Skip to content

benjamcalvin/bootstraps

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

77 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Bootstraps

A Claude Code plugin marketplace of reusable skills, hooks, and project scaffolds.

Prerequisites

This is a private repository. You need access to benjamcalvin/bootstraps on GitHub and git credentials configured (e.g. gh auth login).

For automatic plugin updates at Claude Code startup, set a GitHub token in your environment:

export GITHUB_TOKEN=ghp_your_token_here

Without this, plugins still work — you just need to update manually with /plugin marketplace update bootstraps.

Install

Add the marketplace to Claude Code:

/plugin marketplace add benjamcalvin/bootstraps

Browse Plugins

List available plugins:

/plugin marketplace list

Or open the interactive plugin manager:

/plugin

Navigate to the Discover tab to browse plugins from this marketplace.

Install a Plugin

/plugin install bootstrap-docs@bootstraps
/plugin install implement-lifecycle@bootstraps
/plugin install implement-cli@bootstraps
/plugin install implement-team@bootstraps
/plugin install issue-management@bootstraps
/plugin install bootstrap-worktrees@bootstraps
/plugin install stop-guard@bootstraps

Choose a scope when prompted:

  • user (default) — available in all your projects
  • project — available to anyone working on the current project
  • local — only for you in the current project

Use a Plugin

Invoke a plugin's skill as a slash command:

/bootstrap-docs
/implement #42

Some skills accept arguments:

/bootstrap-docs adr
/implement #42
/implement fix the login bug
/implement 17 just review
/implement-cli #42
/implement-team #42
/bootstrap-worktrees
/draft-issue add user avatar support
/cleanup-issue #42
/refine-issue #42

Update Plugins

/plugin marketplace update bootstraps

Uninstall a Plugin

/plugin uninstall bootstrap-docs@bootstraps
/plugin uninstall implement-lifecycle@bootstraps
/plugin uninstall implement-cli@bootstraps
/plugin uninstall implement-team@bootstraps
/plugin uninstall issue-management@bootstraps
/plugin uninstall bootstrap-worktrees@bootstraps
/plugin uninstall stop-guard@bootstraps

Available Plugins

Plugin Description
bootstrap-docs Set up a comprehensive, AI-readable documentation strategy in any project. Creates AGENTS.md, specs, ADRs, guides, plans, standards, and research templates.
bootstrap-worktrees Set up project-agnostic worktree isolation with per-worktree ports, Docker Compose projects, and config files. Discovers services and generates create/remove scripts plus Claude Code hooks.
implement-lifecycle Full implementation lifecycle with adversarial PR review — plan, implement, PR, review/address loop, docs gate, verify, merge.
implement-cli CLI-based variant of the implementation lifecycle using the Python Agent SDK to orchestrate review/address subprocesses with native async parallelism.
implement-team Experimental. Implementation lifecycle re-architected around Claude Code agent-teams — long-lived implementer and reviewer teammates with shared task list and mailbox messaging.
issue-management Draft, clean up, and refine GitHub issues — optimized for AI agent consumption.
stop-guard Stop hook that evaluates task completion via Gemini CLI and blocks premature stops. Opt-in per session via activation marker.

implement-lifecycle

Provides 6 skills and 5 reviewer agents for the complete implementation lifecycle:

Skills:

Skill Description
/implement Lean orchestrator — 6-phase lifecycle (plan → implement → PR → review loop → verify → merge). Accepts #issue, PR number, or freeform task. Supports trailing instructions like "just review" or "skip planning".
/merge-pr Validate, squash-merge, delete branch, and update linked GitHub issues with delivery status.
/pr-check Pre-flight PR validation — branch naming, title, description, sizing, commits, references.

Subagent skills (invoked by the orchestrator, not directly):

Skill Description
implement-code Explore codebase, plan, write tests first, implement, self-review, commit, and create PR.
implement-address Address filtered review findings from the referee's action plan.
verify End-to-end verification — exercises the real running system, checks downstream effects, regression tests existing flows.

