feat: doc-staleness context — mention-based reverse-reference scan

[aliasunder]

Summary

• Extend the workspace scan to doc files (.md, .json) so unchanged docs that describe changed code reach the LLM prompt and staleness becomes a flaggable correctness finding
• Mention matching is analogous to the import-trace scan: full-path substring matches outrank basename-only, and generic basenames (index, main, config, etc.) require a parent directory segment to avoid false positives
• Separate candidate pool (RELATED_DOCS_MAX = 4) and token budget (consumes what's left after code-related files)

Changes by module

• src/context/doc-mentions.ts (new) — pure leaf module: findMentionedChangedPaths (mention extraction with generic-basename guard), byMentionRelevance (ranking comparator), DOC_EXTENSIONS, GENERIC_BASENAMES
• src/context/workspace.ts — extracted shared BFS walker (scanWorkspaceFiles), new findRelatedDocs method on ContextReader, RELATED_DOCS_MAX = 4
• src/review/prompt.ts — new relatedDocs param, doc section between related files and diff with staleness instruction header
• src/review/phases.ts — extended DIMENSION_CORRECTNESS_SECURITY with unchanged-doc staleness instruction
• src/orchestrate.ts — relatedDocs in ReviewContext, budget chain (remaining after related files → doc scan), wiring through to buildUserPrompt
• AGENTS.md — module map updated
• README.md — doc-staleness moved from "In progress" to "Shipped"

Test plan

• 15 new tests in doc-mentions.test.ts — full-path match, basename match, generic guard, ranking, constants
• 10 new tests in workspace.test.ts — find docs, exclude conventions/changed docs, cap, budget, ranking, binary skip, JSON extension
• 3 new tests in prompt.test.ts — section ordering with docs, absent when empty, staleness header
• 1 new test in phases.test.ts — staleness instruction present
• 5 new tests in orchestrate.test.ts — budget chain, conventionsFile passthrough, relatedDocs in review context, conditional skip
• npm test — 242 passing (31 new)
• npm run lint — clean
• npm run build — clean
• docker build . — succeeds

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features
  • Added doc-mention scanning with relevance ranking for documentation staleness detection.
  • Added token-budgeted relatedDocs to the review context and included a related-docs section in prompts, with stronger documentation verification guidance.
  • Added action inputs for priority_docs and configurable scan/related caps.
• Documentation
  • Updated README and workspace agent structure docs to describe doc-mention scanning and limits.
• Refactor
  • Unified workspace scanning and improved related-candidate handling (including skipping empty content).
• Tests
  • Expanded suites for mention matching/ranking, token budgeting, caps/exclusions, priority behavior, and prompt/context wiring.

[sourcery-ai]

Sorry @aliasunder, you have reached your weekly rate limit of 2500000 diff characters.

Please try again later or upgrade to continue using Sourcery

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

The change adds documentation-file scanning for changed-path mentions, selects related docs within token and count limits, passes them through orchestration, renders them in prompts, and adds documentation-staleness review guidance and coverage.

Changes

Related documentation context

Layer / File(s)	Summary
Configuration and workspace scanning action.yml, src/config.ts, src/main.ts, src/context/workspace.ts, src/context/__tests__/*, .coderabbit.yaml, AGENTS.md	Adds configurable scan and related-context limits, priority documentation inputs, shared workspace scanning, and related-file cap handling.
Documentation mention matching src/context/doc-mentions.ts, src/context/__tests__/doc-mentions.test.ts	Adds full-path and basename mention detection, generic-basename handling, deterministic relevance ordering, and tests for matching behavior.
Workspace documentation discovery src/context/workspace.ts, src/context/__tests__/workspace.test.ts, fixtures/workspace/docs/*	Adds priority and mention-matched document selection, binary and exclusion filtering, token budgeting, result limits, fixtures, and tests.
Orchestration context wiring src/orchestrate.ts, src/__tests__/orchestrate.test.ts	Adds relatedDocs to review context, allocates remaining budget after related files, invokes document discovery, and tests the wiring and disabled behavior.
Prompt and review integration src/review/prompt.ts, src/review/phases.ts, src/review/__tests__/*, README.md	Renders related docs between related files and the diff, adds documentation-staleness instructions, updates tests, and documents the shipped scan behavior.

