v0.7.0

What's New

Added

• scv index subcommand — rebuild ~/.scv/index.md, a summary index of all analyzed repositories
  • AI reads each project's README.md directly to write a one-line summary + functional keywords (up to 5)
  • Generated index includes a directory-structure header and a project table (Project / Summary / Responsibilities)
  • New reference docs: references/index.md (both en and zh-cn)
• Language-aware index generation
  • scv_util.py update-index reads lang from ~/.scv/config.json and generates a fully localized index.md
  • All-Chinese output for zh-cn, all-English for en
• install.sh persists the --lang setting into ~/.scv/config.json at install time
• config.example.json — added "lang" field (default: "en")

Changed

• scv_util.py update-index — replaced Python-based README parsing with an AI-driven workflow
  • New --entries-json parameter accepts a JSON array of {name, summary, keywords} objects provided by the AI
• run.md Step 7.5 — AI reads the new README, writes summary and keywords itself, merges with existing index, then calls the script to write the file
• batchRun.md Step 6.5 — delegates to references/index.md workflow
• SKILL.md (en + zh-cn) — added index subcommand to description, command list, and routing table

v0.6.1

[v0.6.1] - 2026-03-27

Added

• Multi-language installer support

  • install.sh now accepts --lang parameter (en or zh-cn)
  • Default language is English (en)
  • Usage: ./install.sh --lang=zh-cn or ./install.sh --lang zh-cn

• New config options in config.example.json

  • batch_size: Control parallel analysis batch size (default: 5)
  • deep_analysis_enabled: Toggle deep analysis mode (default: false)

v0.6.0

[v0.6.0] - 2026-03-25

Added

• Deep Analysis with codebones Integration

  • New progressive deep dive strategy: skeleton overview → targeted codebones get for full implementations
  • codebones get <symbol> - Fetch complete source code of specific classes/methods on-demand
  • codebones search <query> - Search symbols and trace dependencies across the codebase
  • codebones outline <path> - Get file/directory structure without reading full content
  • 85% token reduction compared to reading all files, while still getting full implementation details

• Enhanced project-analyzer Agent

  • Added Bash tool to execute codebones CLI commands during deep analysis
  • New Phase 2.5 deep analysis workflow with on-demand code fetching
  • Framework-specific extraction patterns (Java/Spring, Python/FastAPI, Go, Node.js/NestJS)
  • Enhanced output sections: Service dependencies, API method chains, business rules from actual code

Changed

• Deep Analysis Workflow Refactored
  • Previously: Only generated skeleton file (85% token savings) but stopped there
  • Now: Skeleton overview → identify key symbols → codebones get for actual implementations
  • Deep analysis is now truly "deep" - fetches complete implementations when needed

Dependencies

• codebones (optional): Required for deep analysis mode
  • Install: pip install codebones or cargo install codebones
  • GitHub: https://github.com/creynir/codebones

v0.5.0

[v0.5.0] - 2026-03-17

Added

• Incremental analysis (skip unchanged repos)

  • batchRun and run now detect the HEAD commit of each repository before analysis
  • Repos whose HEAD commit hasn't changed since the last analysis are skipped automatically
  • Per-repo .scv_metadata.json stores the last-analyzed commit hash and timestamp
  • batchRun plan step marks unchanged repos as skipped, complete step writes metadata on success
  • run.md updated with commit detection (Step 4), metadata writing (Step 6), and updated step numbering

• scv_util.py — new CLI utility script

  • Provides three CLI commands: get-commit-info, check-skip, write-metadata
  • check-skip exits with code 2 when a repo can be skipped (no changes since last run)
  • write-metadata writes .scv_metadata.json to the analysis output directory
  • Also importable as a Python module (used by batch_manager.py)

• git_op.py enhancements

  • Added get_head_commit(repo_path) to retrieve the current HEAD commit hash
  • Added get_commit_info(repo_path) to return commit hash + author + message as a dict

Changed

• batch_manager.py refactored

  • plan() now marks repos with unchanged commits as skipped before dispatching subagents
  • complete() writes .scv_metadata.json via scv_util after successful analysis
  • Imports migrated from removed metadata.py to scv_util.py

• project-analyzer agent updated (en + zh-cn)

