v3.2.0 — Advanced Hooks, LLM Gates & 7 New Skills

What's New in v3.2.0

LLM-Powered Hooks (First in any Claude Code plugin)

• Prompt hooks (type: "prompt") — AI-powered quality gates using Haiku
• Commit message validation — Conventional commits format enforced by LLM
• Secret detection — File writes scanned for hardcoded API keys and tokens
• if conditionals — Precise hook filtering without spawning processes

7 New Skills (17 → 24)

Skill	What It Does
permission-tuner	Analyze denial patterns, generate optimized allow/deny rules
compact-guard	Protect context through compaction (5-file restore limit, 50K budget)
cost-tracker	Session cost awareness with budget benchmarks by task type
auto-setup	Auto-detect project type, configure quality gates (Node/Python/Rust/Go)
mcp-audit	Analyze MCP server token overhead, recommend servers to disable
llm-gate	Configure AI-powered quality gates using prompt and agent hooks
file-watcher	Reactive workflows using FileChanged/CwdChanged with env injection

2 New Agents (6 → 8)

• Permission Analyst — Analyzes denial patterns, generates rules by risk tier
• Cost Analyst — Token usage analysis, cache optimization, model selection

6 New Hook Events (18 → 24)

PermissionDenied, Setup, WorktreeCreate, WorktreeRemove, CwdChanged, TaskCreated

Agent Token Optimization

• omitClaudeMd: true on 6 read-only agents — skips project context loading to save tokens
• Applies to: planner, reviewer, scout, context-engineer, permission-analyst, cost-analyst

5 New Commands (16 → 21)

/permission-tuner, /compact-guard, /cost-tracker, /auto-setup, /mcp-audit

Hook Improvements

• Consolidated git commit/push hooks into single matcher with if conditions
• Scoped FileChanged to config files only (.env|package.json|tsconfig.json|Cargo.toml|go.mod)
• Added cargo test to test failure detection
• All scripts validated for JSON shape safety and error handling

Full Counts

Component	v3.1	v3.2
Skills	17	24
Agents	6	8
Hook Events	18	24
Commands	16	21
Scripts	23	29

Install

# Claude Code Plugin
/plugin marketplace add rohitg00/pro-workflow

# SkillKit
npx skillkit install pro-workflow

Full Changelog: v3.1.0...v3.2.0

pro-workflow v3.1.0

What's Changed

• chore: bump version to 3.1.0
• Add decision framework, safety modes, sprint tracking (#33)

Installation

Claude Plugin

claude plugin marketplace add rohitg00/pro-workflow
claude plugin install pro-workflow@pro-workflow

npm Package

npm install pro-workflow

Build SQLite Support

cd ~/.claude/plugins/*/pro-workflow
npm install && npm run build

pro-workflow v3.0.0

What's Changed

• Fix CI: TypeScript 6.0 compat, sync versions to 3.0.0
• deps: bump typescript from 5.9.3 to 6.0.2
• deps: bump @types/node from 25.3.0 to 25.5.0
• deps: bump better-sqlite3 from 12.6.2 to 12.8.0
• fix: use GITHUB_OUTPUT for release changelog
• v3.0: Context engineering, agent teams, extracted hooks (#32)
• Add Mintlify docs link to README and infographic (#26)
• fix: rebrand visuals to Anthropic colors and update infographic to v2.0 (#24)

Installation

Claude Plugin

claude plugin marketplace add rohitg00/pro-workflow
claude plugin install pro-workflow@pro-workflow

npm Package

npm install pro-workflow

Build SQLite Support

cd ~/.claude/plugins/*/pro-workflow
npm install && npm run build

pro-workflow v2.0.0

What's Changed

$(cat /tmp/changelog.txt)

Installation

Claude Plugin

claude plugin marketplace add rohitg00/pro-workflow
claude plugin install pro-workflow@pro-workflow

npm Package

npm install pro-workflow

Build SQLite Support

cd ~/.claude/plugins/*/pro-workflow
npm install && npm run build

pro-workflow v1.3.0

7 improvements leveraging new Claude Code 2.1.49 features, with full Cursor compatibility.

What's New

Native Worktree Flag

claude -w replaces manual git worktree add for instant parallel sessions. Updated /parallel command and parallel-worktrees skill. Cursor users get the manual git worktree approach side by side.

Background Scout Agent

Scout now runs with background: true and isolation: worktree — confidence scoring happens without blocking your main session.

ConfigChange Hook

New hook type detects when quality gates or hooks are modified mid-session. Ships config-watcher.js with 100KB log rotation.

Smarter Stop Hook

Uses last_assistant_message (2.1.49+) to detect task completion signals and large changes instead of blind response counting. Truncates to last 2000 chars for performance.

Default Plugin Settings

Ships settings.json with quality gate permissions (lint, test, typecheck, git worktree) so users don't get prompted for common commands.