Estimated code review effort: 4 (Complex) | ~45 minutes

Possibly related PRs

• aliasunder/umm-actually#3: Introduces the workspace context reader later extended with findRelatedDocs.
• aliasunder/umm-actually#7: Introduces orchestration and prompted generation later extended with related-document context.
• aliasunder/umm-actually#1: Introduces the correctness/security instruction constant extended here with documentation-verification checks.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title is concise and correctly highlights the main change: adding doc-staleness context via a mention-based scan.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch worktree-doc-reverse-reference

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Findings beyond the diff

These are in code the changes touch or depend on, outside the diff's line ranges:

• [medium/correctness] Pre-existing: BFS walk follows symlinked directories outside workspace — src/context/workspace.ts:279 (confidence: high)
  scanWorkspaceFiles uses entry.isDirectory() to decide whether to recurse, which returns true for directory symlinks. A symlink like docs → /etc would be pushed onto the queue and the walk would descend into the external filesystem. This pre-existing issue now affects the doc scan as well as the source scan. The directory-level pruned check catches .name patterns but not symlink targets.
  Failure scenario: A repo contains docs → /etc as a symlink (committed or created by a prior step). The BFS walk enters /etc, reads all files with matching extensions, and includes their contents in the prompt — potentially exposing sensitive host filesystem data.

Suggested fix

Check `entry.isSymbolicLink()` before recursing into directories, or resolve and validate the real path of each directory entry before pushing it onto the BFS queue.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

umm-actually re-reviewed at 4400830

1 new finding(s) posted as inline comments (29 total across all reviews).

---

umm-actually · deepseek/deepseek-v4-pro

[coderabbitai]

Actionable comments posted: 3

🧹 Nitpick comments (1)

src/context/__tests__/doc-mentions.test.ts (1)

189-193: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Assert the complete generic-basename set.

This test still passes when expected entries such as helpers or constants are removed. Assert the whole deterministic Set instead.

As per coding guidelines, “Prefer exact whole-value or whole-call-parameter assertions over loose matchers.”

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/context/__tests__/doc-mentions.test.ts` around lines 189 - 193, Update
the GENERIC_BASENAMES test to assert the entire Set value against the complete
expected deterministic set, including helpers and constants, rather than
checking individual memberships in a loop. Preserve the test’s coverage of all
expected common stems through one exact whole-value assertion.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/context/__tests__/workspace.test.ts`:
- Around line 653-677: Strengthen the exclusion tests around
contextReader.findRelatedDocs by adding deterministic fixtures for both the
excluded document and a matching document that must remain, then assert the
complete returned document value with an exact whole-value assertion. Update
both “excludes the conventions file from results” and “excludes changed docs
from results” so they fail if findRelatedDocs returns an empty array, while
preserving the intended exclusions.

In `@src/context/doc-mentions.ts`:
- Around line 44-69: Update the mention matching in the changed-path loop of
src/context/doc-mentions.ts (lines 44-69) to require complete path-like token
boundaries for full paths, generic two-segment paths, and basenames, preventing
matches inside longer suffixes or prefixed filenames while preserving valid
mentions. Add regression tests in src/context/__tests__/doc-mentions.test.ts
(lines 11-137) covering longer path suffixes and prefixed basenames.
- Around line 76-87: Update byMentionRelevance in src/context/doc-mentions.ts
(lines 76-87) to return 0 when a.path === b.path, while preserving the existing
relevance and path ordering. Add an assertion in
src/context/__tests__/doc-mentions.test.ts (lines 175-181) verifying equal
candidate paths compare as zero.

