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

Hyper-Handoff

Write a handoff document summarising the current conversation so a fresh agent can continue the work without needing this conversation history.

---

Step 1 — Determine Output Path

Resolve the OS temp directory by running:

python3 -c "import tempfile; print(tempfile.gettempdir())"

Name the file: handoff_<YYYYMMDD>.md using the current date (date only — no time component). Save it in the temp directory resolved above — not in the current workspace.

Disambiguate same-day handoffs with a version suffix. If a file with that exact name already exists in the temp directory, append _v2, _v3, … to the end of the filename — before the .md extension — using the next unused number. The first handoff of the day carries no suffix:

tmp="$(python3 -c 'import tempfile; print(tempfile.gettempdir())')"
base="${tmp}/handoff_$(date +%Y%m%d)"
path="${base}.md"
n=2
while [ -e "$path" ]; do
  path="${base}_v${n}.md"
  n=$((n + 1))
done
# $path is now collision-free, e.g.
#   /tmp/handoff_20260606.md      (first of the day)
#   /tmp/handoff_20260606_v2.md   (second of the day)

---

Step 2 — Read Key Artifacts (Do Not Duplicate)

Before writing, read and note paths/URLs for any existing artifacts that capture ground truth. Reference them by path rather than reproducing their content:

• spec/compiled/architecture.yml — hypergraph state
• spec/active/ — any in-progress PRDs, Red Team reports, or draft specs
• spec/compiled/ — finalized SuperPRDs and MiniPRDs
• .agents/memory/activeContext.md — current sprint state and recent decisions
• Recent git log (git log --oneline -5) — commit history

---

Step 3 — Write the Handoff Document

Structure the document as follows:

# Handoff: <brief title derived from conversation context>

**Generated:** <ISO 8601 timestamp>
**Next session focus:** <from $ARGUMENTS, or "general continuation" if none>

## Context Summary

<2–4 sentences: what problem is being solved and where the work stands.
No fluff. Written for an agent reading cold.>

## What Was Decided

<Bullet list of key decisions made this session. If a decision is captured in
a PRD or commit, reference the path — do not reproduce it.>

## What Was NOT Resolved

<Bullet list of open questions, blocked items, or deferred decisions.
Be specific — vague items like "needs more work" are not useful.>

## Current State of Artifacts

| Artifact | Path | Status |
|---|---|---|
| architecture.yml | spec/compiled/architecture.yml | <clean / has dirty nodes> |
| Active specs | spec/active/ | <list files or "empty"> |
| Compiled specs | spec/compiled/ | <list key files> |
| activeContext.md | .agents/memory/activeContext.md | <one-line summary> |

## Suggested Skills for Next Session

<List the HACF skills the next agent should invoke, in recommended order,
with a one-line reason for each. Choose from:>

- `/hyper-architect` — if requirements still need extraction
- `/hyper-redteam` — if a Draft PRD exists and needs adversarial review
- `/hyper-resolve` — if a Red Team report exists and needs trade-off resolution
- `/hyper-execute` — if a MiniPRD is ready to implement
- `/hyper-audit` — if code was just written and needs verification
- `/hyper-document` — if documentation is out of sync
- `/hyper-contextualize` — if agent files may have HACF framing issues
- `/hyper-status` — to reorient before any phase
- `/hyper-troubleshooting` — if something is broken or desynchronized

## Exact Next Action

<One sentence: the single most important thing the next agent should do first.
Example: "Run /hyper-redteam on spec/active/Draft_PRD.md in a new context window.">