Reviewer agents (invoked in parallel during the review loop):

Agent Focus
review-correctness Logic bugs, edge cases, error handling, race conditions
review-security AuthZ, injection risks, PII handling, spec conformance
review-architecture Pattern consistency, module boundaries, coupling, forward-looking design
review-testing Test coverage, assertion quality, edge cases, test anti-patterns
review-docs Docs compliance gate (Phase 4.5) — missing docs for new behavior, stale docs contradicted by code changes, frontmatter/cross-link correctness

implement-cli

Single-skill plugin that mirrors the /implement lifecycle but delegates heavy work to Python Agent SDK subprocesses.

Skill Description
/implement-cli Same 6-phase lifecycle as /implement (plan → implement → PR → review loop → docs gate → verify → merge), but the orchestrator runs review/address phases as claude-agent-sdk subprocesses with native async parallelism. Accepts the same argument shapes and trailing instructions as /implement.

implement-team

Experimental. Requires Claude Code >= 2.1.32 and CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. Re-architects the implementation lifecycle around long-lived teammates that share a task list and mailbox instead of forked one-shot subagents.

Skill Description
/implement-team Full implementation lifecycle run by an orchestrator ("lead") that coordinates persistent team-implementer, team-verifier, and reviewer teammates. Lead is the sole GitHub publisher; reviewers post findings to the shared task list only. Trades higher token cost for fewer speculative findings and cross-reviewer dedupe.
Agent Role
team-implementer Long-lived implementer. Acts only on lead direction; refuses reviewer work requests and answers batched factual questions only.
team-verifier Long-lived verifier. Runs end-to-end verification on lead invocation and reports PASS/FAIL/PARTIAL/N/A evidence.
team-reviewer-correctness Correctness specialist — same focus as review-correctness, with persistent-team messaging discipline.
team-reviewer-security Security / requirements-conformance specialist.
team-reviewer-architecture Architecture and design specialist.
team-reviewer-testing Test-quality specialist.
team-reviewer-docs Docs compliance specialist (invoked only in the Phase 4.5 gate).

bootstrap-worktrees

Single-skill plugin for project-agnostic worktree isolation.

Skill Description
/bootstrap-worktrees Discover services in a project, assign per-worktree port offsets, generate scripts/create-worktree.sh and scripts/remove-worktree.sh, wire up Docker Compose project isolation, and register Claude Code hooks so new sessions land in isolated environments.

issue-management

Provides 3 skills for GitHub issue quality:

Skill Description
/draft-issue Create well-structured GitHub issues with testable acceptance criteria, optimized for /implement consumption.
/cleanup-issue Fix formatting, fill missing sections, and clarify ambiguity in existing issues.
/refine-issue Deepen an issue with codebase research — sharpen acceptance criteria, add implementation hints, decompose into sub-tasks.

stop-guard

A Stop hook — no skills to invoke. Once installed, it activates when any session transcript contains <!-- stop-guard:active -->.

Requires: Gemini CLI (npm install -g @google/gemini-cli), jq

How it works: When Claude tries to stop, the hook calls Gemini to independently evaluate whether the task is complete. If incomplete, it blocks the stop and tells Claude what to finish. If complete (or if human input is needed), it allows the stop.

Activation: Include <!-- stop-guard:active --> in a skill's output or type it directly in conversation. Optionally provide context:

<!-- stop-guard:context {"task": "implement feature X", "criteria": ["tests pass"]} -->

Safety: Opt-in per session, max 3 continuations (configurable), fail-open on any error, 60s timeout.

Configuration (optional): ~/.config/stop-guard/config.json

{"max_continuations": 3, "model": "gemini-3-flash-preview"}

Testing: Run the evaluator against any session without triggering the hook:

./plugins/stop-guard/hooks/test-stop-guard.sh ~/.claude/projects/<project>/<session>.jsonl

See stop-guard/README.md for full documentation.

License

MIT

About

A curated collection reusable patterns, agentic skills, and hooks that work for the problems I solve.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors