devsplain 2.3.0 → 2.3.2

package/README.md

@@ -18,7 +18,7 @@ Unlike interactive AI editors, `devsplain` is designed for batch documentation p
 - **Comment Preservation & Tagging**: AI-generated comments are tagged with `[ds]`. Your manually written comments are safe and will never be touched by the engine.
 - **Local Deterministic Scrubber**: The `--clean` flag strips AI-generated `[ds]` comments locally using a deterministic lexical state machine—no LLM calls, API keys, or internet required.
 - **Git Hook Automation**: Supports an automated two-commit Git hook workflow (`pre-commit` for quality, `post-commit` for auto-generated documentation commits) that prevents recursive commit loops.
-- **Bring Your Own LLM**: Native setup wizard for Groq, Gemini, OpenAI, or any OpenAI-compatible API endpoint (like Ollama or LMStudio).
+- **Bring Your Own LLM**: Native setup wizard for Groq, Gemini, OpenAI, Claude, DeepSeek, or any OpenAI-compatible API endpoint (like Ollama or LMStudio).
 - **Exponential Backoff**: Resilient AI request handler that automatically retries rate-limited requests with exponential backoff.
 - **Headless & Override Control**: Configure via environment variables or override global config settings dynamically on the fly with command-line flags.
 
@@ -109,10 +109,12 @@ devsplain <file-or-directory> [options]
 | `--force` | Bypasses the safety block check that prevents running `devsplain` on a dirty Git working tree. |
 | `--clean` | Scrubber mode. Deterministically removes only devsplain-generated comments tagged with `[ds]`, preserving your manual comments. |
 | `--prune` | Destructive scrubber mode. Removes ALL comments and docstrings from source files, including your own manual comments. |
-| `--provider <name>`| Temporary one-off override for the AI provider (`gemini`, `groq`, `openai`, `custom`) for this command run only (does not modify the saved config file). |
+| `--provider <name>`| Temporary one-off override for the AI provider (`gemini`, `groq`, `openai`, `claude`, `deepseek`, `custom`) for this command run only (does not modify the saved config file). |
 | `--model <name>` | Temporary one-off override for the model name for this command run only. |
 | `--api-key <key>` | Temporary one-off override for the API key for this command run only. |
 | `--base-url <url>` | Temporary one-off override for the API base URL for this command run only. |
+| `--concurrency <num>` | Temporary one-off override for file concurrency level (1–5, default: 2). |
+| `--chunk-size <lines>` | Temporary one-off override for chunk line size (e.g. 200, 600, 1000). |
 | `--config` | Relaunches the configuration setup wizard. |
 | `--setup-hook` | Installs Git pre-commit and post-commit hooks in the repository. |
 | `--help, -h` | Displays the help menu. |

package/bin/cli.js

@@ -10,7 +10,10 @@ const { execSync } = require('child_process');
 let rl;
 let askQuestion;
 
-// Checks if the Git repository is dirty [ds]
+/** [ds]
+ * Determines if the git working tree has uncommitted changes.
+ * Silently returns false if not in a git repo or git is unavailable.
+*/
 function isGitDirty() {
     try {
         const gitDir = execSync('git rev-parse --is-inside-work-tree', { stdio: ['ignore', 'pipe', 'ignore'], encoding: 'utf8' }).trim();
@@ -23,13 +26,21 @@ function isGitDirty() {
     return false;
 }
 
