echoVic/boss-skill

What it solves

Boss transforms a single coding agent into a structured, auditable engineering team. It solves the problem of "prompt-only" orchestration, where agents often hallucinate state or fail to follow a plan without proof. Boss provides a verifiable runtime that ensures requirements, design, and tests are actually executed and recorded as evidence.

How it works

Boss operates as a "skill" installed into coding agents (such as Claude Code, Codex, or OpenClaw). It implements a four-stage pipeline that simulates a full software engineering team (PM, Architect, UI Designer, Tech Lead, Scrum Master, Frontend, Backend, QA, and DevOps).

Key technical mechanisms include:

• Event-Sourced Runtime: All pipeline state is appended to an event log (events.jsonl), making the process replayable and auditable.
• Quality Gates: Verifiable checks (e.g., linting, test evidence, and deployment checks) must be passed before the workflow can progress.
• Artifact Generation: The system generates a traceable folder (.boss/<feature>/) containing PRDs, architecture docs, and QA reports.
• Zero-Network Boundary: By default, it performs no outbound network requests, ensuring privacy and security.

Who it’s for

Developers and teams using AI coding agents who need more than simple code edits. It is designed for those building new features or medium-sized products that require formal requirements, technical design, and verifiable test evidence.

Highlights

• Auditable Workflow: Replaces vague prompts with a deterministic event stream and verifiable artifacts.
• Multi-Role Simulation: Orchestrates roles from PM to DevOps to ensure a complete engineering lifecycle.
• Agent-Agnostic: Compatible with multiple agents including Claude Code, Codex, OpenClaw, Antigravity, and Hermes.
• Security-First: Features a zero-network default policy and prevents shell injection via structured argument verification.
• Deterministic Evals: Allows scoring of captured transcripts without needing to call a live LLM.