Skip to content

shrwnsan/base

v1.12.1Apache-2.0

Essential Claude Code workflow and context engineering tools for productive development

GitHub

Changelog

All notable changes to the Base Plugin will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

  • branch-hygiene skill: squash-merge-aware branch and worktree cleanup classifier
    • Detects content-level merges that git branch --merged cannot see (subject/PR matching, file presence, feature diffs)
    • Verdicts: merged-ancestry, merged-content, remote-gone, stale-base, superseded, unmerged, active, worktree-prunable/clean-merged/dirty, orphan-dir
    • Classifier script is strictly read-only; tiered action policy (auto local delete / confirm worktree+remote removal / never touch dirty or unmerged)

[1.12.1] - 2026-11-03

Changed

  • Skills sync to the shrwnsan/agents hub is now gated on a fail-closed portability allowlist (.github/sync-skills.json; no entry = no sync). base opts in — its four skills keep syncing as before.

1.11.0 - 2026-02-11

Added

  • Test output filter: added npx/bunx support (vitest, jest, playwright, cypress, mocha, ava, tape)
  • Supports compound commands (e.g., npm run build && npx vitest)
  • Covers common one-off test runs, CI/CD workflows, and tool evaluation patterns
  • Added 11 new test cases for npx/bunx pattern validation

1.10.1 - 2026-02-10

Changed

  • Test output filter: removed ^ anchor from patterns to match npm test -- <args>, compound commands like npm test && build
  • Test output filter: added negative grep to exclude application logs (ISO timestamps, time-only stamps)
  • Test output filter: removed redundant subshell wrapper and outer redirection for cleaner command syntax

1.10.0 - 2026-02-06

Added

  • PreToolUse hook for automatic test output filtering (npm, pytest, cargo, go, yarn, pnpm, bun, rspec, rails test, mvn test, gradle test)
  • Reduces token costs by showing only errors, failures, and key test results
  • Enabled by default, disable with VIBEKIT_BASE_TEST_FILTER=none or VIBEKIT_BASE_TEST_FILTER=disable
  • Graceful degradation when jq is not installed
  • Comprehensive documentation in docs/test-output-filtering.md

1.9.0 - 2026-02-05

Skill Frontmatter Improvements

Added

  • user-invocable: true to all skills for reliable slash command invocation
  • disable-model-invocation: true to handoff-context to prevent accidental activation
  • Dynamic context injection (! syntax) to systematic-debugging and crafting-commits skills
    • systematic-debugging: git log, recent test logs, build output
    • crafting-commits: git status, recent commits, branch information
  • Expanded allowed-tools lists with explicit tool declarations for clarity

Changed

  • Skills now support reliable /skill-name slash command invocation
  • systematic-debugging skill pre-loads error context for faster debugging
  • crafting-commits skill pre-loads git history for style consistency
  • handoff-context no longer auto-activates during normal conversation

1.8.0 - 2026-02-04

handoff-context skill

P1/P2 improvements from eval-021 analysis:

Added

  • Confidence/completeness scoring (0.3-0.95 scale) with missing context detection
  • Enhanced session tracking (unique session IDs, timestamps, duration)
  • Learnings/debugging section for tracking patterns and techniques
  • Quick start section (package manager detection, project type detection)
  • Approaches tracking (what worked, what didn't, what's left to try)
  • Iterative validation with validate-context.sh script
  • YAML-based configuration system with multi-location support
    • ~/.config/agents/ (cross-tool standard)
    • ~/.claude/ (Claude Code specific)
    • .agents/ (project-local)
  • Monorepo support with multi-type project detection

Changed

  • Updated templates with new YAML sections
  • Updated documentation (eval-021, tasks-003, SKILL.md, templates.md)
  • Improved package manager detection (npm, pnpm, yarn, bun with fallback chain)
  • macOS-compatible random generation using openssl rand

1.7.7 - 2026-02-03

handoff-context skill

Added

  • /handoff-context slash command as the most reliable invocation method
  • "Let's handoff" and "Lets handoff" trigger patterns
  • Confidence scoring system (0.3-0.95 scale) for quality evaluation
  • Validation script (validate-context.sh) for checking completeness
  • HANDOFF_FILE output variable for easy file path capture

Changed

  • Restructured to follow agentskills.io standard
  • Improved workflow to populate template in-place (single file)
  • Tightened bash tool permissions for security
  • Enhanced documentation with progressive disclosure structure

Fixed

  • Script execution issue with imperative "you must:" language
  • Cross-directory compatibility with find ~/.claude/plugins
  • Double file creation issue
  • Missing trigger patterns causing skill to be ignored

1.7.0 - 2026-02-03

handoff-context skill

Added

  • Progressive disclosure implementation with capture-context.sh
  • Automated git state capture (branch, staged/unstaged files)
  • Structured YAML context generation with unique timestamps
  • Error handling for non-git repos and script failures
  • Quick reference examples (examples-quick.md)
  • Evaluation test suite for validation scenarios
  • Screenshots demonstrating usage

Changed

  • Refactored documentation with progressive disclosure
  • Simplified SKILL.md with quick start section
  • Moved detailed workflow to references/workflow.md

CI workflow

Added

  • docs-bypass.yml for auto-merging documentation-only changes

1.6.0 - 2026-01-26

handoff-context skill

Added

  • Natural language thread continuation
  • Detects handoff triggers: "handoff and", "handoff to", "start a new thread with this"
  • Captures git state, conversation summary, active work, next steps
  • Writes to /tmp/ with private temp directory for security
  • Agent-agnostic output compatible with agentskills.io standard
  • Complements Claude Code Tasks system for clean slate continuation

1.5.0 - 2026-01-11

systematic-debugging skill

Added

  • Structured debugging approach with 5-step workflow: capture context, reproduce, isolate, fix, verify
  • Guardrails against common debugging pitfalls
  • Multi-component system debugging guidance
  • Architecture questioning when 3+ fixes fail
  • Auto-activates on error language and debugging context

/review-arch command

Added

  • Comprehensive architecture reviews
  • Analyzes components, data flows, and risks
  • Provides prioritized improvement opportunities
  • Supports optional focus areas (backend, frontend, auth, performance)
  • Delivers actionable insights with concrete next steps

1.3.0 - 2025-12-08

/commit command

Added

  • Flexible CLI modes: -f/--fast for quick commits, -v/--verbose for detailed analysis
  • Auto-detection based on change complexity (default behavior)
  • Support for both short (-f, -v) and long (--fast, --verbose) flags
  • Progressive disclosure design - simple by default, detailed when needed
  • Enhanced argument-hint showing all available options

Changed

  • Improved command description to include flag hints
  • Better UX with user choice over verbosity level based on context
  • Updated terminology from "complex" to "detailed" for positive framing

Documentation

Added

  • Comprehensive evaluation document (eval-014) comparing with Anthropic's official commit-commands plugin
  • Reframed positioning to emphasize "adaptive productivity" for VibeKit's brand

[1.2.0] - Previous

Added

  • Base plugin workflow orchestration capabilities
  • Crafting-commits skill for intelligent commit message creation
  • Essential Claude Code workflow tools