-// Determines if a line is inside a string literal [ds]
+/** [ds]
+ * Determines whether a given line index resides inside a string literal or comment
+ * block for a given file extension. This is a hand-written heuristic lexer used to
+ * avoid mis-attributing comment insertions that land inside multi-line strings or
+ * block comments. It intentionally supports a wide menu of languages (Python,
+ * HTML-family, Rust/Swift nested block comments, C++ raw strings, JS/TS regex literals).
+*/
 function isLineInsideString(lines, targetLineIndex, ext = '') {
+    // Language-family flags derived from file extension drive all downstream branching. [ds]
     const isPython = ext.toLowerCase() === '.py';
     const isHTML = ['.html', '.vue', '.svelte'].includes(ext.toLowerCase());
     const isRustOrSwift = ['.rs', '.swift'].includes(ext.toLowerCase());
     const isCpp = ['.cpp', '.cc', '.cxx', '.c', '.h', '.hpp'].includes(ext.toLowerCase());
     const isJS = ['.js', '.jsx', '.ts', '.tsx'].includes(ext.toLowerCase());
+    // Tracks lexical context across the entire prefix of the file up to targetLineIndex. [ds]
     let inBacktick = false;
     let inTripleDouble = false;
     let inTripleSingle = false;
@@ -41,10 +52,12 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
     let inCppRawString = false;
     let cppRawDelimiter = '';
     let inRegex = false;
+    // Scan every prior line character-by-character, carrying state forward. [ds]
     for (let i = 0; i < targetLineIndex; i++) {
         const line = lines[i];
         let j = 0;
         while (j < line.length) {
+            // Inside /* */ comments: Rust/Swift allow nesting, so we track depth. [ds]
             if (inBlockJS) {
                 if (line.slice(j, j + 2) === '/*') {
                     if (isRustOrSwift) blockDepthJS++;
@@ -64,6 +77,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                 j++;
                 continue;
             }
+            // C++ raw strings (R"delim(...)delim") suppress all quote/comment parsing until terminator. [ds]
             if (inCppRawString) {
                 if (line.slice(j, j + 2 + cppRawDelimiter.length) === ')' + cppRawDelimiter + '"') {
                     inCppRawString = false;
@@ -73,6 +87,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                 j++;
                 continue;
             }
+            // JS regex literal: must count preceding backslashes to detect escaped slashes. [ds]
             if (inRegex) {
                 let escaped = false;
                 let k = j - 1;
@@ -95,10 +110,12 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                 j++;
                 continue;
             }
+            // Comment detection runs BEFORE quote toggling so comment markers inside strings are not misread. [ds]
             // Check if comment starts (skip processing quotes if we are entering a comment)
             if (!inSingle && !inDouble && !inBacktick && !inTripleSingle && !inTripleDouble) {
                 if (isPython) {
                     if (line[j] === '#') {
+                        // Python comment: bail out of rest of the line entirely. [ds]
                         break; // Ignore rest of line
                     }
                 } else if (isHTML) {
@@ -134,6 +151,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                         if (ext.toLowerCase() === '.php' && line[j] === '#') {
                             break; // Ignore rest of line
                         }
+                        // Detect C++ raw string opener R"delim( — delimiter can be up to 16 chars, no whitespace/parens/backslash. [ds]
                         if (isCpp && line[j] === 'R' && line[j+1] === '"') {
                             const match = line.slice(j).match(/^R"([^()\\\s]{0,16})\(/);
                             if (match) {
@@ -143,6 +161,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                                 continue;
                             }
                         }
+                        // Disambiguate '/' as regex-literal vs division by inspecting the prior non-whitespace token. [ds]
                         if (isJS && line[j] === '/') {
                             let k = j - 1;
                             while (k >= 0 && /\s/.test(line[k])) k--;
@@ -151,6 +170,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                                 isRegex = true;
                             } else {
                                 const prevChar = line[k];
+                                // Slash after these punctuators/keywords is a regex literal, not division. [ds]
                                 if (/[=({\[:,;!+*&|?<>-]/.test(prevChar)) {
                                     isRegex = true;
                                 } else {
@@ -169,6 +189,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                     }
                 }
             }
+            // Python triple-quoted strings may span multiple lines and toggle on triple delimiters. [ds]
             if (isPython) {
                 if (!inTripleSingle && !inSingle && !inDouble) {
                     if (line.slice(j, j + 3) === '"""') {
@@ -185,12 +206,14 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
                     }
                 }
             }
+            // Backtick template literals are exclusive to non-Python languages; honor escape sequences. [ds]
             if (!inTripleSingle && !inTripleDouble) {
                 if (!isPython) {
                     if (!inSingle && !inDouble) {
                         if (line[j] === '`') {
                         let escaped = false;
                         let k = j - 1;
+                        // Count consecutive backslashes by toggling; odd count means escaped [ds]
                         while (k >= 0 && line[k] === '\\') {
                             escaped = !escaped;
                             k--;
@@ -227,6 +250,7 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
             }
             j++;
         }
+        // C-family and similar languages reset string/quote state per line (unlike backtick templates or Python triple-quotes) [ds]
         const resetsAtLineEnd = ['.js', '.jsx', '.ts', '.tsx', '.java', '.c', '.cpp', '.cs', '.go', '.swift', '.kt', '.dart'].includes(ext.toLowerCase());
         if (resetsAtLineEnd) {
             inSingle = false;
@@ -237,7 +261,18 @@ function isLineInsideString(lines, targetLineIndex, ext = '') {
     return inBacktick || inTripleDouble || inTripleSingle || inSingle || inDouble || inCppRawString || inRegex;
 }
 
