ToolInfo` carries none): `cpplint` (`CPPLINT.cfg`), `hadolint` (`.hadolint.yaml`, `.hadolint.yml`, matching `idockerfile.rs:223`), `clang_tidy` (`defaults: None`; `.clang-tidy`, `_clang-tidy`, `compile_commands.json`), `cmake` (`.cmakelintrc`), `htmllint` (`.htmllintrc`), `markdownlint` (missing `.markdownlint.yml`), `mdl` (missing `.mdl_style.rb`; `.mdlrc` is not parsed for its `style`), `pylint` (missing `pylintrc`, `pyproject.toml`, `setup.cfg`), `rumdl` (missing `rumdl.toml`, `pyproject.toml`), and `luacheck` (missing `output_depends_on_input_name: true`, unlike `iluacheck.rs:355`).
- `src/processor/generator/mod.rs:90-102`, `src/processor/mod.rs:433, 559-581, 652-655, 1355-1379`, and custom generators (`src/processor/generator/cc_single_file.rs:648`, `src/processor/generator/ijinja2.rs:345`, `src/processor/generator/ipdfunite.rs:253`, `src/processor/generator/iprotobuf.rs:194`, `src/processor/generator/jinja2.rs:107`, `src/processor/generator/mako.rs:115`, `src/processor/generator/pdfunite.rs:121`, `src/processor/generator/requirements.rs:127`, `src/processor/generator/rust_single_file.rs:99`, `src/processor/generator/tags.rs:148`, `src/processor/generator/tera.rs:175`, `src/processor/mass_generator/jekyll.rs:344`, `src/processor/mass_generator/linux_module.rs:318`) - `dep_auto` is a `StandardConfig` known field accepted (and existence-validated by `validate_dep_auto_exist`) for all these processors, but their `discover()` implementations only call `resolve_extra_inputs(&self.config.standard.dep_inputs)` and never reference `scan.dep_auto`. `discover_checker_products` merges it via `config_file_inputs` (`src/processor/mod.rs:433, 652-655`); `discover_directory_products` (`src/processor/mod.rs:559-581`) and `SimpleGenerator` (`DiscoverParams` has no `dep_auto` field, `src/processor/generator/mod.rs:90-102`) ignore it too, so every data-driven generator is affected as well. `src/processor/generator/generic.rs:133-140` handles it correctly. Merge `dep_auto` via `config_file_inputs` in each, or add `dep_auto` to `omit_standard_fields` where unsupported.
- `src/processor/generator/tags.rs:154-167, 384, 457-463, 1544-1550, 1575-1583` and `src/processor/checker/terms.rs:141-155` - Bugs in `tags` and `terms`: (1) `TagsProcessor::discover` (`tags.rs:154-167`) and `TermsProcessor::discover` (`terms.rs:141-155`) iterate `fs::read_dir` without sorting paths before pushing `.txt` files into `inputs` / `dep_inputs`, and nothing sorts them later (`insert_product` keeps order, `combined_input_checksum` is order-sensitive), making the cache key filesystem-dependent (spurious rebuilds if readdir order changes); (2) `TagsProcessor::check_tags` (`tags.rs:1544-1550`, used by `rsconstruct tag check`) unconditionally fails on unused tags even when `config.check_unused` is `false` (unlike `execute()` at line 384 which checks `if self.config.check_unused`), and ignores `serde_json::Value::Array` values in `unique_fields` at line 1575-1583 (unlike `execute()` at lines 457-463).
- `src/processor/generator/cc_single_file.rs:351-360, 409, 476-500, 561, 580-592, 741, 781-786` - Four issues in `cc_single_file`: (1) `include_scanner` field and `IncludeScanner` enum (lines 351-360, 409, 781) are dead config — never read anywhere in the codebase (no `-MM` code path exists); (2) `find_source_files` (line 498) and `execute` (line 741) only check `extension == Some("cc")` to select C++ mode; the default `src_extensions` is `[".c", ".cc"]` so this is moot by default, but a user who adds `.cpp`/`.cxx` to `src_extensions` gets those files compiled with the C compiler and `cflags`; (3) `source_dir()` / `should_process` (lines 476-490) only check whether `src_dirs.first()` exists and ignore `src_dirs[1..]`; (4) `compile_source` (lines 580-591) prints `[cc_single_file...] <command>` to stdout when `ctx.show_child_processes()` is set, and then `run_command` -> `log_command` (`src/processor/mod.rs:123-135, 159`) prints `[exec] <command>` a second time to stderr.