---

Nitpick comments:
In `@src/context/__tests__/doc-mentions.test.ts`:
- Around line 189-193: Update the GENERIC_BASENAMES test to assert the entire
Set value against the complete expected deterministic set, including helpers and
constants, rather than checking individual memberships in a loop. Preserve the
test’s coverage of all expected common stems through one exact whole-value
assertion.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 83e78efe-ce8f-405d-ac75-e95d8bb7ce3f

📥 Commits

Reviewing files that changed from the base of the PR and between af70f65 and ede0fa1.

📒 Files selected for processing (15)

• AGENTS.md
• README.md
• fixtures/workspace/docs/api.md
• fixtures/workspace/docs/config.json
• fixtures/workspace/docs/internal.md
• src/__tests__/orchestrate.test.ts
• src/context/__tests__/doc-mentions.test.ts
• src/context/__tests__/workspace.test.ts
• src/context/doc-mentions.ts
• src/context/workspace.ts
• src/orchestrate.ts
• src/review/__tests__/phases.test.ts
• src/review/__tests__/prompt.test.ts
• src/review/phases.ts
• src/review/prompt.ts

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[aliasunder]

Re: umm-actually "Findings beyond the diff" — BFS walk follows symlinked directories

The finding claims entry.isDirectory() returns true for directory symlinks, allowing the BFS walk to follow symlinks outside the workspace. This is incorrect for Dirent objects returned by readdir({withFileTypes: true}):

• Dirent.isDirectory() returns false for symlinks
• Dirent.isSymbolicLink() returns true for symlinks
• Both isFile() and isDirectory() return false — symlinks are their own entry type

The BFS walk at scanWorkspaceFiles (workspace.ts:287) checks entry.isDirectory() for recursion — symlinked directories are not recursed into. And entry.isFile() at line 293 filters out symlinked files. The scan is already safe against symlink escapes at both the directory and file level.

---

🔍 ship-check · pr-monitor · claude-opus-4-6[1m]

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/context/workspace.ts`:
- Around line 223-239: Update readPriorityDocOrNull to resolve each priority
document through the existing realPathIfSafe check before reading it, rather
than directly joining resolvedRoot and docPath. Skip and return null when the
resolved path is outside the reviewable workspace tree or otherwise unsafe,
while preserving the existing missing-file and unreadable-file logging behavior.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: fb41c15c-ff49-4317-a2cc-6f7cb3631c19

📥 Commits

Reviewing files that changed from the base of the PR and between ede0fa1 and 9123a5a.

📒 Files selected for processing (14)

• AGENTS.md
• README.md
• action.yml
• src/__tests__/config.test.ts
• src/__tests__/orchestrate.test.ts
• src/config.ts
• src/context/__tests__/doc-mentions.test.ts
• src/context/__tests__/workspace.test.ts
• src/context/doc-mentions.ts
• src/context/workspace.ts
• src/main.ts
• src/orchestrate.ts
• src/review/__tests__/prompt.test.ts
• src/review/prompt.ts

🚧 Files skipped from review as they are similar to previous changes (7)

• AGENTS.md
• src/context/doc-mentions.ts
• src/orchestrate.ts
• src/context/tests/doc-mentions.test.ts
• src/review/tests/prompt.test.ts
• src/review/prompt.ts
• src/tests/orchestrate.test.ts

[umm-actually]

Reviewed — 3 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[aliasunder]

Phase 1: PR Review

1 finding across 1 file.

Verdict: ship-with-minor-fixes

---

🔍 ship-check · Phase 1 (PR Review) · deepseek/deepseek-v4-pro

[aliasunder]

Phase 2: Code Quality

4 findings across 2 files.

Category	Count
Naming	3
Structure	1
Module conventions	0
Comments	0
Simplicity	0
Error handling	0

---

🔍 ship-check · Phase 2 (Code Quality) · deepseek/deepseek-v4-pro

[aliasunder]

Phase 3: Test Audit

1 coverage gap across 1 file. All 306 tests pass (18 test files, 0 failures).