-// Analyzes comments in the given lines of code [ds]
+/** [ds]
+ * Analyzes source lines to determine if the file ends inside an unterminated
+ * string, template literal, raw string, regex, or block comment.
+ *
+ * Handles language-specific quoting rules: Python triple-quotes, C++ raw
+ * strings, JS template literals, JS regex literals, HTML block comments,
+ * and Rust/Swift nested block comments.
+ *
+ * @param {string[]} lines - Source lines to scan
+ * @param {string} ext - File extension used to select lexer rules
+ * @returns {boolean} True if the file ends with an unterminated token
+*/
 function analyzeComments(lines, ext = '') {
     const isPython = ext.toLowerCase() === '.py';
     const isHTML = ['.html', '.vue', '.svelte'].includes(ext.toLowerCase());
@@ -264,6 +299,7 @@ function analyzeComments(lines, ext = '') {
         while (j < line.length) {
             if (inBlockJS) {
                 if (line.slice(j, j + 2) === '/*') {
+                    // Rust and Swift allow nested /* */ block comments; track depth instead of a boolean [ds]
                     if (isRustOrSwift) blockDepthJS++;
                     j += 2;
                     continue;
@@ -282,6 +318,7 @@ function analyzeComments(lines, ext = '') {
                 continue;
             }
             if (inCppRawString) {
+                // C++ raw string terminator is )"delimiter" — delimiter must be matched exactly [ds]
                 if (line.slice(j, j + 2 + cppRawDelimiter.length) === ')' + cppRawDelimiter + '"') {
                     inCppRawString = false;
                     j += 2 + cppRawDelimiter.length;
@@ -336,6 +373,7 @@ function analyzeComments(lines, ext = '') {
                         break;
                     }
                 } else {
+                    // Shell and Ruby use '#' for line comments [ds]
                     const isShellOrRuby = ['.sh', '.rb'].includes(ext.toLowerCase());
                     if (isShellOrRuby) {
                         if (line[j] === '#') {
@@ -358,6 +396,7 @@ function analyzeComments(lines, ext = '') {
                             commentStartIndex = j;
                             break;
                         }
+                        // C++11 raw string literal: R"delim(...)delim" where delim is up to 16 non-special chars [ds]
                         if (isCpp && line[j] === 'R' && line[j+1] === '"') {
                             const match = line.slice(j).match(/^R"([^()\\\s]{0,16})\(/);
                             if (match) {
@@ -367,6 +406,7 @@ function analyzeComments(lines, ext = '') {
                                 continue;
                             }
                         }
+                        // Distinguish regex literal /.../  from division operator by inspecting the preceding non-whitespace token [ds]
                         if (isJS && line[j] === '/') {
                             let k = j - 1;
                             while (k >= 0 && /\s/.test(line[k])) k--;
@@ -384,6 +424,7 @@ function analyzeComments(lines, ext = '') {
                                     }
                                 }
                             }
+                            // Enter regex mode; the closing '/' will be matched during subsequent char scanning. [ds]
                             if (isRegex) {
                                 inRegex = true;
                                 j++;
@@ -395,6 +436,7 @@ function analyzeComments(lines, ext = '') {
             }
             if (isPython) {
                 if (!inTripleSingle && !inSingle && !inDouble) {
+                    // Triple-quoted Python strings can span lines; only enter/exit when not already inside another quote type to avoid premature termination. [ds]
                     if (line.slice(j, j + 3) === '"""') {
                         inTripleDouble = !inTripleDouble;
                         j += 3;
@@ -409,6 +451,7 @@ function analyzeComments(lines, ext = '') {
                     }
                 }
             }
+            // Attempts to detect escaped quote characters (backslash-balanced, ignoring double escaping). [ds]
             if (!inTripleSingle && !inTripleDouble) {
                 if (!isPython) {
                     if (!inSingle && !inDouble) {
@@ -452,12 +495,14 @@ function analyzeComments(lines, ext = '') {
             }
             j++;
         }
+        // C-family languages reset string/comment state at each newline (no line-spanning single/double quotes or regex). Python and JS template literals are excluded to allow multi-line constructs. [ds]
         const resetsAtLineEnd = ['.js', '.jsx', '.ts', '.tsx', '.java', '.c', '.cpp', '.cs', '.go', '.swift', '.kt', '.dart'].includes(ext.toLowerCase());
         if (resetsAtLineEnd) {
             inSingle = false;
             inDouble = false;
             inRegex = false;
         }
+        // A line is 'entirely inside a block' if we're at a block start with no trailing code, or fully wrapped by a block comment with no code before/after. [ds]
         const isEntirelyInsideBlock = isInsideBlockStart && (inBlockJS || inBlockHTML || (commentStartIndex === -1));
         let isPureComment = false;
         if (isEntirelyInsideBlock) {
@@ -479,11 +524,17 @@ function analyzeComments(lines, ext = '') {
     return analysis;
 }
 
