By joestein
Spec-driven SDLC plugin for Claude Code — orchestrates specialist agents through a full development lifecycle with security-first engineering practices
Set up build tooling and CI/CD using the build agent
Pick up the next available step from the roadmap and execute it through the full agent pipeline
Generate documentation for the specified target using the documentation agent
Check GitHub CI status for a PR, fix regressions, alert on pre-existing failures
Clean up on-loop workspace, switch to main, and pull latest
Generates specifications, architecture decision records, and system design documents from user prompts
Sets up Makefile, GitHub Actions CI/CD, linter configuration, and security scanning
Implements features according to the spec and plan with security-first practices, handles remediation from test/security/review feedback
Generates READMEs, CLAUDE.md files, API docs, guides, and architecture diagrams
Conducts the full SDLC loop — manages phase transitions, quality gates, retry logic, and agent coordination
Manages on-loop phase state transitions, validation, and persistence
Defines pass/fail criteria for phase transitions in the on-loop pipeline
File-level locking with TTL, heartbeat, and stale lock detection for multi-session roadmap coordination
Manages roadmap state files — initialization, phase transitions, step tracking, and schema validation
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Spec-driven SDLC plugin for Claude Code that orchestrates specialist agents through a full development lifecycle with security-first engineering practices.
/on-loop takes a prompt and runs it through a complete software development pipeline:
Prompt → Worktree → Branch → Spec → Plan → Code → Test → Security → Docs + Build → Review → Commit + Push + PR → Done
Each phase is handled by a specialist agent operating as a Staff Engineer with ISC2 certifications, building for regulated financial environments and critical infrastructure.
Each session operates in its own git worktree, so multiple sessions can run concurrently on the same repo without interference. Session logs are persisted in the repo as an audit trail.
Add the on-loop marketplace, then install the plugin:
/plugin marketplace add joestein/on-loop
This registers the marketplace from the repo's marketplace.json. Then install the plugin:
/plugin install on-loop
That's it — all /on-loop commands are now available in your Claude Code sessions.
How it works: The
marketplace.jsonat the repo root declares available plugins. When you run/plugin marketplace add, Claude Code fetches this manifest and makes the listed plugins available for install./plugin installthen activates the plugin, loading its commands, agents, skills, and hooks.
git clone https://github.com/your-org/on-loop.git ~/.claude/plugins/on-loop
git clone https://github.com/your-org/on-loop.git ~/dev/on-loop
ln -s ~/dev/on-loop ~/.claude/plugins/on-loop
Clone or copy into your project and reference it in your project's Claude Code configuration.
| Command | Description |
|---|---|
/on-loop <prompt> | Run full SDLC loop with all agents |
/on-loop-check [PR number or branch] | Check GitHub CI status, fix regressions, alert on pre-existing failures |
/on-loop-debug-fix [description or image] | Debug and fix issues from infrastructure logs or user-provided context |
/on-loop-status | Check progress of current and past sessions |
/on-loop-resume [--from=phase] [--session=<id>] | Resume an interrupted loop |
/on-loop:clear [--include-logs] | Clean up worktrees, optionally remove session logs |
/on-loop:main-resolve | Pull main, merge into branch, resolve conflicts |
/on-spec <description> | Standalone spec generation |
/on-test <target> | Standalone test generation |
/on-security <target> | Standalone security audit |
/on-doc <target> | Standalone documentation generation |
/on-build <target> | Standalone build/CI setup |
/on-review <target> | Standalone code review |
| Command | Description |
|---|---|
/on-prepare <prompt> | Generate a roadmap with phases, steps, and acceptance criteria |
/on-plan [feature-slug] | Read roadmap, produce detailed implementation plan |
/on-continue [feature-slug] | Pick up next available step and execute through agent pipeline |
/on-pause [feature-slug] | Release locks, commit WIP, write handoff summary |
| Agent | Model | Role |
|---|---|---|
| Orchestrator | Opus | Pipeline control, quality gates, retry logic, worktree/session lifecycle |
| Architect | Opus | Spec generation, ADRs, system design |
| Coding | Opus | Implementation with security-first practices |
| Testing | Sonnet | Unit, integration, and E2E tests |
| Security | Opus | OWASP/STRIDE audit, compliance checks (read-only) |
| Documentation | Sonnet | READMEs, guides, CLAUDE.md files |
| Build | Sonnet | Makefile, GitHub Actions, lint/security configs |
| Reviewer | Opus | Final code review gate (read-only) |
graph TD
START["/on-loop prompt"] --> INIT["INIT: Session + Worktree + Branch"]
INIT --> SPEC["SPEC: Architect Agent (in worktree)"]
SPEC --> PLAN["PLAN: Orchestrator writes plan"]
PLAN --> CODE["CODE: Coding Agent (in worktree)"]
CODE --> TEST["TEST: Testing Agent (in worktree)"]
TEST -->|Pass| SEC["SECURITY: Security Agent (in worktree)"]
TEST -->|"Fail (max 3x)"| CODE
SEC -->|Pass| PAR["DOC + BUILD (parallel, in worktree)"]
SEC -->|"Blockers (max 2x)"| CODE
PAR --> REVIEW["REVIEW: Reviewer Agent (in worktree)"]
REVIEW -->|Approve| GIT["GIT: Commit + Push + PR (from worktree)"]
REVIEW -->|"Changes (max 2x)"| CODE
GIT --> DONE["COMPLETE: Summary + PR link + Worktree cleanup"]
Each session creates a git worktree at .claude/worktrees/<branch-slug>/, providing an independent working directory. This means:
npx claudepluginhub joestein/on-loop --plugin on-loopPersonal knowledge wiki system following Karpathy's LLM Wiki pattern. Research any domain, build structured wiki pages, ingest into pgvector, and chat with your knowledge base.
Scan, migrate, and validate codebases for post-quantum cryptographic security using NIST FIPS 203/204/205 standards (ML-KEM, ML-DSA, SLH-DSA).
loophaus — Control plane for coding agents
Universal CLI orchestrator with multi-runner support. Autonomous spec-driven development with dependency DAG, parallel worktree execution, two-stage review gates, and modular merge hardening.
SDLC enforcement for AI agents — TDD, planning, self-review, CI shepherd
Harness engineering for Claude Code — hook-enforced dual review, state-machine gates, and fail-closed safety where it counts.
SPEC-First development workflow with TDD, Ralph Loop, and autonomous agent coordination for Claude Code
Persona-driven AI development team: orchestrator, team agents, review agents, skills, slash commands, and advisory hooks for Claude Code