- `src/processor/checker/duplicate_files.rs:19-29, 70-94` - `DuplicateFilesProcessor::discover` appends `dep_inputs` and `dep_auto` to `product.inputs` with no dedup, and `execute()` hashes all paths in `product.inputs` to check for duplicate file contents without excluding `dep_inputs`/`dep_auto` or deduplicating identical paths. If a file listed in `dep_inputs` is also matched by `scan` (same path twice, same hash) or has the same content as a scanned file, `duplicate_files` falsely fails the build. Deduplicate `files` or exclude `dep_inputs`/`dep_auto` from the duplicate-content comparison.
- `src/processor/mass_generator/linux_module.rs:175-230, 301-303`, `src/processor/mass_generator/gem.rs:118, 173`, `src/processor/mass_generator/jekyll.rs:117, 146-149`, `src/processor/mass_generator/generic.rs:84, 148-151`, `src/processor/mass_generator/cargo.rs:180-192, 245-255, 324-330, 400`, `src/processor/mass_generator/mdbook.rs:163-201`, `src/processor/mass_generator/npm.rs:97-98` - Mass generator issues: (1) `linux_module.rs:301-303` hardcodes `vec!["make", "uname"]` in `required_tools()`, ignoring `manifest.make` (which `build_module` actually runs, line 175) and `standard.required_tools`, and `build_module` (lines 189-190 `?`) leaves the generated `Kbuild` file and kbuild intermediates in the source tree if `make modules` fails before the cleanup at lines 206-225; (2) `gem.rs:118` sets `siblings.extensions: &[".gemspec"]`, and `Gemfile.lock` is covered by no other field (`dep_auto` is in `omit_standard_fields`, line 173), so only user `dep_inputs` can track it — whether the lockfile should be an input is a design decision; (3) `jekyll.rs:117, 146-149` and `mass_generator/generic.rs:84, 148-151` cache their plan in a `Mutex` on Round 0 of fixed-point discovery and return it unchanged on later rounds even though both are `WholeIndex` and are re-run when upstream generators add virtual files to `file_index`; (4) `cargo.rs:180-192` uses `[workspace].members` verbatim (a glob like `members = ["crates/*"]` yields a nonexistent path and is skipped silently), and because discovery scans every `Cargo.toml` (line 400) with no workspace detection, a separately discovered member crate predicts `crates/foo/target/debug/foo` (lines 324-330) while `cargo` (run without `--target-dir`, lines 245-255) writes to `<workspace_root>/target/debug/foo`; (5) `mdbook.rs:163-201` (`plan_book`) does not include files from a custom `theme/` directory in `sources`; (6) `npm.rs:97` includes `".js"` and `".ts"` in `siblings.extensions`, and the sibling query is recursive under the package directory (excluding only `node_modules`, `.git`, `out`, `.rsconstruct`), so `npm install` re-runs whenever any `.js`/`.ts` under the package directory is edited — for a root `package.json`, the whole repo.
- `src/processor/generator/pdfunite.rs:72, 208-219`, `src/processor/generator/ipdfunite.rs:71, 328-339`, `src/processor/generator/pandoc.rs:64-70, 116`, `src/processor/generator/jinja2.rs:49`, `src/processor/generator/iprotobuf.rs:53-60, 82, 103, 194-196` - Generator bugs: (1) `pdfunite` and `ipdfunite` pass `&[], &[], &[], &[]` (`src_exclude_dirs`, `src_exclude_files`, `src_exclude_paths`, `src_files`) to `file_index.query` instead of the `StandardConfig` exclude/filter fields, and both plugins carry `scan_defaults` of `["course.yaml"]` and a `source_dir` doc "Directory containing course YAML files listing PDFs to merge" although the default `source_ext` is `.md` and no YAML is read; (2) `pandoc.rs:64-70` omits `pdflatex` from `required_tools()` when `pdf_engine` is empty and `"pdf"` is in `formats` (nothing inspects `formats`); (3) `jinja2.rs:49` renders templates with `**os.environ`, making output depend on untracked environment variables; (4) `iprotobuf.rs:82` recompiles `Regex::new` on every call and matches `package foo;` inside `/* ... */` block comments, and `iprotobuf.rs:103, 194-196` sets `compiler.include_imports(true)` without tracking imported `.proto` files in product `inputs` — `include_paths` defaults to `src_dirs` so imports under `src_dirs` are already scanned inputs; only imports resolved from a user-set `include_paths` outside `src_dirs` go untracked.