-// Splices comments into the provided data [ds]
+/** [ds]
+ * Rewrites source by applying user-supplied comments.
+ * Handles 'default' (insert annotations), 'prune' (strip inline comments only),
+ * and 'clean' (remove auto-generated [ds] doc blocks).
+ * Preserves the original line-ending style (CRLF vs LF).
+*/
 function spliceComments(data, comments, mode = 'default', ext = '') {
     const hasCRLF = data.includes('\r\n');
     const lineEnding = hasCRLF ? '\r\n' : '\n';
     const originalLines = data.split(/\r?\n/);
+    // Process in descending line order so earlier insertions don't invalidate line numbers of subsequent edits. [ds]
     const sortedComments = [...comments].sort((a, b) => b.line - a.line);
     const validComments = sortedComments.filter(c => c.line >= 1 && c.line <= originalLines.length + 1);
 
@@ -491,9 +542,11 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
     let analysis = null;
     let dsBlocks = new Set();
 
+    // 'clean' and 'prune' modes both require lexical analysis to know which lines/regions consist solely of comments. [ds]
     if (mode === 'clean' || mode === 'prune') {
         analysis = analyzeComments(originalLines, ext);
         const finalDeletions = new Set();
+        // Pre-scan for [ds]-tagged comment blocks (including the line preceding the block) so entire generated docs can be removed atomically in 'clean' mode. [ds]
         if (mode === 'clean') {
             let i = 0;
             while (i < originalLines.length) {
@@ -524,6 +577,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
             const lineStr = originalLines[i];
             const lineAnalysis = analysis[i];
 
+            // Preserve shebang lines unconditionally across all modes. [ds]
             if (lineStr.trim().startsWith('#!')) {
                 continue;
             }
@@ -542,6 +596,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
                             newText = lineStr.slice(0, idx) + remainder.slice(endIdx + 2);
                         }
                     } else if (remainder.startsWith('<!--')) {
+                        // Handle closing of inline block comment: slice out '-->' terminator to preserve any trailing code after the comment [ds]
                         const endIdx = remainder.indexOf('-->');
                         if (endIdx !== -1) {
                             newText = lineStr.slice(0, idx) + remainder.slice(endIdx + 3);
@@ -550,10 +605,12 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
                     
                     annotated[i].text = newText.trimEnd();
                 }
+            // Clean mode removes comments only when tagged with [ds] marker (inline or inside a [ds] block), unlike prune which removes all comments [ds]
             } else if (mode === 'clean') {
                 const isDsBlockLine = dsBlocks.has(lineNum);
                 const hasDsInline = lineStr.includes('[ds]');
 
