tsdoc/ts/aidocs_classes/commit.ts

import * as plugins from '../plugins.js';
import { AiDoc } from '../classes.aidoc.js';
import { ProjectContext } from './projectcontext.js';
import { DiffProcessor } from '../classes.diffprocessor.js';

export interface INextCommitObject {
  recommendedNextVersionLevel: 'fix' | 'feat' | 'BREAKING CHANGE'; // the recommended next version level of the project
  recommendedNextVersionScope: string; // the recommended scope name of the next version, like "core" or "cli", or specific class names.
  recommendedNextVersionMessage: string; // the commit message. Don't put fix() feat() or BREAKING CHANGE in the message. Please just the message itself.
  recommendedNextVersionDetails: string[]; // detailed bullet points for the changelog
  recommendedNextVersion: string; // the recommended next version of the project, x.x.x
  changelog?: string; // the changelog for the next version
}

export class Commit {
  private aiDocsRef: AiDoc;
  private projectDir: string;

  constructor(aiDocsRef: AiDoc, projectDirArg: string) {
    this.aiDocsRef = aiDocsRef;
    this.projectDir = projectDirArg;
  }

  public async buildNextCommitObject(): Promise<INextCommitObject> {
    const smartgitInstance = new plugins.smartgit.Smartgit();
    await smartgitInstance.init();
    const gitRepo = await plugins.smartgit.GitRepo.fromOpeningRepoDir(
      smartgitInstance,
      this.projectDir
    );

    // Define comprehensive exclusion patterns
    // smartgit@3.3.0+ supports glob patterns natively
    const excludePatterns = [
      // Lock files
      'pnpm-lock.yaml',
      'package-lock.json',
      'npm-shrinkwrap.json',
      'yarn.lock',
      'deno.lock',
      'bun.lockb',

      // Build artifacts (main culprit for large diffs!)
      'dist/**',
      'dist_*/**',           // dist_ts, dist_web, etc.
      'build/**',
      '.next/**',
      'out/**',
      'public/dist/**',

      // Compiled/bundled files
      '**/*.js.map',
      '**/*.d.ts.map',
      '**/*.min.js',
      '**/*.bundle.js',
      '**/*.chunk.js',

      // IDE/Editor directories
      '.claude/**',
      '.cursor/**',
      '.vscode/**',
      '.idea/**',
      '**/*.swp',
      '**/*.swo',

      // Logs and caches
      '.nogit/**',
      '**/*.log',
      '.cache/**',
      '.rpt2_cache/**',
      'coverage/**',
      '.nyc_output/**',
    ];

    // Pass glob patterns directly to smartgit - it handles matching internally
    const diffStringArray = await gitRepo.getUncommittedDiff(excludePatterns);

    // Process diffs intelligently using DiffProcessor
    let processedDiffString: string;

    if (diffStringArray.length > 0) {
      // Diagnostic logging for raw diff statistics
      const totalChars = diffStringArray.join('\n\n').length;
      const estimatedTokens = Math.ceil(totalChars / 4);

      console.log(`📊 Raw git diff statistics:`);
      console.log(`   Files changed: ${diffStringArray.length}`);
      console.log(`   Total characters: ${totalChars.toLocaleString()}`);
      console.log(`   Estimated tokens: ${estimatedTokens.toLocaleString()}`);
      console.log(`   Exclusion patterns: ${excludePatterns.length}`);

      // Use DiffProcessor to intelligently handle large diffs
      const diffProcessor = new DiffProcessor({
        maxDiffTokens: 100000,      // Reserve 100k tokens for diffs
        smallFileLines: 300,         // Most source files are under 300 lines
        mediumFileLines: 800,        // Only very large files get head/tail treatment
        sampleHeadLines: 75,         // When sampling, show more context
        sampleTailLines: 75,         // When sampling, show more context
      });

      const processedDiff = diffProcessor.processDiffs(diffStringArray);
      processedDiffString = diffProcessor.formatForContext(processedDiff);

      console.log(`📝 Processed diff statistics:`);
      console.log(`   Full diffs: ${processedDiff.fullDiffs.length} files`);
      console.log(`   Summarized: ${processedDiff.summarizedDiffs.length} files`);
      console.log(`   Metadata only: ${processedDiff.metadataOnly.length} files`);
      console.log(`   Final tokens: ${processedDiff.totalTokens.toLocaleString()}`);

      if (estimatedTokens > 50000) {
        console.log(`✅ DiffProcessor reduced token usage: ${estimatedTokens.toLocaleString()} → ${processedDiff.totalTokens.toLocaleString()}`);
      }
    } else {
      processedDiffString = 'No changes.';
    }

    // Use DualAgentOrchestrator for commit message generation
    // Note: No filesystem tool needed - the diff already contains all change information
    const commitOrchestrator = new plugins.smartagent.DualAgentOrchestrator({
      smartAiInstance: this.aiDocsRef.smartAiInstance,
      defaultProvider: 'openai',
      guardianPolicyPrompt: `
You validate commit messages for semantic versioning compliance.