- `src/processor/checker/icheckpatch.rs:70-76, 100`, `src/processor/checker/icheckstyle.rs:130-136, 164`, `src/processor/checker/iperlcritic.rs:124-130, 158` vs `docs/src/processor/checker/icheckpatch.md:53-55`, `docs/src/processor/checker/icheckstyle.md:62-64`, `docs/src/processor/checker/iperlcritic.md:62-64` - All three docs state that batch support is *"Not used"* with files *"checked in parallel as separate products"* (matching `ishellcheck`), but the code sets `supports_batch: true` and implements `execute_batch` via `execute_checker_batch_per_file` (`src/processor/mod.rs:795-803`, a sequential `products.iter().map(check_fn)`). Batching is on by default (`StandardConfig.batch = true`, `[build] batch_size = 0` = unlimited), so in practice all rebuilding files run sequentially on a single worker. Set `supports_batch: false` and remove `execute_batch` in those three processors (or update their docs if sequential batching was intended).
- `src/processor/` (33 plugins) - Unused `StandardConfig` fields (`command`, `args`, `formats`, `output_dir`, `batch`) are not listed in `omit_standard_fields` across 33 processor plugins, so users can set ignored fields in `rsconstruct.toml` without a config error and `rsconstruct processor defconfig` displays them. Notably, `src/processor/mass_generator/linux_module.rs:63, 369`, `src/processor/generator/tera.rs:22, 1008`, `src/processor/generator/ijinja2.rs:26, 385`, `src/processor/generator/jinja2.rs:15, 147`, and `src/processor/generator/mako.rs:15, 155` explicitly state `Unused StandardConfig fields: ...` in their doc comments while leaving `omit_standard_fields: &[]` (or incomplete). Other affected plugins: `icheckpatch`, `icheckstyle`, `idockerfile`, `ijq`, `ijsonlint`, `iperlcritic`, `isvglint`, `itaplo`, `itidy`, `ixmllint`, `iyamllint`, `iyamlschema`, `iyq`, `imarkdown2html`, `ipdfunite`, `iprotobuf`, `isass`, `ascii`, `duplicate_files`, `encoding`, `json_schema`, `license_header`, `marp_images`, `terms`, `zspell`, `markdown2html`, `pdfunite`, `yaml2json`.
- `docs/src/commands.md:11-23, 68-91, 118, 188, 275-288, 318-335, 448-478, 577-591`, `README.md:69`, `docs/src/advanced.md:36-40, 63-72`, `docs/src/completions.md:43-109`, `docs/src/getting-started.md:22, 49, 66-68, 74, 83-87, 109, 137, 155, 168-170`, `docs/src/processors.md:26-27`, `docs/src/project-structure.md:16-25, 44-47, 55` - User-facing CLI and config documentation errors:
  - `docs/src/commands.md`: missing global flags `--graph-stats`, `--no-mtime-cache`, `--color <COLOR>` (lines 11-23); missing `build` flags `--trace <FILE>`, `--iset`, `--pset` (lines 68-91); documents removed `@creator` alias (line 118; `ProcessorType::parse("creator")` returns `None`, and the code comment at `src/builder/build.rs:20` still lists it too) and `Creators` clean behavior (line 188); duplicate `rsconstruct smart auto` section (lines 275-288, plus the `smart` table at 448-463) and `rsconstruct smart enable/disable/only` examples (lines 467-474) using short names (`pylint`, `ruff`) which fail `smart::validate_name` at runtime (`parse_name` requires the `processor.` prefix); missing `graph lookup-fwd`, `graph lookup-rev`, and `graph unreferenced` subcommands (lines 318-335); `tool list-configured` missing from the `tool` table (lines 577-585, though it appears in the examples at 590-591), and `tool list` documented as "Uses config if available" while the CLI says "no config needed" (the other `Config required?` entries match).
  - `README.md:69` and `docs/src/advanced.md:36-40`: use non-existent `rsconstruct graph --view` and `rsconstruct graph --format ...` instead of `rsconstruct graph view` and `rsconstruct graph show --format ...` (bare `rsconstruct graph` lists subcommands rather than printing a text graph).
  - `docs/src/advanced.md:63-72`: documents non-existent numeric verbosity levels `-v 0`, `-v 1`, `-v 2`, `-v 3` (`-v` is a boolean flag, `src/cli.rs:33, 932`).