+                // Pure-comment lines are dropped entirely; partial (inline) comments require surgical extraction of the surrounding code [ds]
                 if (lineAnalysis.isPureComment) {
                     if (isDsBlockLine || hasDsInline) {
                         finalDeletions.add(lineNum);
@@ -564,6 +621,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
                         const remainder = lineStr.slice(idx);
                         let newText = lineStr.slice(0, idx).trimEnd();
                         
+                        // Block comment: reattach any content following the closing delimiter since it may be live code on the same line [ds]
                         if (remainder.startsWith('/*')) {
                             const endIdx = remainder.indexOf('*/');
                             if (endIdx !== -1) {
@@ -583,6 +641,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
         }
 
 
+        // Deletions are collected into a Set to dedupe (a line may be flagged by multiple passes), then sorted descending so splice offsets remain valid [ds]
         for (const c of validComments) {
             const lineIdx = c.line - 1;
             if (lineIdx >= 0 && lineIdx < originalLines.length) {
@@ -598,10 +657,12 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
             const trimmedLine = targetLine.trim();
 
             const lineAnalysis = analysis[lineNum - 1];
+            // Never strip shebang lines - removing them would break the executable script [ds]
             if (trimmedLine.startsWith('#!')) {
                 continue;
             }
 
+            // Defense-in-depth guard: before deleting, re-verify the line actually looks like a comment or blank. Prevents catastrophic data loss if upstream analysis is buggy [ds]
             const isCommentLine = 
                 lineAnalysis.isInsideBlock ||
                 lineAnalysis.isPureComment ||
@@ -622,8 +683,10 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
 
             annotated.splice(lineNum - 1, 1);
         }
+    // Insertion mode: for prune/annotate-style operations we splice new comment blocks into the source [ds]
     } else {
         for (const c of validComments) {
+            // Skip insertion if the target line offset falls inside a string literal, which would corrupt the string [ds]
             if (isLineInsideString(originalLines, c.line - 1, ext)) {
                 console.warn(`[devsplain] Skipping comment insertion at line ${c.line} to avoid string literal corruption.`);
                 continue;
@@ -637,6 +700,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
                 let trimmed = line.trimStart();
                 if (!trimmed) return '';
 
+                // Classify comment style to know where to append the [ds] marker: line comments get it at EOL, block comments get it before the closing delimiter [ds]
                 const isSingleLine = trimmed.startsWith('//') || trimmed.startsWith('#') || trimmed.startsWith('--');
                 const isBlockEnd = trimmed.endsWith('*/') || trimmed.endsWith('-->');
 
@@ -644,12 +708,14 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
                     trimmed = trimmed + ' [ds]';
                 } else if (idx === 0) {
                     if (isBlockEnd) {
+                        // Place [ds] marker just before the block-close delimiter so the comment remains syntactically valid [ds]
                         trimmed = trimmed.replace(/(\*\/|-->)$/, '[ds] $1');
                     } else {
                         trimmed = trimmed + ' [ds]';
                     }
                 }
 
+                // JSDoc-style continuation lines (' * ...') need an extra leading space to align the asterisk under the opening '/**' [ds]
                 if (trimmed.startsWith('*') && !trimmed.startsWith('*/') && !trimmed.startsWith('/*')) {
                     return indentation + ' ' + trimmed;
                 }
@@ -661,16 +727,18 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
         }
     }
 
+    // Split annotated stream back into original-indexed lines vs newly-inserted lines so we can validate both independently [ds]
     const filtered = annotated.filter(line => line.originalIndex !== -1);
     const filteredText = filtered.map(line => line.text);
     const filteredIndices = filtered.map(line => line.originalIndex);
 
-    // Validate that all inserted lines (originalIndex === -1) are valid comments or empty lines [ds]
+    // Validate that every inserted line is a genuine comment (or continuation of a block comment), never stray executable code [ds]
     const insertedLines = annotated.filter(line => line.originalIndex === -1);
     let inInsertedBlock = false;
     for (const item of insertedLines) {
         const trimmed = item.text.trim();
         if (!trimmed) continue;
+        // Track whether we are inside a multi-line block comment so inner lines are not individually required to start with a comment token [ds]
         if (inInsertedBlock) {
             if (trimmed.includes('*/') || trimmed.includes('-->')) {
                 inInsertedBlock = false;
@@ -687,22 +755,26 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
         if (!isValidComment) {
             throw new Error(`Safety Assertion Failed: Refused to insert non-comment code: "${trimmed}"`);
         }
+        // An opening block-comment delimiter without a matching close on the same line flips us into 'inside block' mode [ds]
         if ((trimmed.startsWith('/*') && !trimmed.includes('*/')) || (trimmed.startsWith('<!--') && !trimmed.includes('-->'))) {
             inInsertedBlock = true;
         }
     }
 
