compound_skill

Overview

This skill is a Meta-Engineer agent that enforces a strict Plan → Delegate → Assess → Codify loop to improve development outcomes. It biases effort distribution: 80% on planning and assessment, 20% on execution, and guarantees every task produces reusable knowledge. The agent owns the Codify step, turning outcomes into rules, patterns, or snippets stored in the Knowledge Base.

How this skill works

The agent automates and enforces the four-phase loop: it generates detailed specs and plans, delegates execution to role-specific agents, runs quality assessment and scoring, and then codifies learnings into the KB. It integrates with the local tooling (asdlc.py commands) to research similar work, trigger agent roles (@DEV, @UIUX, @TESTER), run observations and scoring, and append new rules or templates. Commands like compound plan, compound assess, and compound codify drive each phase so the process is repeatable and auditable.

When to use it

• Starting any non-trivial feature or bug that benefits from formal planning and reuse
• When you want to capture institutional knowledge and prevent repeated fixes
• Before handing work to multiple contributors or external agents
• When quality gates and measurable assessment are required
• When building a self-learning SDLC layer that must evolve with each task

Best practices

• Always complete and approve specification.md and implementation_plan.md before task execution
• Spend at least 60–80% of effort on research and assessment to reduce rework
• Use the role delegates (@DEV, @UIUX, @TESTER) for focused execution and avoid ad-hoc coding
• Require assessment score > 8/10 before codifying; rerun tests if criteria fail
• Write concise KB entries: pattern name, context, triggers, solution, and tests

Example use cases

• Onboarding a new feature: generate specs, delegate to dev/testers, ensure tests pass, then codify the design pattern
• Fixing recurring bugs: research similar bugs in KB, plan a robust fix, validate, and record the prevention rule
• Introducing architecture patterns: prototype via delegated agents, assess compliance, then publish reusable templates
• Automating release checklists: plan checks, delegate verification agents, assess results, and add checklist templates to KB