Claude Code plugin: turns a multi-repo microservice project into a coordinated AI agent team that proposes (never applies) cross-repo change plans
Command
Adversarial multi-repo root-cause mode: spawn all repo specialists as an Agent Team, force them to challenge each other's assumptions with evidence, and surface the true cross-repo root cause of an incident. Max 3 deliberation rounds with a tie-break rule.
Command
Open and guide editing of a repo's context file. Ingests frontmatter changes back into registry.json on completion.
Command
Build or refresh the Claude-native knowledge summary for one or all repos. Produces .repo-orchestrator/graphs/ /summary.json used by /triage to pre-populate specialist context, reducing token consumption. No external dependencies — runs entirely within the Claude session.
Command
Bootstrap: discover all repos in the workspace, index them, generate editable context docs and specialist agents (each with a unique name and per-repo skill file), and register them with the master. Pauses for user review before writing agents.
Command
First-time setup: a no-build, zero-friction guided install — scans prerequisites, enables optional tiers, auto-wires settings (Agent Teams + MCP), then bootstraps. Works with no toolchain; the compiled runner is an optional accelerator, never required.
Command
Show at-a-glance status of the repo-orchestrator workspace: registered repos, last index timestamps, graph availability, userEdited flags, and Agent Teams enablement.
Command
Incremental refresh: detect repo drift via fingerprint, re-index changed repos, preserve userEdited content, ingest context frontmatter back into registry.json.
Command
Master controller: route a ticket to responsible repo specialists, have them deliberate, and return a single consolidated change plan with confidence aggregate, current-state baseline, ordered steps, and rollback guidance. Propose-only — no files are modified.