+    // Safety assertion: verify the spliced output is equivalent to the original minus the removed comments. Prevents silent corruption of source code. [ds]
     const textEqual = filteredText.every((text, idx) => {
         const origIdx = filteredIndices[idx];
         const originalLine = originalLines[origIdx];
         if (text === originalLine) {
             return true;
         }
+        // In clean/prune modes, allow text to differ from the original where a [ds] comment was deliberately stripped from an otherwise-live line [ds]
         if ((mode === 'clean' || mode === 'prune') && analysis) {
             const lineAnalysis = analysis[origIdx];
             if (lineAnalysis && lineAnalysis.commentStartIndex !== -1 && !lineAnalysis.isPureComment) {
                 const isDsBlockLine = dsBlocks.has(origIdx + 1);
                 const hasDsInline = originalLine.includes('[ds]');
+                // Reconstruct expected output after removing the tagged comment and compare - any drift indicates the splice damaged real code [ds]
                 if (mode === 'prune' || (mode === 'clean' && (hasDsInline || isDsBlockLine))) {
                     const idx = lineAnalysis.commentStartIndex;
                     const remainder = originalLine.slice(idx);
@@ -730,6 +802,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
         return false;
     });
 
+    // Ensure original indices remain strictly increasing; a duplicate or reorder would signal that splice offsets were miscomputed [ds]
     let indicesSequential = true;
     for (let i = 1; i < filteredIndices.length; i++) {
         if (filteredIndices[i] <= filteredIndices[i - 1]) {
@@ -738,6 +811,7 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
         }
     }
 
+    // Hard failure if either content drifted or ordering broke - better to abort than to write corrupted files [ds]
     if (!textEqual || !indicesSequential) {
         throw new Error("Safety Assertion Failed: Spliced code does not match original code minus comments!");
     }
@@ -745,7 +819,11 @@ function spliceComments(data, comments, mode = 'default', ext = '') {
     return annotated.map(line => line.text).join(lineEnding);
 }
 
-// Runs the CLI interface for the application [ds]
+/** [ds]
+ * Entry point for the CLI. Parses argv, merges CLI overrides with saved config, and dispatches to the appropriate
+ * command (help/version/config/hook install/remove) or the main comment-generation pipeline.
+ * Sets up a shared readline interface used by `askQuestion` for interactive prompts.
+*/
 async function runCLI() {
     rl = readline.createInterface({ input: process.stdin, output: process.stdout });
     askQuestion = (query) => new Promise((resolve) => rl.question(query, resolve));
@@ -766,10 +844,12 @@ Options:
   --force             Bypass the dirty Git tree safety check
   --clean             Scrub only devsplain-generated [ds] comments
   --prune             Destructively scrub ALL comments from files
-  --provider <name>   Override AI provider (gemini, groq, openai, custom)
+  --provider <name>   Override AI provider (gemini, groq, openai, claude, deepseek, custom)
   --model <name>      Override AI model name
   --api-key <key>     Override API key for the provider
   --base-url <url>    Override base URL for custom APIs
+  --concurrency <num> Override file concurrency level (1-5, default: 2)
+  --chunk-size <num>  Override chunk line size (e.g. 200, 600, 1000)
   --config            Force run the configuration setup wizard
   --setup-hook        Install Git pre-commit and post-commit hooks in repository
   --remove-hook       Remove devsplain-installed Git hooks from repository
@@ -808,6 +888,7 @@ Options:
         return;
     }
 
+    /** Reads the value immediately following a `--flag` in argv, or null if absent/trailing. [ds] */
     const getArgValue = (flag) => {
         const index = args.indexOf(flag);
         if (index !== -1 && index + 1 < args.length) {
@@ -816,8 +897,9 @@ Options:
         return null;
     };
 