Summary

• Test files audited: 6 (config, orchestrate, doc-mentions, workspace, phases, prompt)
• New/modified it() blocks reviewed: ~35 across all files
• Two-bar rule: All new tests pass both bars — no silent no-ops, no wrong-error passes, no early-return passes
• Assertion quality: Consistent use of toEqual/toBe on deterministic output; substring matching (toContain) used appropriately for large prompt strings and log assertions on non-deterministic fields (error messages, timestamps)
• Test hygiene: const per test, explicit callback parameter names, no ! assertions, onTestFinished cleanup in temp-workspace tests, beforeEach not abused
• Coverage regressions: None — no removed tests, no weakened assertions, no .skip'd tests
• Coverage gaps: 1 (soft) — see inline

---

🔍 ship-check · Phase 3 (Test Audit) · deepseek/deepseek-v4-pro

[aliasunder]

Phase 4: Bug Check

3 bugs found across 2 files. All would-fix (trivial, high confidence).

By dimension:

• D1 (Description mismatch): 1 — "always included" claim for priority docs is contradicted by budget skip
• D3 (Type safety / Boundary): 1 — dotfiles silently skipped in mention detection due to lastIndexOf(".") edge case
• D5 (Behavioral consistency): 1 — readPriorityDocOrNull lacks symlink resolution (asymmetric with readConventions)

Confidence: 3 high, 0 medium, 0 low

---

🔍 ship-check · Phase 4 (Bug Check) · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 2 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 2 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 2 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Findings beyond the diff

These are in code the changes touch or depend on, outside the diff's line ranges:

• [high/security] Pre-existing: Scan candidates read through symlinks without verifying the real path stays within the workspace — src/context/workspace.ts:196 (confidence: high)
  readScannedFileOrNull calls readFile on a path built from the workspace BFS without resolving symlinks. A symlinked file pointing outside the workspace or into .git would be read and its content included in the LLM prompt. The new findRelatedDocs inherits this gap. Changed-file and priority-doc paths are already protected by realPathIfSafe.
  Failure scenario: A repository contains a symlink docs/secret.md → /etc/passwd. The workspace BFS finds docs/secret.md, findRelatedDocs reads it without checking the symlink target, and the content of /etc/passwd appears in the prompt and potentially in the public review.

Suggested fix

Apply realPathIfSafe in readScannedFileOrNull so that scan candidate content cannot escape the workspace via symlinks. Degrade to null (skip) when the real path is outside the safe zone, logging a warning as done for changed files.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/context/__tests__/workspace.test.ts (1)

922-939: 📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win

Make the negative-path tests fail on a no-op implementation.

Each test currently passes if its method always returns an empty result. Add a valid companion document and assert the complete result after the excluded candidate is skipped.

• src/context/__tests__/workspace.test.ts#L922-L939: include a second matching document and assert it remains after README.md is excluded.
• src/context/__tests__/workspace.test.ts#L1087-L1101: include a smaller priority document and assert it is read after the oversized document is skipped.
• src/context/__tests__/workspace.test.ts#L1115-L1120: include a valid priority document and assert it is read after the binary document is skipped.

As per coding guidelines, “Every test must fail when the behavior breaks” and must guard against silent no-op passes.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/context/__tests__/workspace.test.ts` around lines 922 - 939, Strengthen
the negative-path tests in src/context/__tests__/workspace.test.ts at lines
922-939, 1087-1101, and 1115-1120 so they cannot pass when the implementation
always returns an empty result: add a valid companion document to the
mention-matched test and assert it remains after README.md is excluded; add a
smaller-priority document and assert it is read after the oversized document is
skipped; and add a valid-priority document and assert it is read after the
binary document is skipped. Preserve each test’s existing exclusion behavior
while asserting the complete expected result.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/context/workspace.ts`:
- Around line 528-535: Update the exclusion logging around rankedCandidates and
relatedDocs so it reports only candidates rejected specifically because the
related-document cap was reached. Track or derive the actual cap-rejected
candidates during selection rather than comparing lengths or slicing
rankedCandidates, ensuring budget-skipped candidates and later included
documents are not reported as excluded.