APPROVE if:
- Version level (fix/feat/BREAKING CHANGE) matches the scope of changes in the diff
- Commit message is clear, professional, and follows conventional commit conventions
- No personal information, licensing details, or AI mentions (Claude/Codex) included
- JSON structure is valid with all required fields
- Scope accurately reflects the changed modules/files

REJECT with specific feedback if:
- Version level doesn't match the scope of changes (e.g., "feat" for a typo fix should be "fix")
- Message is vague, unprofessional, or contains sensitive information
- JSON is malformed or missing required fields
`,
    });

    await commitOrchestrator.start();

    const commitTaskPrompt = `
You create a commit message for a git commit.
Project directory: ${this.projectDir}

Analyze the git diff below to understand what changed and generate a commit message.

You should not include any licensing information or personal information.
Never mention CLAUDE code, or codex.

Important: Answer only in valid JSON.

Your answer should be parseable with JSON.parse() without modifying anything.

Here is the structure of the JSON you should return:

interface {
  recommendedNextVersionLevel: 'fix' | 'feat' | 'BREAKING CHANGE'; // the recommended next version level
  recommendedNextVersionScope: string; // scope name like "core", "cli", or specific class names
  recommendedNextVersionMessage: string; // the commit message (don't include fix/feat prefix)
  recommendedNextVersionDetails: string[]; // detailed bullet points for the changelog
  recommendedNextVersion: string; // the recommended next version x.x.x
}

For recommendedNextVersionDetails, only add entries that have obvious value to the reader.

Here is the git diff showing what changed:

${processedDiffString}

Generate the commit message based on these changes.
`;

    const commitResult = await commitOrchestrator.run(commitTaskPrompt);
    await commitOrchestrator.stop();

    if (!commitResult.success) {
      throw new Error(`Commit message generation failed: ${commitResult.status}`);
    }

    const resultObject: INextCommitObject = JSON.parse(
      commitResult.result.replace('```json', '').replace('```', '')
    );

    const previousChangelogPath = plugins.path.join(this.projectDir, 'changelog.md');
    let previousChangelog: plugins.smartfile.SmartFile;
    if (await plugins.fsInstance.file(previousChangelogPath).exists()) {
      previousChangelog = await plugins.smartfileFactory.fromFilePath(previousChangelogPath);
    }

    if (!previousChangelog) {
      // lets build the changelog based on that
      const commitMessages = await gitRepo.getAllCommitMessages();
      console.log(JSON.stringify(commitMessages, null, 2));

      // Use DualAgentOrchestrator for changelog generation with Guardian validation
      const changelogOrchestrator = new plugins.smartagent.DualAgentOrchestrator({
        smartAiInstance: this.aiDocsRef.smartAiInstance,
        defaultProvider: 'openai',
        guardianPolicyPrompt: `
You validate changelog generation.

APPROVE if:
- Changelog follows proper markdown format with ## headers for each version
- Entries are chronologically ordered (newest first)
- Version ranges for trivial commits are properly summarized
- No duplicate or empty entries
- Format matches: ## yyyy-mm-dd - x.x.x - scope

REJECT with feedback if:
- Markdown formatting is incorrect
- Entries are not meaningful or helpful
- Dates or versions are malformed
`,
      });

      await changelogOrchestrator.start();

      const changelogTaskPrompt = `
You are building a changelog.md file for the project.
Omit commits and versions that lack relevant changes, but make sure to mention them as a range with a summarizing message instead.

A changelog entry should look like this:

    ## yyyy-mm-dd - x.x.x - scope here
    main descriptiom here

    - detailed bullet points follow

You are given:
* the commit messages of the project

Only return the changelog file content, so it can be written directly to changelog.md.

Here are the commit messages:

${JSON.stringify(commitMessages, null, 2)}
`;

      const changelogResult = await changelogOrchestrator.run(changelogTaskPrompt);
      await changelogOrchestrator.stop();

      if (!changelogResult.success) {
        throw new Error(`Changelog generation failed: ${changelogResult.status}`);
      }

      previousChangelog = plugins.smartfileFactory.fromString(
        previousChangelogPath,
        changelogResult.result.replaceAll('```markdown', '').replaceAll('```', ''),
        'utf8'
      );
    }

    let oldChangelog = previousChangelog.contents.toString().replace('# Changelog\n\n', '');
    if (oldChangelog.startsWith('\n')) {
      oldChangelog = oldChangelog.replace('\n', '');
    }
    let newDateString = new plugins.smarttime.ExtendedDate().exportToHyphedSortableDate();
    let newChangelog = `# Changelog\n\n${`## ${newDateString} - {{nextVersion}} - {{nextVersionScope}}
{{nextVersionMessage}}

{{nextVersionDetails}}`}\n\n${oldChangelog}`;
    resultObject.changelog = newChangelog;

    return resultObject;
  }
}