handoff-pack

Multi-model Agent context handoff packet CLI

Turn OpenAI-compatible messages[] into a paste-ready resume pack for the next model — with secret redaction and before/after token stats.

Local-first · IDE-agnostic · zero npm deps · Node ≥ 18.

git clone https://github.com/ktz03/handoff-pack.git
cd handoff-pack
node bin/handoff-pack.js preview examples/long.session.json --max-tokens 200
node bin/handoff-pack.js pack examples/long.session.json --max-tokens 200 --out examples/RESUME.long.md

On the bundled long session (--max-tokens 200): 508 → 181 tokens (~64% cut), demo secrets redacted, system/tool/error + recent turns kept.

npm test   # or: npm run demo:long

messages.json → redact secrets → priority compress → RESUME.md / JSON packet

Keep order: all system → high-signal tool/error turns → most recent user/assistant until budget fills.

• task_state — goal / last assistant / blockers / open questions
• next_actions — continue checklist
• plans — A / B / C fallback paths
• compressed_messages — budget-fitted history

Coding / Agent sessions are context silos. Switching models needs a handoff layer, not another IDE plugin.

MIT