+    // Identify the positional file path by skipping over flags and their argument values (flagKeys consume the next token). [ds]
     let filepath = '.';
-    const flagKeys = ['--provider', '--model', '--api-key', '--base-url', '--concurrency'];
+    const flagKeys = ['--provider', '--model', '--api-key', '--base-url', '--concurrency', '--chunk-size'];
     for (let i = 0; i < args.length; i++) {
         const arg = args[i];
         if (arg.startsWith('--')) {
@@ -846,6 +928,7 @@ Options:
     const hasOverwriteFlag = args.includes('--overwrite');
     const hasKeepFlag = args.includes('--keep');
 
+    // Guard against modifying an uncommitted working tree. Skipped in tests and during dry-runs since nothing is written. [ds]
     if (process.env.NODE_ENV !== 'test' && isGitDirty() && !isForce && !isDryRun) {
         console.error("Error: Git working tree is dirty. Please commit or stash your changes, or use --force to bypass this check.");
         rl.close();
@@ -859,13 +942,14 @@ Options:
     const cliApiKey = getArgValue('--api-key');
     const cliBaseUrl = getArgValue('--base-url');
 
+    // When the user overrides the provider via CLI, backfill sensible default model and base URL so partial overrides still work. [ds]
     if (cliProvider) {
         config.provider = cliProvider;
         if (!cliModel) {
-            config.model = cliProvider === 'gemini' ? 'gemini-2.0-flash' : (cliProvider === 'claude' ? 'claude-3-5-sonnet-20240620' : 'llama-3.3-70b-versatile');
+            config.model = cliProvider === 'gemini' ? 'gemini-2.0-flash' : (cliProvider === 'claude' ? 'claude-3-5-sonnet-20240620' : (cliProvider === 'deepseek' ? 'deepseek-chat' : (cliProvider === 'openai' ? 'gpt-4o' : 'llama-3.3-70b-versatile')));
         }
         if (!cliBaseUrl) {
-            config.baseUrl = cliProvider === 'gemini' ? null : (cliProvider === 'groq' ? 'https://api.groq.com/openai' : (cliProvider === 'openai' ? 'https://api.openai.com' : (cliProvider === 'claude' ? 'https://api.anthropic.com' : '')));
+            config.baseUrl = cliProvider === 'gemini' ? null : (cliProvider === 'groq' ? 'https://api.groq.com/openai' : (cliProvider === 'openai' ? 'https://api.openai.com' : (cliProvider === 'claude' ? 'https://api.anthropic.com' : (cliProvider === 'deepseek' ? 'https://api.deepseek.com' : ''))));
         }
     }
     if (cliModel) config.model = cliModel;
@@ -877,15 +961,25 @@ Options:
 
     const isOverwrite = (hasOverwriteFlag || config.autoPrune) && !hasKeepFlag;
 
-    // Parse --concurrency flag (default: 2, max: 5, min: 1) [ds]
+    // Clamp concurrency to the 1-5 range; default to 2 to avoid hammering provider rate limits. [ds]
     const cliConcurrency = parseInt(getArgValue('--concurrency'), 10);
     const concurrencyLevel = (cliConcurrency && cliConcurrency >= 1 && cliConcurrency <= 5) ? cliConcurrency : 2;
 
+    // Derive chunk threshold/overlap from the requested chunk size so large files split predictably without cutting mid-function. [ds]
+    const cliChunkSize = parseInt(getArgValue('--chunk-size'), 10);
+    if (cliChunkSize && cliChunkSize >= 50 && cliChunkSize <= 2000) {
+        config.chunkSize = cliChunkSize;
+        config.chunkThreshold = Math.round(cliChunkSize * 1.25);
+        config.chunkOverlap = Math.max(10, Math.round(cliChunkSize * 0.1));
+    }
+
+    // Load user-defined ignore patterns from `.devsplainignore`; silently ignore read/parse errors since the file is optional. [ds]
     let userIgnorePatterns = [];
     try {
         const ignorePath = path.join(process.cwd(), '.devsplainignore');
         if (fs.existsSync(ignorePath)) {
             const ignoreContent = fs.readFileSync(ignorePath, 'utf8');
+            // Strip trailing whitespace, drop blank lines and `#` comments to mirror .gitignore semantics. [ds]
             userIgnorePatterns = ignoreContent.split(/\r?\n/)
                 .map(line => line.trim())
                 .filter(line => line && !line.startsWith('#'));
@@ -901,10 +995,17 @@ Options:
     ];
     const allIgnored = [...defaultIgnoredFolders, ...userIgnorePatterns];
 