Updated Model Preferences

• Opus 4.6 and Sonnet 4.6 both support adaptive thinking and 1M-token context (beta via context-1m-2025-08-07 header; default remains 200K)
• Sonnet 4.5 (200K) retired from Max plan
• Removed opus-thinking from config (adaptive thinking is built-in)

Updated /learn Command

New keybindings (Ctrl+F kill agents), Simple mode includes Edit tool, Shift+Down only for teammate navigation, ConfigChange hook docs, Plugins section.

Cursor Compatibility

• Platform-specific sections marked with (Claude Code) labels
• Commands table includes Cursor skill equivalents
• 7 of 9 skills are fully platform-agnostic
• background: true and isolation: worktree in agent frontmatter are silently ignored by Cursor
• .mdc rules provide equivalent enforcement to Claude Code hooks

Installation

Claude Code Plugin

claude plugin marketplace add rohitg00/pro-workflow
claude plugin install pro-workflow@pro-workflow

Cursor Plugin

/add-plugin pro-workflow

npm

npm install pro-workflow

Build SQLite Support

cd ~/.claude/plugins/*/pro-workflow
npm install && npm run build

Files Changed

• 14 files changed, 383 additions, 114 deletions
• 2 new files: .claude-plugin/settings.json, scripts/config-watcher.js

Full Changelog: v1.2.0...v1.3.0

pro-workflow v1.2.0

What's New in v1.2.0

New Features

• Scout Agent — Confidence-gated exploration. Scores readiness (0-100) across 5 dimensions before implementation. If confidence < 70, automatically gathers more context. Integrates with learnings DB to reduce confidence for historically problematic patterns.

• /replay — Surface past learnings for current task. Uses FTS5 BM25 ranking to find relevant learnings + checks session history for correction rates on similar work. Like muscle memory for coding sessions.

• /handoff — Structured session handoff documents. Generates a copy-paste-ready briefing (status, done, pending, decisions, gotchas, resume command) for seamless continuation in the next session.

• Drift Detection — New UserPromptSubmit hook that tracks original task intent and warns after 6+ edits with low relevance to the original goal. Recognizes explicit task switches.

• Adaptive Quality Gates — Quality gate thresholds now adjust based on your correction history from the DB. High correction rate (>25%) = gates at 3 edits. Low rate (<5%) = gates at 10 edits. Gets smarter over time.

• Correction Heatmap — New /insights heatmap option showing correction distribution by category and project, adaptive gate status, hot learnings (frequently corrected, rarely captured), and cold learnings (captured but never applied).

Fixes

• Added YAML frontmatter to planner and reviewer agents (fixes #7)
• Added .gitattributes enforcing LF line endings for Windows compatibility
• Synced all version files to 1.2.0

Commits

• feat: add scout agent, /replay, /handoff, drift detection, adaptive gates, correction heatmap
• fix: add YAML frontmatter to planner and reviewer agents
• fix: bump version to 1.2.0 for next release

Installation

Claude Plugin

claude plugin marketplace add rohitg00/pro-workflow
claude plugin install pro-workflow@pro-workflow

npm Package

npm install pro-workflow

Build SQLite Support

cd ~/.claude/plugins/*/pro-workflow
npm install && npm run build

v1.1.0 - Smart Commit, Insights & Agent Teams

What's New

New Commands

• /commit - Smart commit with pre-commit quality gates (lint, typecheck, tests), staged change review, conventional commit message drafting, and post-commit learning capture
• /insights - Session analytics, learning database analytics (top categories, most applied, stale learnings), correction pattern analysis, and productivity metrics

New Features

• Agent Teams (Experimental) - Coordinate multiple Claude Code sessions with shared task lists, inter-agent messaging, delegate mode, and split panes (tmux/iTerm2)
• Custom Subagents - Create project or user-level subagents with custom tools, permission modes, persistent memory, hooks, and skill preloading
• Adaptive Thinking - Opus 4.6 calibrates reasoning depth per task automatically
• Context Compaction - Auto-compaction at ~95% capacity with configurable threshold and PreCompact hooks

Updated

• Model selection updated: Haiku 4.5, Sonnet 4.5, Opus 4.6
• /learn command expanded with agent teams, subagents, adaptive thinking, compaction, and new CLI shortcuts
• SKILL.md updated across all 8 patterns
• 8 total commands: /wrap-up, /learn, /learn-rule, /parallel, /search, /list, /commit, /insights

Install

Plugin:

/plugin marketplace add rohitg00/pro-workflow
/plugin install pro-workflow@pro-workflow

SkillKit:

npx skillkit install pro-workflow

pro-workflow v0.1.0

What's Changed

$(cat /tmp/changelog.txt)

Installation

Claude Plugin

claude plugin marketplace add rohitg00/pro-workflow
claude plugin install pro-workflow@pro-workflow

npm Package

npm install pro-workflow

Build SQLite Support

cd ~/.claude/plugins/*/pro-workflow
npm install && npm run build