  - `docs/src/completions.md:43-109`: lists removed `config` top-level command (line 43); claims `build -p` / `watch -p` complete pnames from the plugin registry (lines 47-53, 104) when the bash patch injects inames read from `rsconstruct.toml` (`src/cli.rs:1233-1246`, bash only); claims `analyzer config` (line 68) and `analyzer clean --analyzer` (line 69) have completions (neither has a `value_parser`, `src/cli.rs:656-665`) and cites a non-existent `AnalyzersAction::Config` with a literal `value_parser` list (lines 85-90; the enum is `AnalyzerAction`); shows outdated bash completion labels `rsconstruct__processor__config)` etc. (lines 100-102) where the code targets `rsconstruct__subcmd__processor__subcmd__{files,config,delete,disable,enable})` (`src/cli.rs:1156-1162`; `defconfig` is not an injection target).
  - `docs/src/getting-started.md:22, 66-68, 74, 83-87, 109, 155, 168-170` and `docs/src/project-structure.md:16-25, 44-47, 55`: Getting Started processor stanzas omit `src_dirs` (which defaults to `[]`, scanning nothing — `src/config/mod.rs:299-312`), so the "Expected output ... (1 product)" at lines 49 and 137 would not happen; `project-structure.md` shows non-existent checker stub directories (`out/ruff/`, `out/pylint/`, ...; checkers write nothing, as line 51 of the same doc says, and the cc output dir is `out/processor.generator.cc_single_file/`), legacy `.rsconstruct/index.json` and `deps/` (actual contents: `db.redb deps.redb descriptors mtime.redb objects`), and bare `rsconstruct clean` (`clean` requires a subcommand; should be `rsconstruct clean outputs`).
  - `docs/src/processors.md:26-27`: says `rsconstruct processor list` lists processors in the current project and `processor list --all` lists all built-in processors (in reality `processor list` always lists all built-in processors, `--all` is the global show-all-columns flag, and `processor used` lists the current project's processors).
- `docs/src/analyzers/cpp.md:25`, `docs/src/processor/explicit/generic.md:39, 52, 82-84, 95`, `docs/src/processor/generator/a2x.md:21-22, 30-31`, `docs/src/processor/mass_generator/cc.md:138, 160, 162`, `docs/src/processor/mass_generator/linux_module.md:110, 119, 121`, `docs/src/processor/checker/clippy.md:47`, `docs/src/processor/checker/make.md:93`, `docs/src/processor/checker/script.md:42`, `docs/src/processor/mass_generator/cargo.md:56`, `docs/src/processor/checker/aspell.md:19-32`, `docs/src/processor/checker/zspell.md:37-54`, `docs/src/processor/generator/marp.md:23-40`, `docs/src/processor/generator/pdflatex.md:20-37`, `docs/src/processor/generator/requirements.md:99-104`, `docs/src/processor/generator/tags.md:233-243, 257-285`, `docs/src/processor/checker/README.md:7-9`, `docs/src/processor/generator/README.md:8-9` - Per-processor and per-analyzer documentation mismatches against code schemas:
  - `docs/src/analyzers/cpp.md:25` documents `include_scanner = "builtin"` under `[analyzer.cpp]`, which does not exist on `CppAnalyzerConfig` (`#[serde(deny_unknown_fields)]`, `src/config/analyzer_configs.rs:8-37`) and fails config parsing (the field belongs to `processor.generator.cc_single_file`; `icpp` is a separate analyzer).
  - `docs/src/processor/explicit/generic.md:39, 52, 82-84, 95` documents a non-existent `outputs` config field and `--outputs` CLI flag; the actual config fields are `output_files` and `output_dirs`, and the flags passed to `command` are `--output-files` and `--output-dirs` (the doc's own Clean section at line 134 uses the correct names).
  - `docs/src/processor/generator/a2x.md:21-22, 30-31` documents non-existent `a2x` and `format` fields (actual field is `command`, and format is fixed to `"pdf"`, `src/processor/generator/a2x.rs:31`).
  - `docs/src/processor/mass_generator/cc.md:138, 160` and `docs/src/processor/mass_generator/linux_module.md:110, 119` document a non-existent `loose_manifest` field (exists only on mdbook/jekyll/zola), and `cc.md:162`, `linux_module.md:121`, `clippy.md:47`, `make.md:93`, `script.md:42`, and `cargo.md:56` list the `src_dirs` default as `[""]` instead of `[]`.
