Pepper Orchestrator for Claude Code

One coordinator for a multi-repository program

Pepper Orchestrator 0.11.0 is a free, MIT-licensed preview for Claude Code. It turns a program spanning several repositories, or parallel areas of one repository, into explicit work packages, dependencies, locks, evidence and decisions.

The coordinator plans and records program state. Independent module sessions implement the work in their own repositories or worktrees. Review and verification compare the actual PR, commit and test evidence before a task is accepted. The owner queue keeps decisions and privileged actions visible.

Included

• Program initialization, planning, dispatch, review, verification and recovery instructions.
• Shared Python runtime, templates, state management, atomic IDs and file locks.
• Native Claude plugin, 13 short commands and three scoped review/research agents.
• English and Russian methodology resources.

Install and start

Extract the ZIP into ~/.claude/skills/, preserving the complete top-level folder. Restart the agent and ask: “Use Pepper Orchestrator. Initialize a Claude program for these repositories, then prepare a plan.” Recent Claude Code versions load the bundled .claude-plugin automatically from the personal skills directory. The plugin may alternatively be loaded with --plugin-dir.

The owner launches module sessions using the prepared commands. Read the bundled runtime contract before operating a program. Install one copy per host; avoid simultaneously installing this ZIP and the same skill from a marketplace.

Validation and scope

Package/frontmatter checks passed. The shared packaged runtime passed 11 local regression tests covering IDs, locks, recovery, project instructions and Codex worktrees. The Claude manifest passed native validation and the CLI detected 14 skill/command components and three agents. These are package and runtime checks, not a recorded end-to-end multi-repository release.

This is a 0.11 preview. It does not provision accounts, authorize MCP services or grant production access. Irreversible actions require the owner’s explicit decisions and the workflow’s gates. Free download; model usage and optional third-party tools may have their own costs.

Permissions and compact packaging

Agensi revision 2 includes the full native commands, agents, runtime and safety gates in fewer than 50 files. English/Russian/settings template texts are consolidated into a readable JSON dictionary and mode instructions into one Markdown reference. Development test runners, fixtures and source history remain upstream.

Environment Variables are required for named runtime configuration and ordinary child-process environment. Optional network uses github.com and api.github.com. Other repository origins and live verification/MCP endpoints require explicit owner configuration and authorization. Owner-configured checks may execute shell commands. Installing this package grants no permission to merge, deploy or access production. Read the root SKILL.md for exact access and previously flagged detector expressions; no private keys are bundled.