Self-improving documentation system with agent evolution, metrics-driven self-modification, and stateless command protocol. Brainstorm, plan, orchestrate, review, reflect, evolve, complete.
npx claudepluginhub donovan-yohan/chalk-bag --plugin harnessUse when executing a living plan via Claude Code's /batch tool for worktree-isolated parallel execution, or when user says "batch", "batch execute", "run batch", "parallel batch", "run the plan in parallel", "execute without checkpoints", "hands-off execution", or "use batch mode"
Use when starting new feature work, creative design, or when user says "brainstorm", "design a feature", or "let's think about"
Use when investigating a bug, diagnosing an error, or when user says "debug", "fix bug", "investigate issue", or "root cause"
Use when ready to archive the plan and create the PR, when user says "we're done", "complete this", "wrap up", or after /harness:reflect finishes
Use after reflect to classify learnings, update metrics, and propose agent evolution, or when user says "evolve", "self-improve", "classify learnings"
Use when initializing structured documentation for a repository, when CLAUDE.md exceeds 120 lines, or when user says "set up docs" or "initialize harness"
Use when executing a living plan with agent teams, or when user says "orchestrate", "execute the plan", "start building", or "run the plan"
Use when creating an implementation plan from a design doc, bug analysis, or refactor scope, or when user says "create a plan", "plan this", or "write the plan"
Use when auditing docs for staleness, broken links, or bloat. Also use when user says "docs feel stale", "prune docs", or when CLAUDE.md exceeds 120 lines.
Use when resuming an in-progress refactoring, checking goal progress, or when user says "refactor status", "where are we", or "continue refactoring"
Use when planning an incremental refactoring, extracting responsibilities from a large class, or when user says "refactor", "extract", "strangler fig", or "decompose"
Use when capturing learnings and updating docs after review, when user says "reflect", "retrospective", "update docs", or after /harness:review completes
Use when implementation is done and code needs quality review, when user says "review the code", "check the code", or after /harness:orchestrate completes all tasks
Use when proposing modifications to agent definitions based on review escapes, metric anomalies, or universal learnings — invoked by /harness:evolve Phase 3
Use when auditing documentation health, finding stale or orphaned guides, checking CLAUDE.md bloat, or when /harness:prune is invoked
Use during /harness:review Phase 4 to check code changes against active learnings from docs/LEARNINGS.md for violations
Reusable agent skills for structured software development workflows. Two plugins:
A 3-tier documentation system with living execution plans, adversarial code review, conversation mining, and self-improving agent evolution.
Workflow: brainstorm -> plan -> orchestrate -> review -> reflect -> evolve -> complete
| Command | Purpose |
|---|---|
/harness:init | Initialize 3-tier documentation structure |
/harness:brainstorm | Design through collaborative dialogue |
/harness:bug | Systematic bug investigation with architecture review |
/harness:refactor | Scope incremental refactoring with strangler fig patterns |
/harness:plan | Create living execution plans from design docs |
/harness:orchestrate | Execute plans with agent teams and micro-reflects |
/harness:batch | Execute plans via worktree-isolated parallel batch |
/harness:review | Multi-agent code review with adversarial production review |
/harness:evolve | Classify learnings, update metrics, propose agent evolution |
/harness:reflect | Full reflection, conversation mining, retrospective |
/harness:complete | Archive plan, prune check, and create PR |
/harness:prune | Audit docs for staleness, broken links, bloat |
Agents: harness-pruner, learnings-reviewer, harness-evolver
Skills: strangler-fig (incremental refactoring patterns)
Pull request lifecycle management with multi-perspective automated review.
| Command | Purpose |
|---|---|
/pr:author | Create PRs with quality gates |
/pr:automate | Full automated lifecycle: author -> review -> resolve -> merge |
/pr:review | Multi-agent PR review (6 specialized agents) |
/pr:resolve | Analyze and address PR review comments |
/pr:update | Sync PR description with current changes |
Install as a marketplace. From any project directory:
claude /plugins add https://github.com/donovan-yohan/chalk-bag
This registers both plugins. You can also install individual plugins:
# Install just the harness plugin
claude /plugins add https://github.com/donovan-yohan/chalk-bag/plugins/harness
# Install just the pr plugin
claude /plugins add https://github.com/donovan-yohan/chalk-bag/plugins/pr
Hermes can load these skills through its plugin and profile system. See docs/hermes.md for integration instructions.
The command and skill files are standard markdown with YAML frontmatter. Any agent that can load markdown-based instructions can use these skills directly. The key integration points:
plugins/*/commands/*.md) — Procedural workflows triggered by explicit user commandsplugins/*/agents/*.md) — System prompts for specialized background agentsplugins/*/skills/*/SKILL.md) — Pattern-triggered capabilities with frontmatter descriptionsplugins/harness/scripts/*.sh) — Shell scripts for persistence and metricsplugins/harness/references/*.md) — Shared reference documents loaded by commandsThe harness plugin works standalone. Some commands optionally integrate with:
The pr plugin requires pr-review-toolkit for the review and automate commands.
The harness plugin manages a 3-tier documentation system:
The harness plugin tracks review effectiveness across sessions:
claude -pMIT
Complete developer workflow toolkit. Includes 34 reference skills, 34 specialized agents, and 21 slash commands covering TDD, debugging, code review, architecture, documentation, refactoring, security, testing, git workflows, API design, performance, UI/UX design, plugin development, and incident response. Full SDLC coverage with MCP integrations.
Uses power tools
Uses Bash, Write, or Edit tools
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
AI-powered wiki generator for code repositories. Generates comprehensive, Mermaid-rich documentation with dark-mode VitePress sites, onboarding guides, deep research, and source citations. Inspired by OpenDeepWiki and deepwiki-open.
Claude + Obsidian knowledge companion. Sets up a persistent, compounding wiki vault. Covers memory management, session notetaking, knowledge organization, and agent context across projects. Based on Andrej Karpathy's LLM Wiki pattern. Optional DragonScale Memory extension adds hierarchical log folds, deterministic page addresses, embedding-based semantic tiling lint, and boundary-first autoresearch topic selection.
Comprehensive C4 architecture documentation workflow with bottom-up code analysis, component synthesis, container mapping, and context diagram generation
Build and maintain an LLM-curated personal knowledge base in your project — Andrej Karpathy's LLM Wiki pattern, designed to scale to thousands of pages without becoming a context bottleneck. Now with an optional compiled graph layer for typed, provenance-backed relationships.