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

Execute

This skill implements a plan precisely, minimally, and in full — adhering to existing conventions, updating progress tracking, and ensuring the hypergraph is kept in sync after every modification.

When to use this skill

• When the user has a compiled MiniPRD_*.md in spec/compiled/ and is ready to build.
• When the user explicitly runs /hyper-execute or asks to "implement the plan."
• After /hyper-resolve has compiled the final MiniPRDs and the workspace has been archived.

How to use it

1. Memory Check

   • Read .agents/memory/activeContext.md before anything else.

   • Check whether the target MiniPRD is listed as complete or already audited.

   • If the MiniPRD is marked complete or audited: HALT. Display the warning, then use AskUserQuestion:

     ⚠️ [MiniPRD name] appears already complete per activeContext.md.
     Run /hyper-audit to archive it, or move it manually:
       mv spec/compiled/[MiniPRD].md spec/archive/[MiniPRD]_AUDITED.md
     
     Confirm you want to re-implement it?
     
     - Option A: Yes, re-implement — proceed even though it appears done
     - Option B: No, stop — I will archive or review it instead
     

2. Read the MiniPRD

   • Read the target MiniPRD_*.md from spec/compiled/.

   • Confirm a Confidence Score (1–10) against the MiniPRD's "Confidence Mandate" section. If below 9, list the clarifying questions, then use AskUserQuestion:

     Confidence is [N]/10. How do you want to proceed?
     
     - Option A: Answer my questions — provide clarifications so I can proceed
     - Option B: Proceed anyway — accept the risk and implement with current information
     - Option C: Stop — halt until the MiniPRD is revised
     

2.5. Codebase Orientation (Haiku Sub-Agent)

• Before implementing, launch a read-only sub-agent to pre-load all relevant file contents:
  • Use the Agent tool with subagent_type: "general-purpose" and model: "haiku"
  • Prompt the sub-agent: "Read spec/compiled/[MiniPRD filename] in full. From the Implementation Plan task list, identify every file path that must be created or modified. For each file that already exists in the codebase, read its full content and return it verbatim. For new files, list the path and expected purpose only. Return a structured report mapping file paths to their current content."
• Wait for the sub-agent result before proceeding to Step 3.
• Use the returned file contents as your working context for implementation.
• Fallback: If the sub-agent returns an error or empty result, proceed to Step 3 and read files directly as normal.
• Do not re-read files the sub-agent already returned unless you need to verify a specific line.