
from superclaude-plusplus13
An always-active orchestration agent that analyzes requests, auto-delegates to sub-agents, and runs PDCA-style project workflows across MCP tools.
PM is a Project Manager agent designed to act as the default orchestration layer in an assistant stack. It restores session context, analyzes user requests, selects strategies (brainstorm, direct, wave), auto-delegates tasks to specialist sub-agents, and documents progress using PDCA patterns.
Use PM for multi-domain projects, vague feature requests that need discovery, and complex workflows that require orchestration across planners, implementers, testers, and writers. It is intended as a default layer active at session start or when explicit PM invocation is used.
Fits developer and productivity-focused agent frameworks that can orchestrate sub-agents and load MCP tools dynamically (e.g., Claude Code, multi-agent orchestrators, MCP-aware frameworks).
Orchestration skill for SuperClaude framework that manages sub-agents via PDCA cycles and MCP tool coordination. Pure instruction-based with no scripts — nothing to execute. Very verbose SKILL.md (mixed Japanese/English) with redundant sections, overly prescriptive error investigation protocols, and no separation of concerns. Heavily tied to specific MCP tools (Serena, context7, sequentialthinking) making it non-portable.
Monolithic SKILL.md with no scripts/ or references/ directories. Memory key schema and PDCA templates could be extracted to references/. No destructive or malicious content — purely orchestration instructions. The skill's value depends heavily on the SuperClaude ecosystem being present.
Project Audit Rules
Run project-specific verification rules that check business logic, architecture patterns, and naming conventions beyond linters.
Session Save / Context Persistence
Persist session context and create checkpoints: save learnings, summaries, and project state for cross-session continuity.