  • Added Current Commit as an input parameter passed by the orchestrator
  • Clarified that the agent displays the commit in its output but does not write metadata itself

• batchRun.md updated (en + zh-cn)

  • Plan output example now includes skipped_repos and per-repo skipped flag
  • Step 4 documents the skip logic and exit-code handling
  • Step 5 summary includes skipped repo count

Removed

• metadata.py — logic fully consolidated into scv_util.py

v0.4.1

[v0.4.1] - 2026-03-09

Fixed

• batchRun concurrency control
  • Added explicit batch processing pseudocode with TaskOutput(block=true) to enforce waiting
  • Added strict execution requirements to prevent forking more than 5 subagents per turn
  • Added warning about why concurrency must be limited (resource consumption, API rate limiting)
  • Updated both English and Chinese versions of batchRun.md

v0.4.0

[v0.4.0] - 2026-03-13

Changed

• Migrated from Commands to Skill architecture

  • Converted scv.run, scv.batchRun, scv.gather commands into a unified scv skill
  • New subcommand syntax: /scv run, /scv batchRun, /scv gather
  • Updated install.sh to install skill instead of commands
  • Old commands are automatically removed during installation

• Language-specific skill structure

  • Separated English and Chinese skills into skills/en/ and skills/zh-cn/
  • Each language skill is self-contained with its own templates
  • Install script copies selected language skill as ~/.claude/skills/scv
  • Removed prompts/ and commands/ directories (now integrated into skills)

• Unified analysis engine with project-analyzer subagent

  • Both run and batchRun now use the same project-analyzer subagent
  • Consistent analysis quality across single and batch operations
  • Context isolation for all repository analyses

Added

• New agent definition with bilingual support: agents/{en,zh-cn}/project-analyzer.md

  • Specialized subagent with optimized tool set (Read, Write, Glob, Grep, LSP)
  • Single-purpose deep code analysis agent
  • Generates 4 standardized documents (README, SUMMARY, ARCHITECTURE, FILE_INDEX)
  • Language-specific agent installed based on --lang parameter

• Subagent-based parallel execution for batchRun

  • Each repository analysis runs in an independent subagent
  • Maximum 5 concurrent subagents with batched processing
  • TodoWrite-based task tracking for progress visibility
  • True parallel processing with isolated contexts
  • Prevents context bloat when analyzing multiple repositories

Architecture

• Unified flow:

  /scv run <path>     → Agent(project-analyzer) → 4 docs
  /scv batchRun       → Agent(project-analyzer) × N (max 5 parallel) → 4 docs × N
  

• Directory structure:

  SCV/
  ├── agents/
  │   ├── en/
  │   │   └── project-analyzer.md    # English agent definition
  │   └── zh-cn/
  │       └── project-analyzer.md    # Chinese agent definition
  ├── skills/
  │   ├── en/                        # English skill
  │   │   ├── SKILL.md
  │   │   └── references/
  │   │       ├── run.md             # Uses project-analyzer subagent
  │   │       ├── batchRun.md        # Uses project-analyzer subagent (max 5 parallel)
  │   │       ├── gather.md
  │   │       └── templates/
  │   └── zh-cn/                     # Chinese skill (same structure)
  └── install.sh                     # Installs agent + skill based on --lang
  

v0.3.0

[v0.3.0] - 2026-03-06

Added

• Project Analyzer Enhancements
  • Added "Recommended Tools" section with guidance on using Glob, Grep, Read, and LSP
  • Added "Template Syntax Reference" table explaining placeholder, loop, and conditional syntax
  • Added "Edge Cases Handling" section covering 7 common scenarios
  • Added "Output Checklist" for quality verification before completion
  • Added token optimization tips for large projects

Changed

• Project Analyzer Improvements
  • Enhanced "Execution Instructions" with concrete 3-step workflow
  • Moved output language setting to the beginning of the document
  • Expanded "Analysis Principles" with more actionable guidance
  • Improved overall structure and clarity for both English and Chinese versions

Fixed

• Installation Script
  • Fixed install.sh to only remove scv.* commands instead of entire ~/.claude/commands directory
  • Prevents accidental deletion of other user-installed Claude commands