orchestrate

Orchestrates multi-step coding work through a git-ignored markdown store and specialist agents.

Updated May 5, 2026
One-click install
npx skills add https://github.com/josippapez/ai-setup --skill orchestrate-josippapez
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: orchestrate
Source: https://github.com/josippapez/ai-setup/tree/main/claude/plugins/orchestrate/skills/orchestrate
Command: npx skills add https://github.com/josippapez/ai-setup --skill orchestrate-josippapez

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Non-trivial coding work spanning multiple files or steps loses track of decisions, dependencies, and progress across sessions. This Skill decomposes work into tracked chunks, persists state in a local .orchestration/ markdown store, and coordinates specialist agents so an epic can be paused, resumed, and verified without an external tracker. ## Core Features & Use Cases - Phased state machine: Runs Gate, Intake, Explore, Refine, Design, Decompose, Plan, Execute, Relay, and Converge phases with explicit transition checkpoints and prohibited transitions. - Durable markdown store: Persists PROJECT.md, EPIC.md, and per-issue specs under .orchestration/ with append-only comments, status ownership rules, and resume-from-frontmatter semantics. - Conditional specialist routing: Dispatches repo scouts, planners, reviewers, standards checkers, and WCAG reviewers only when routing predicates match, with per-chunk model selection (haiku through opus) driven by a complexity field. - Use Case: Ask the agent to "track this" refactor across a dozen files; it scouts the repo, decomposes the work into waved issues, dispatches workers with review gates, and resumes exactly where it left off in a later session. ## Quick Start Ask the agent to orchestrate and track the multi-file task you describe, for example by saying "orchestrate this migration and break it down into tracked chunks".

Frequently Asked Questions about orchestrate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I track a multi-step coding task across sessions?

Ask the agent to orchestrate or track the work. It creates a `.orchestration/` store at the repo root with PROJECT.md, EPIC.md, and per-issue files, so any later session can resume from frontmatter status without an external tracker.

How does the orchestrator decide which specialist agents to run?

A routing matrix in references/routing.md defines exact predicates per specialist, such as non-empty test surfaces or documented standards. Each row is recorded as dispatched or skipped with a reason, and empty inputs always mean skip rather than a synthetic pass.

Does the orchestration store get committed to git?

No. The store lives in `.orchestration/` with its own `.gitignore` containing `*`, and the repository root `.gitignore` is never modified. The store is never committed, and no branch, commit, or PR happens without explicit user approval.

What happens when a worker agent keeps failing a chunk?

Workers get at most two fix-and-review rounds. At the retry cap the issue is labeled blocked, the fix list is persisted, and the orchestrator returns to the user with concrete resolution options instead of grinding or guessing.

When should I not use orchestration for a task?

Skip it for trivial one-off edits handled inline, and for requests where the outcome or approach itself is undecided. Undecided work goes through the brainstorm skill first; orchestration executes an already-decided plan.