Long Task Handoff
/install long-task-handoff
Long Task Handoff
Keep this skill lightweight. Do not manually reproduce the full handoff template in context. Use the bundled Python manager and keep user-facing messages short.
Detailed protocol is deferred to references/protocol.md. Do not load it during normal operation. Load it only if the manager is unavailable, the skill is being audited or modified, or a manual fallback is required.
Use The Manager
Prefer:
python scripts/handoff_manager.py update --workspace . --task "Task name" --event context_compaction --compaction-count N
Installed script paths:
- Codex:
.codex/skills/long-task-handoff/scripts/handoff_manager.py - Hermes:
/root/.hermes/skills/software-development/long-task-handoff/scripts/handoff_manager.py
Useful commands:
python scripts/handoff_manager.py create --workspace . --task "Task name"
python scripts/handoff_manager.py update --workspace . --task "Task name" --event context_compaction --compaction-count N
python scripts/handoff_manager.py recover --workspace .
python scripts/handoff_manager.py suggest --compaction-count N --json
python scripts/handoff_manager.py validate handoffs/session-handoff-*.md
Pass rich task facts with --input-json when available. Otherwise pass only concise CLI facts such as --completed, --test-result, --key-file, --unfinished, --next-action, --risk, and --do-not-do.
Automatic Behavior
- First compaction: update handoff quietly.
- Second compaction: update handoff and verify
handoffs/ACTIVE.md. - Third compaction: update handoff and briefly say restart is advisable.
- Fourth compaction or state loss: update handoff and strongly recommend restart.
- Fresh session with "continue", "keep going", "继续", or "接着做": run
recoverbefore asking the user for context.
The user should not need to know the handoff path. A new session prompt like "continue this task" should be enough when ACTIVE.md exists.
Boundaries
The handoff is a restart packet, not a project wiki. Include only current restart-critical state: goal, branch/commit/worktree, delta, tests, key files, unfinished items, next actions, risks, and do-not-do items.
Do not include secrets, .env contents, private keys, long chat logs, stale plans, unverified guesses as facts, or descriptions that conflict with the current workspace. Put durable project knowledge in repo docs and reference it by path.
If Script Fails
Do not silently continue. Report the concrete failure, keep the message short, and avoid asking the user to restate context unless recover cannot find or read a handoff.
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install long-task-handoff - After installation, invoke the skill by name or use
/long-task-handoff - Provide required inputs per the skill's parameter spec and get structured output
What is Long Task Handoff?
Automatically create a compact, standardized handoff file for long tasks to enable seamless session restarts or agent transfers without losing context or pro... It is an AI Agent Skill for Claude Code / OpenClaw, with 49 downloads so far.
How do I install Long Task Handoff?
Run "/install long-task-handoff" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Long Task Handoff free?
Yes, Long Task Handoff is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Long Task Handoff support?
Long Task Handoff is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Long Task Handoff?
It is built and maintained by hollis9087 (@hollis9087); the current version is v0.3.3.