sdd-riper  by huisezhiyin

AI agent harness for steerable, observable coding

Created 3 months ago
272 stars

Top 94.6% on SourcePulse

GitHubView on GitHub
Project Summary

Lightweight AI Agent Harness for agentic coding, SDD-RIPER enables strong AI models to advance work while humans maintain control over goals, boundaries, and acceptance. It provides a structured approach to managing AI-driven development, making processes observable, recoverable, and safe. This toolkit is designed for engineers and researchers working with sophisticated coding agents who need a robust control plane.

How It Works

The harness operates on a "context-first" principle, emphasizing small active contexts and persisting durable state externally. It comprises four complementary skills: sdd-riper-one-light for daily tasks with low-friction checkpoints, sdd-riper-one for high-risk work requiring stricter gates, codemap for indexing unfamiliar codebases, and new-chat-ready for managing chat handoffs and recovery. This design prioritizes observability, recoverability, and safety by building a control plane around AI agency.

Quick Start & Requirements

Usage involves specific prompts tailored to each skill (e.g., "Use sdd-riper-one-light for this task."). The toolkit integrates with existing agentic coding environments like Codex and Claude Code. Recommended reading paths include skills/sdd-riper-one-light/README.md for immediate use and docs/README.md for deeper dives. A system-level template for agent rules is available at examples/global-agents.md.

Highlighted Details

  • Features four distinct skills: sdd-riper-one-light (default), sdd-riper-one (strict), codemap (code indexing), and new-chat-ready (handoffs).
  • Enforces a core workflow: read context, restate goal/risk, checkpoint, execute, validate, reverse sync.
  • Adheres to strict harness rules: Restate First, No Spec, No Code, No Approval, No Execute, Done by Evidence, Reverse Sync.
  • Defines three distinct layers for spec, project knowledge, and system defaults to manage context and memory boundaries.
  • Mandates rigorous repository hygiene, prohibiting commits of runtime data, credentials, or sensitive user information.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels like Discord or Slack.

Licensing & Compatibility

The repository's license is not specified in the README, making it difficult to assess compatibility for commercial use or closed-source integration without further clarification.

Limitations & Caveats

The "light" version (sdd-riper-one-light) offers lower friction but may be less suitable for extremely high-risk tasks compared to sdd-riper-one. Setup relies on existing agentic coding environments, and specific platform or LLM compatibility beyond general categories is not detailed.

Health Check
Last Commit

5 days ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
2
Star History
54 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.