v3.7.0

What's Changed

New skills

• ticket-grooming — deep-dive ticket investigation via parallel sub-agents (codebase explorer, domain analyst, staff engineer reviewer). Auto-sets story points, @mentions PM on questions, fills severity for security tickets (#11)
• grill-me — interview the user relentlessly about a plan or design until ready for implementation (#13)
• write-a-prd — create a PRD through user interview, codebase exploration, and structured output (#13)
• improve-codebase-architecture — explore a codebase to find opportunities for architectural improvement (#13)
• future-code-search — tiered model routing for codebase search: delegates search/exploration to Haiku/Sonnet sub-agents, keeps Opus as orchestrator (#21)

Review reliability improvements

• Verify-the-Mechanism rule — sub-agents must now cite exact file:line where a failure is signaled, how it propagates, and why the claimed symptom actually occurs before flagging CRITICAL/IMPORTANT findings. Prevents false-positive findings based on pattern-matching without verification (#16)
• Self-Critique Pass — mandatory counterargument step for every CRITICAL/IMPORTANT finding before submission (#16)
• Exact-Name Citation — class, method, and file names in findings must match the diff verbatim; no substitution from memory (#16)
• Precomputed Failure Semantics Context — orchestrator gathers framework failure behaviour (e.g., Interactor gem context.fail! vs uncaught exceptions) once and forwards to both sub-agents (#16)
• Staff Engineer review step added to ticket-grooming skill (#12)
• Author attribution added to skill frontmatter; IMPORTANT findings must now be fixed before merge (#10)

Architecture

• Consolidated 5 review sub-agents into 2 (code-quality-reviewer + security-reviewer) for token efficiency — reduces context overhead while maintaining full coverage (#15)
• Fixed agent frontmatter and corrected subagent_type references across skills (#14)

Full Changelog

v3.6.0...v3.7.0

v3.6.0

What's Changed

Lifecycle improvements

• Consolidated separate Code Review and SQL Review phases into a single Phase 7 using the comprehensive-code-review skill, which dispatches 5 parallel sub-agents (code quality, pattern consistency, SQL, security, simplification)
• Added explicit Phase 8 (FIX FINDINGS) to address review output before committing
• CRITICAL findings: hard block on commit; IMPORTANT findings: must fix or explicitly defer with tracking issue + user approval; MINOR: track, don't block
• Phase count: 11 → 11 (PUSH and VERIFY CI preserved)

Skills added (in recent releases)

• comprehensive-code-review — orchestrated parallel review across all dimensions
• pull-request-description — structured PR format with summary, background, test plan, and rollback plan

Agents added

• security-reviewer — OWASP Top 10 (2021) aligned security audit agent

v3.5.0

Changelog

• c059bcc feat: add comprehensive-code-review skill (#8)
• bb5e948 feat: add pull-request-description skill (#6)
• f1dadf0 fix: address security and reliability review findings (#7)

v3.4.0

Changelog

• a77c22c feat: add Claude Desktop as installation target (#4)
• 41bcff7 feat: add codebase-memory-mcp installation support (#5)
• aadc635 feat: add sql-optimization-patterns skill and sql-reviewer agent (#2)
• 73bb916 feat: store project init state in .claude/project.json (#3)

v3.3.0

Changelog

• 98ed9b2 feat: project detection heuristics for CLAUDE.md generation (#1)

v3.2.0

Changelog

• 30de954 feat: add --mode flag and constants for install modes
• 22fd451 feat: add Mode field to config struct
• 4424555 feat: implement askInstallMode interactive question
• 4409c1d feat: implement runAgentsOnly for agents-only mode
• fd55d4f feat: implement runConfigOnly for config-file-only mode
• aa9f52a refactor: dispatch runInstall by mode, extract runFullInstall

v3.1.0

What's Changed

README Rewrite

• Complete restructure with table of contents and logical section ordering
• Dedicated sections for GitHub Workflow, Skills & Agents, CLI Installer, and Installation
• Added workflow lifecycle overview, review modes, issue tracking, and git worktrees documentation
• Moved Attribution to the end; added License section

CLAUDE-BASE Template Sync

Synced the base template with the latest learning-journey CLAUDE.md patterns:

• Compact format: Replaced verbose ASCII diagrams and long prose with a concise phase table
• Autonomous PR review: Added auto-merge workflow with safety limits (max 3 review cycles, max 3 CI fixes)
• Unaddressed work tracking: Any deferred finding must become a GitHub issue — never silently drop work
• Actionable issues: Issues must include specific file paths and line numbers
• Streamlined sections: Consolidated duplicate content (Pre-Push Workflow merged into lifecycle, Manual/Autonomous review condensed)

Init Command Updates

• Expanded permissions setup from gh-only to full autonomous agent workflow
• New permissions cover: git, node, npm, npx, and common shell utilities (sed, cat, cp, echo, head, tail, diff, grep, mkdir, pwd, for, ls)
• Permissions are written to .claude/settings.local.json (gitignored, per-developer)

v3.0.0

Changelog

• f4a34b9 ci: add test workflow for PRs and main branch
• 4eecaf2 feat: add GitHub workflow integration with slash commands
• 3b7a03e feat: add code-simplifier and code-review enforcement to CLAUDE-BASE template
• 9c43f42 feat: add staff engineer plan review enforcement to CLAUDE-BASE template

v2.0.0

Changelog

• bb06860 Initial commit: futuregerald-claude-plugin
• c74da15 ci: add GoReleaser for automated binary releases
• fb4ab1b feat: add tags to all skill frontmatter for filtering
• 33e2988 fix: normalize SKILL.md casing for all skills
• 368cf2c refactor: rewrite installer for directory-based skills with fs.FS interface
• 2238cee refactor: rewrite runInstall with scope prompt, config generation, command installation
• 210704f refactor: simplify config struct to match new flag structure
• f9213c3 refactor: update targets, embed directives, remove packs, add scope prompt