Hypergraph-Coding-Agent-Framework: Skill for Claude Code

/hyper-sop — Master Standard Operating Procedure Guide

This skill guides the user through the Master Standard Operating Procedure (SOP) of the Hypergraph Coding Agent Framework.

The framework relies on a strict sequential model — Spec-First, Deterministic Memory, and Automated Auditing — to prevent context collapse and hallucinated requirements.

Step 1: Introduction

Briefly introduce the Hypergraph framework's core paradigm:

• The workflow is strictly sequential.
• Specialized agents handle distinct phases (Architect, Red Team, Resolution, Auditor).
• Deterministic Python scripts manage state so LLMs never traverse the graph probabilistically.
• New context windows (new conversations) prevent cross-contamination between adversarial agents.

Step 2: Determine User State

Use AskUserQuestion to determine the user's current state:

Which phase of the framework are you in?

- Option A: Phase -1: Legacy Onboarding — Integrating the framework into an existing project
- Option B: Phase 0: Initialization — Starting a brand new greenfield project
- Option C: Phase 1: Specification — Planning and designing a new feature
- Option D: Phase 2: Execution — Writing code against a compiled MiniPRD

Step 3: Provide Phase-Specific Guidance

Wait for the user's response, then provide the exact steps:

Phase -1 (Legacy Onboarding):

1. Run /hyper-discover — scans the codebase and populates spec/compiled/architecture.yml.
2. Review the generated YAML for accuracy.
3. Run /hyper-baseline — generates the initial SuperPRD.md.
4. Once verified, proceed to Phase 1.

Phase 0 (System Initialization):

1. Ensure the template is cloned with .agents/ and spec/ directories intact.
2. Install the dependency: pip install pyyaml
3. Give execution permissions: chmod +x .agents/scripts/*.py
4. The workspace is ready. Proceed to Phase 1.

Phase 1 (Specification):

1. Run /hyper-architect — begins the requirements extraction interview (creates Draft_PRD.md in spec/active/).
2. Start a new conversation and run /hyper-redteam — adversarial analysis (creates RedTeam_Report.md in spec/active/).
3. Start a new conversation and run /hyper-resolve — resolves trade-offs and compiles SuperPRD.md and MiniPRD files into spec/compiled/.
4. The resolve agent will automatically run python .agents/scripts/archive_specs.py [Feature_Name] to flush active drafts.