A Claude Code plugin for knowledge work — spec-driven phased execution with compounding institutional knowledge
RDEL-Group/compound-knowledge-work
Agent
Knowledge extraction specialist. Dual-mode — solution extraction (capture what worked) and full retrospective (project close-out). Produces reusable solutions for the knowledge repo.
RDEL-Group/compound-knowledge-work
Agent
Content drafting specialist. Produces deliverable content from structured task plans in a fresh context — one task per invocation.
RDEL-Group/compound-knowledge-work
Agent
Task planning specialist. Breaks knowledge work phases into atomic, executable tasks with context budgets and dependency ordering.
RDEL-Group/compound-knowledge-work
Agent
Domain investigation specialist. Reviews imported references, searches solutions library, and produces RESEARCH.md with reference recommendations and token budgets for the planner.
RDEL-Group/compound-knowledge-work
Agent
Stress-test specialist. Attacks drafted content from hostile perspectives — devil's advocate, risk assessment, audience pressure testing. Runs alongside the standard reviewer on high-stakes phases.
RDEL-Group/compound-knowledge-work
Agent
Quality gate specialist. Reviews drafted content against requirements, classifies issues by P1/P2/P3 severity, and generates revision task plans for critical problems.
RDEL-Group/compound-knowledge-work
Agent
Project scoping specialist. Conducts structured interviews, searches prior solutions, and generates the complete file set needed to start a CKW project.