  - Missing custom `FieldSpec` fields in docs: `aspell.md` omits `auto_add_words` and `words_file`; `zspell.md` omits `dict_dir`; `marp.md` omits `timeout_secs` and `max_attempts`; `pdflatex.md` omits `shell_escape`; `requirements.md` omits `python_paths`; `tags.md` omits `check_unused`, `similar_files_limit`, `suggested_tags_limit`, `common_tags_limit`, and the `rsconstruct tag collect` subcommand.
  - `docs/src/processor/checker/README.md:7-9` (11 names) and `docs/src/processor/generator/README.md:8-9` (2 names) claim 13 processors have no dedicated doc page yet, all of which now have dedicated `.md` pages.

## Low

- `src/object_store/checksums.rs:1` (`src/object_store/mod.rs:2`), `src/object_store/mod.rs:142`, and `src/processor/mod.rs:562-563, 606-609` - Dead code and stale `#[allow(dead_code)]`: `src/object_store/checksums.rs` is a 1-line placeholder file (`// Checksum logic has moved to src/checksum.rs`); `remote_pull: bool` at `src/object_store/mod.rs:142` still carries `#[allow(dead_code)]` (the only one left in `src/`) even though it is now read in `src/object_store/operations.rs:60, 92`; and `DirectoryProductOpts::output_dir_name` is `None` at both call sites (`src/processor/checker/make.rs:75-86`, `src/processor/checker/clippy.rs:90-101`), making the `output_dir_name.map(|d| anchor_dir.join(d))` branch at `src/processor/mod.rs:606-609` dead code (its doc comment at 562-563 still says "enabling directory-level caching for creators").
- `src/config/processor_configs.rs:197-309` and `src/processor/checker/{ijq.rs:4, ijsonlint.rs:4, itaplo.rs:4, ascii.rs:4, encoding.rs:4, duplicate_files.rs:6, json_schema.rs:6, marp_images.rs:6}` - `src/config/processor_configs.rs` contains 26 obsolete `// FooConfig lives in ...` comment tombstones, 14 empty `// --- tidy (HTML validator) ---` style section headers, and 8 `pub type X = CheckerConfig;` aliases (`IjqConfig`, `IjsonlintConfig`, `ItaploConfig`, `AsciiConfig`, `EncodingConfig`, `DuplicateFilesConfig`, `JsonSchemaConfig`, `MarpImagesConfig`), each used by exactly one processor file, that violate the One-File Processor Rule in `CLAUDE.md` (compare `icheckpatch.rs`, which uses `StandardConfig` directly).
- `src/config/tests.rs:772-813`, `tests/processor/markdown.rs:1-95`, and `tests/processor/` - Test suite hygiene and completeness issues:
  - `tests/processor/markdown.rs` tests `processor.generator.markdown2html` but is placed at `tests/processor/markdown.rs` instead of `tests/processor/generator/markdown2html.rs` (requiring `"processor.generator.markdown2html"` in `TESTS_ALLOWLIST` at `src/config/tests.rs:784`) and names its test functions `markdown_*` instead of `markdown2html_*`.
  - `every_plugin_has_docs_and_tests` (`src/config/tests.rs:772-813`) does not assert that entries in `TESTS_ALLOWLIST` actually lack a test file, and 9 entries in `TESTS_ALLOWLIST` need no external tool: `encoding`, `ijq`, `ijsonlint`, `imarkdown2html`, `ipdfunite`, `itaplo`, `marp_images`, `yaml2json` are in-process (`required_tools()` empty), and `explicit.generic` only runs whatever command the test configures.
  - Test function naming convention violations (`docs/src/internal/coding-standards.md:79-83`, "Test functions for a processor must be prefixed with the processor name"): `tests/processor/checker/iyamlschema.rs:229, 263`, `tests/processor/checker/script.rs:230, 258`, `tests/processor/generator/generic.rs:7` (prefixed `generator_`), `tests/processor/mass_generator/generic.rs:92-529` (21 functions prefixed `mass_generator_`), and `tests/processor/generator/tera.rs:86-1448` (35 of 36 functions missing `tera_`).
  - `tests/processor/generator/libreoffice.rs:21-31` writes `"placeholder"` to `slides.odp` and never asserts `output.status.success()` (its comment admits "This may fail if the ODP content is invalid"); `tests/processor/mass_generator/jekyll.rs:1` and `tests/processor/checker/slidev.rs:1` use `no_project` (generating only a `*_no_project_discovered` test); and the real-input tests in `tests/processor/generator/{pdfunite.rs:51, mermaid.rs:26, drawio.rs:27}` only run `--dry-run` (each file's `*_no_project_discovered` test does run a real `build` on an empty project).