---

Outside diff comments:
In `@src/context/__tests__/workspace.test.ts`:
- Around line 922-939: Strengthen the negative-path tests in
src/context/__tests__/workspace.test.ts at lines 922-939, 1087-1101, and
1115-1120 so they cannot pass when the implementation always returns an empty
result: add a valid companion document to the mention-matched test and assert it
remains after README.md is excluded; add a smaller-priority document and assert
it is read after the oversized document is skipped; and add a valid-priority
document and assert it is read after the binary document is skipped. Preserve
each test’s existing exclusion behavior while asserting the complete expected
result.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 10080b78-397c-4c69-ae09-d8a2af868c9f

📥 Commits

Reviewing files that changed from the base of the PR and between 24f0823 and 23d4492.

📒 Files selected for processing (9)

• README.md
• src/__tests__/orchestrate.test.ts
• src/config.ts
• src/context/__tests__/workspace.test.ts
• src/context/doc-mentions.ts
• src/context/workspace.ts
• src/orchestrate.ts
• src/review/__tests__/prompt.test.ts
• src/review/prompt.ts

🚧 Files skipped from review as they are similar to previous changes (5)

• src/review/tests/prompt.test.ts
• src/review/prompt.ts
• src/config.ts
• src/context/doc-mentions.ts
• src/tests/orchestrate.test.ts

[umm-actually]

Reviewed — 2 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Findings beyond the diff

These are in code the changes touch or depend on, outside the diff's line ranges:

• [high/correctness] Update findRelatedFiles return type to include excludedByCapPaths — src/context/workspace.ts:27 (confidence: high)
  Pre-existing: The ContextReader interface and createContextReader factory both declare findRelatedFiles returning Promise<PromptFile[]>, but the implementation since the related-docs refactor now returns { files: PromptFile[]; excludedByCapPaths: string[] }. The interface is out of sync with the implementation. Callers using the interface type (e.g. through dependency injection in tests) receive a type mismatch.
  Failure scenario: When the ContextReader interface is used as a type constraint, a stub returning PromptFile[] would pass type checking but the real implementation returns the wrapper object. Consumers that destructure files from the result get undefined (array destructuring on an object), leading to silent failures or undefined is not iterable at runtime.

Suggested fix

Update the `findRelatedFiles` return type in both the `ContextReader` interface and `createContextReader` return type to:
```typescript
findRelatedFiles: (params: {
  changedPaths: string[]
  budgetTokens: number
}) => Promise<{ files: PromptFile[]; excludedByCapPaths: string[] }>
```

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

Context notes

• Priority docs not included: README.md (missing, unreadable, or over budget)

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

Reviewed — 1 finding(s) posted as inline comments.

Context notes

• Priority docs not included: README.md (missing, unreadable, or over budget)

---

umm-actually · deepseek/deepseek-v4-pro

[umm-actually]

[medium/correctness] Add byte-size guard for priority docs before reading (confidence: medium)

readPriorityDocs reads the entire file without checking its size first. A large priority doc can exhaust memory before the token budget check. readChangedFiles already guards against this with a stat-based byte check; readPriorityDocs should do the same.

Failure scenario: A user sets priority_docs to a path pointing to a multi-GB file (e.g., a binary blob). The action reads the entire file into memory, causing an OOM crash or excessive memory usage before the token budget check can skip it.

Suggested fix

Before reading the file, stat it and skip if `fileStats.size > remainingTokens * CHARS_PER_TOKEN`, similar to the guard in `readChangedFiles`.