By metalspawn
Orchestrated agentic workflows for Claude Code — plan, execute, and submit with automated quality gates
Convention and structure reviewer. Checks code against project CLAUDE.md conventions. Reports pass/fail with specific issues. Does not fix code.
Critical plan reviewer. Reviews plans to find flaws, edge cases, and problems BEFORE execution.
Semantic reviewer. Checks that naming and comments communicate intent clearly. Binary PASS/FAIL. Does not fix code.
Quality assurance agent. Runs tests and type checks. Reports pass/fail with specific errors. Does not fix issues.
Focused implementation agent. Executes ONE specific task completely. Does not decide what to build - implements what it's told.
Create a feature branch from the current branch. Invoked directly or from /orc:tasks. No-op if already on a feature branch.
Break a feature into PR-sized, independently shippable steps. Use before /orc:plan when work spans multiple PRs — architectural changes, large features, or multi-PR rollouts.
Execute tasks through the worker, review, validate, and commit pipeline. Run after /orc:tasks, or directly with a task description for simple work.
Develop a structured implementation plan through codebase exploration, planning, and critical review. Use for non-trivial work: multi-file changes, architectural decisions, new features, or ambiguous requirements.
Create or update a pull request with project convention detection. Invoked from /orc:submit or directly.
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.
A Claude Code plugin that provides structured, multi-agent workflows for non-trivial code changes. Plan collaboratively, then execute autonomously with parallel quality gates.
Multi-PR feature:
/orc:decompose → approve → then per step ↓
Single PR:
/orc:plan → approve → /orc:tasks → confirm → /orc:execute → /orc:submit
│ │ │
creates branch worker → review push → PR → self-review
(if needed) gates → commit │
findings? → fix → /orc:submit
clean? → ready for human review
│
/orc:pull-comments
│
triage → fix → /orc:submit
Direct task:
/orc:execute <task> → review gates → commit → /orc:submit
Eight skills. Five specialised agents.
| Phase | Skill | What happens |
|---|---|---|
| Decompose | /orc:decompose | Break a feature into PR-sized steps — clarifies boundaries with you first |
| Plan | /orc:plan | Explore codebase → clarify ambiguity → produce plan → critical review → present for approval |
| Branch | /orc:branch | Create a feature branch (invoked by /orc:tasks or directly) |
| Tasks | /orc:tasks | Branch setup → create tasks with acceptance criteria from the plan |
| Execute | /orc:execute | For each task: worker implements → three parallel review gates → commit |
| Submit | /orc:submit | Push → create/update PR → self-review → triage findings |
| PR | /orc:pr | Create or update a PR with project convention detection (invoked by /orc:submit or directly) |
| Review | /orc:pull-comments | Fetch external PR comments → categorise → triage → route to next step |
For simple, direct tasks: /orc:execute <task> skips planning and creates a single task inline.
When work spans multiple PRs, start with /orc:decompose to break the feature into ordered, independently shippable steps. After approving the decomposition, run each step through the standard /orc:plan → /orc:tasks → /orc:execute → /orc:submit cycle.
| Agent | Role | Used by |
|---|---|---|
worker | Implements code for a specific task | /orc:execute |
plan-reviewer | Critiques plans for flaws before execution | /orc:plan |
code-reviewer | Checks code against project CLAUDE.md conventions | /orc:execute |
semantic-reviewer | Checks naming clarity and comment accuracy | /orc:execute |
validator | Runs tests and type checks | /orc:execute |
After the worker implements each task, three agents run in parallel:
worker (implement)
│
▼
┌──────────┬───────────┬───────────┐
│ code- │ semantic- │ validator │
│ reviewer │ reviewer │ │
└────┬─────┴─────┬─────┴─────┬─────┘
│ │ │
▼ ▼ ▼
PASS/FAIL PASS/FAIL PASS/FAIL
All PASS → commit
Any FAIL → worker fixes → all three re-run
They check orthogonal concerns — conventions, naming, and correctness — so all three must re-run after any fix.
Plans include a Verification Plan — concrete, product-level scenarios that prove the feature works end-to-end. These aren't restated unit tests; they're things a human (or integration test) can run against the system:
curl localhost:3000/api/health → 200 OK"/orc:tasks creates a final verification task (blocked by all implementation tasks) that the validator runs after everything is built. Manual-only scenarios are flagged without blocking.
Both /orc:decompose and /orc:plan pause after codebase exploration to check in with you when:
When everything is clear, the step is skipped automatically — no unnecessary interruptions.
Once all tasks pass their review gates and are committed, /orc:submit handles the PR lifecycle:
npx claudepluginhub metalspawn/claude-ops --plugin orcMulti-agent orchestration for code that matters.
Orchestration plugin. v1 use case: async development - turn ready issues into pull requests, then iterate on review feedback until a human takes over. Designed to host more orchestration use cases (refactoring, docs, audits) in future versions.
End-to-end development workflow: design → draft-plan → orchestrate → review → pr-create → pr-review → pr-merge
Long-running agent harness with 5-layer memory architecture, GitHub integration, autonomous batch processing, Agent Teams with ATDD, 9 hooks (safety, quality gates, team coordination), and 6 Agent Skills
Persona-driven AI development team: orchestrator, team agents, review agents, skills, slash commands, and advisory hooks for Claude Code
Autonomous GitHub Project Management workflow with orchestrator-agent coordination for parallel task execution via git worktrees