- `scripts/migrate_processor_names.py:5-8, 46-54, 112-122` and `scripts/ci-install-tools.repo.sh:25` - `scripts/migrate_processor_names.py` still maps `out/{cc,linux-module,gem}` to `out/processor.creator.*` (should be `out/processor.mass_generator.*`) and rewrites `[processor.creator.X]` to `[processor.creator.generic.X]`; `scripts/ci-install-tools.repo.sh:25` still refers to "the Build step's cargo creator" (cargo is `processor.mass_generator.cargo`).
- `docs/src/future-features.md:9-17, 89-90` and `docs/src/internal/licensing.md:1-2` - `docs/src/future-features.md` claims remote cache pull is unimplemented `#[allow(dead_code)]` (lines 9-17; in fact `try_fetch_object_from_remote` is called from the read path at `src/object_store/operations.rs:68`, gated by `remote_pull`) and has stray literal `</content>` and `</invoke>` tags at the end of the file (lines 89-90); `docs/src/internal/licensing.md` is an empty stub containing only `# Licensing of Ported Engines` (linked from `docs/src/SUMMARY.md:156`).
- `CLAUDE.md:11, 28, 29, 30`, `ARCHITECTURE.md:245, 256-258, 287, 295`, `plan.md:127, 195`, `docs/src/internal/{coding-standards.md:32-72, 252, processor-contract.md:51, 63-67, 74, 92-105, architecture.md:50, 313-320, cache.md:38, 70, 173, 187, 239, internal-processors.md:26-37, 239-307, strictness.md:45-103, shared-output-directory.md:9-171, output-routing.md:6-355, output-prediction.md:3-293, processor-ordering.md:100-162, distributed-execution.md:210}`, `docs/src/processor/mass_generator/generic.md:14-23, 178-197, 214`, `docs/src/nomenclature.md:52`, `docs/src/configuration.md:110-136, 326`, `src/builder/build.rs:20`, `src/graph.rs:35, 388, 799-800`, `src/registries/{mod.rs:7, processor.rs:3, 24, 51, 69, 132}`, `src/executor/{execution.rs:214, 233, 296, handlers.rs:229}`, `src/object_store/descriptors.rs:223`, `src/processor/{mod.rs:451, 477, 495, 563, 983, 1487-1488, mass_generator/generic.rs:60}`, `tests/processor/shared_output_dir.rs:3-22, 184` - Stale comments and documentation. The `creator` processor type is fully removed (no `src/processor/creator/`, no `ProcessorType::Creator`, no `cache_output_dir` anywhere in `src/`), yet: `CLAUDE.md:30` still says the creators `cargo, cc, gem, generic, jekyll, linux_module, mdbook, npm, pip` "are to be converted one by one" although all of them already live in `src/processor/mass_generator/`; `CLAUDE.md:11, 29` list `creator` among the processor types; `CLAUDE.md:28` says "the four config-driven processors are all called `generic`" (there are three: `explicit`, `generator`, `mass_generator`); all the cited `src/` and `tests/` lines are comments or test names mentioning creators / `processor.creator.generic` / `@creator`; the cited docs describe creators, `cache_output_dir` (`processor-contract.md` has a whole "Output directory caching (creators)" section), `--type creator`, and `[processor.creator.venv]`; `plan.md:127` lists a `uv` / `pip` creator and `plan.md:195` counts a `creator` column. `ARCHITECTURE.md:245, 256-258, 287, 295` reference removed `fix()` / `can_fix` / `config_has_fix()` / `registries::can_fix`. `docs/src/internal/coding-standards.md:32-68` describes the pre-`inventory` multi-place registration (`processor_enabled_field()`, `default_processors()`, `create_builtin_processors()`, `pub {name}: {PascalCase}Config` on `ProcessorConfig`), none of which exist (registration is 111 `inventory::submit!` entries), and lines 72, 252 still list `src/processor/creator/`. `docs/src/internal/strictness.md:45` says "5 allows remain" (1 actual) and line 100 documents an allow on `ProcessorPlugin.processor_type` that no longer exists, plus allows on `try_fetch_*` that are gone, and lists `Creator` as a type. `docs/src/internal/internal-processors.md:26-37, 239-307` lists `ijq`/`ijsonlint`, `itaplo`, `ijson_schema`, `imarkdown2html`, `iyamlschema`, `yaml2json`, and `isass` under `## Candidates` as proposals although all exist in `src/processor/`; only `ipdfunite` is under `## Implemented`.