+    /** [ds]
+     * Returns true if `targetPath` matches any ignored folder name, glob suffix, or is nested under an ignored directory.
+     * Compares both the basename and the cwd-relative path so patterns work whether matched at any depth or rooted.
+    */
     function isPathIgnored(targetPath) {
         const filename = path.basename(targetPath);
+        // Normalize Windows backslashes to forward slashes for consistent pattern matching across platforms. [ds]
         const relPath = path.relative(process.cwd(), targetPath).replace(/\\/g, '/');
+        // Trim trailing path separators so `node_modules/` also matches the bare directory name. [ds]
         for (const pattern of allIgnored) {
+            // Normalize trailing separators so "foo/" and "foo" compare equal [ds]
             const cleanPattern = pattern.replace(/\/$/, '').replace(/\\$/, '');
             if (filename === cleanPattern || relPath === cleanPattern || relPath.startsWith(cleanPattern + '/')) return true;
             if (pattern.startsWith('*') && filename.endsWith(pattern.slice(1))) return true;
@@ -918,7 +1019,11 @@ Options:
         '.swift', '.kt', '.dart', '.sh', '.sql'
     ];
 
-    // Separate file discovery from processing for concurrency support [ds]
+    /** [ds]
+     * Recursively walks the filesystem from targetPath, returning a flat list of
+     * source files whose extensions are in validExtensions and that are non-empty.
+     * Empty files are skipped because there is nothing to comment on.
+    */
     function collectFiles(targetPath) {
         const collected = [];
         const stats = fs.statSync(targetPath);
@@ -932,6 +1037,7 @@ Options:
                 collected.push(...collectFiles(path.join(targetPath, item)));
             }
         } else if (stats.isFile()) {
+            // Extensions are compared case-insensitively (e.g. .JS matches .js) [ds]
             const ext = path.extname(targetPath).toLowerCase();
             if (!validExtensions.includes(ext)) return collected;
 
@@ -945,6 +1051,16 @@ Options:
         return collected;
     }
 
+    /** [ds]
+     * Runs the full pipeline on a single file: extract existing comments, generate
+     * new comments via the AI, splice them back into the source, and persist atomically.
+     *
+     * In overwrite mode, an initial "prune" pass strips prior comments before they
+     * are regenerated, otherwise previously inserted comments would be duplicated.
+     *
+     * Writes are atomic (temp file + rename) to avoid corrupting the source if the
+     * process is interrupted mid-write.
+    */
     async function processSingleFile(targetPath) {
         const filename = path.basename(targetPath);
         const ext = path.extname(targetPath).toLowerCase();
@@ -954,7 +1070,9 @@ Options:
         try {
             let comments = [];
             let commentedCode;
+            // 'clean'/'prune' modes skip comment extraction since the output is comment-stripped [ds]
             if (mode !== 'clean' && mode !== 'prune') {
+                // Overwrite needs prune (not clean) so existing non-generated comments are preserved as context [ds]
                 const preProcessMode = isOverwrite ? 'prune' : 'clean';
                 const cleanData = spliceComments(data, [], preProcessMode, ext);
                 comments = await getComments(cleanData, filename, config, mode);
@@ -988,10 +1106,9 @@ Options:
         }
     }
 
-    // Collect all eligible files, then process with adaptive concurrency [ds]
     const filesToProcess = collectFiles(filepath);
 
-    // Dry-run mode processes files serially to allow interactive prompts [ds]
+    // Dry-run and clean operations must stay sequential because they prompt the user per file [ds]
     if (isDryRun || mode === 'clean' || mode === 'prune') {
         for (const file of filesToProcess) {
             await processSingleFile(file);