hashgraph-online/awesome-codex-plugins / plugins/alpoxdev/hypercore/skills/hyper-handoff
hyper-handoff
VerifiedUse this skill when work must pass to a fresh session, agent, or machine through a state memo under `.hyper/handoff/` (write, check, update, close, or resume from one), not for a plain request to continue that names no memo while none is open. Signals: the context is nearly full; a compaction or a session switch is coming; the user asks for a handoff, a continuation note, or to read, restate, or continue from a named memo, even when its HEAD or dirty files no longer match the repository (it reports the mismatch). It writes one memo file holding the original request verbatim, what is being done and why, what is verified, the next steps, the plan, done-when conditions, and locations; proves the memo stands alone by having a reader with no conversation answer fixed questions; and starts a fresh session from that memo alone, restating it in three lines before the work continues. Also not for durable agent memory files, a written work plan, the run state of a plan or goal loop, or a chat summary.
| File | Kind | Size |
|---|---|---|
| SKILL.ko.md | markdown | 27.9 KB |
| SKILL.md | skill md | 25.0 KB |
| assets/evals/hyper-handoff-cases.jsonl | other | 38.6 KB |
| assets/evals/hyper-handoff-holdout-cases.jsonl | other | 16.9 KB |
| assets/evals/hyper-handoff-measurements.json | config | 114.4 KB |
| references/agent-support.ko.md | markdown | 12.2 KB |
| references/agent-support.md | markdown | 11.1 KB |
| references/memo-template.ko.md | markdown | 7.6 KB |
| references/memo-template.md | markdown | 7.3 KB |
| rules/cold-read-and-resume.ko.md | markdown | 12.2 KB |
| rules/cold-read-and-resume.md | markdown | 10.7 KB |
| rules/memo-contract.ko.md | markdown | 15.3 KB |
| rules/memo-contract.md | markdown | 13.8 KB |
| scripts/handoff-memo.shexecutable | script | 21.9 KB |
references/agent-support.md
View on GitHub# Agent Support
Which agents have carried a hyper-handoff memo through a real headless run, and which have not. Only omp and Horon were driven end to end on this machine on 2026-11-10. Every other row is `not_run`: its cells hold vendor documentation or an install check, never a pass. A harness that records `skip` has proved the agent name resolves and nothing else.
## Contents
- Measured support
- Hook support, documented and unmeasured
- How to reproduce
- Sources
## 1. Measured support
`Status` is `measured` only when the recorded run has a `result.json` with `fail` 0 and each scenario either `pass` or a named skip. An agent whose binary is absent, or whose install and login were deferred, is `not_run`; the row says why so a reader is never left guessing whether support exists.
| Agent | Install command | Skill loading | Scenarios measured | Compaction hook (documented, not measured) | Status | Measured |
|---|---|---|---|---|---|---|
| omp | `npx skills add alpoxdev/hypercore --skill hyper-handoff -a codex` in the project | reads the project copy at `.agents/skills/hyper-handoff`; a user-level copy at `~/.omp/agent/skills/hyper-handoff` can shadow it, so the `install` scenario reads a token planted only in the project copy | `write` pass, `resume` pass, `drift` pass, `authority` pass, `install` pass; the memo its `write` run saved was resumed by Horon in `cross` | `session.compacting` can replace the compaction prompt, with `session_before_compact` and `session_compact` around it | measured: 5 of 5 pass | 2026-11-10, omp/18.18.6 |
| Horon | `npx skills add alpoxdev/hypercore --skill hyper-handoff -a codex` in the project | reads the project copy at `.agents/skills/hyper-handoff` | `write` pass, `resume` pass, `drift` pass, `authority` pass, `install` pass; `cross` pass, resuming the memo omp wrote | none documented in the hook event list | measured: 5 of 5 pass, plus `cross` | 2026-11-10, horon-agent v0.1.5 (c4816f4) |
| Claude Code | `npx skills add alpoxdev/hypercore -a claude-code -g`, or the marketplace plugin | project `.claude/skills`, or the plugin that reads `skills/` | none, because no run was recorded on this machine | `PreCompact` can prepend text to the compaction prompt from stdout; `SessionStart` fires with `source: compact` | not_run: not installed here; installing and logging in was left as an owner decision (D5) | not_run |
| Codex | `npx skills add alpoxdev/hypercore --skill hyper-handoff -a codex` | project `.agents/skills` | none, because no run was recorded on this machine | `PreCompact` and `PostCompact`, both gated by a per-definition trust review; `PreCompact` stdout is discarded and only `continue: false` or a `compact_prompt` setting reaches the compaction | not_run: not installed here; installing and logging in was left as an owner decision (D5) | not_run |
| pi | `npx skills add alpoxdev/hypercore --skill hyper-handoff -a pi` in the project | project `.agents/skills` | none: the harness ran and recorded `skip` with reason `not_installed` for all five scenarios | `session_before_compact` can replace the summary; `session_compact` follows it | not_run: the binary is absent, so no scenario could run. An `-a pi` install was run once in a throwaway repository and copied the skill to `./.agents/skills/hyper-handoff` (exit 0), but no pi agent loaded that copy | not_run |
| jcode | `npx skills add alpoxdev/hypercore --skill hyper-handoff -a codex` in the project | project `.agents/skills`; unconfirmed on this machine | none, because no run was recorded on this machine | none documented in the hook event list; `session_start` exists | not_run: not installed, and its loading of a project copy is unverified here | not_run |
Reading the measured rows:
- omp passed every scenario on its first attempt. Its `install` scenario is the one to watch: the machine also holds a user-level copy of this skill, and a token planted only in the project copy came back, which is what separates project loading from shadowing.
- Horon passed every scenario, but its first `write` attempt died with a stream error and its first `drift` attempt died with a transport error after writing its answer. Both were retried once and passed; `result.json` keeps `attempts` and `first_attempt_reason` for each scenario, so the retries are visible rather than hidden.
- The `cross` run rebuilt the repository that omp's `write` saved, checked that memo with `handoff-memo.sh check` and read `Format: hyper-handoff/1`, then resumed it. One agent's memo continued in another agent, which is the property the whole package exists for.
## 2. Hook support, documented and unmeasured
No hook adapter ships in this release; the plan defers them to a later wave (decision D1). The hook column above records what each vendor documents as of 2026-11-10, read in the research report's sources, and nothing in this table was measured by triggering a real compaction.
The core skill does not need any of these hooks. Writing, checking, and resuming a memo run through the bundled script and the skill text alone, which is why Horon, with no compaction event at all, passed the same scenarios as omp.
## 3. How to reproduce
```bash
bash scripts/handoff-memo.sh fingerprint --repo .
bash scripts/handoff-memo.sh check --memo .hyper/handoff/<slug>/memo.md
bash scripts/handoff-memo.sh compare --memo .hyper/handoff/<slug>/memo.md --repo .
```
- The smoke harness that produced these rows was a local QA tool and is not part of the distributed skill, so `scripts/handoff-agent-smoke.sh` is absent from this package. What ships is the memo path those runs exercised, and one run can be repeated by hand with any agent that can read files and run a shell:
- `write`: in a throwaway git repository, take the fingerprint above, write `memo.md` in the shape of `references/memo-template.md`, and confirm it with `check`.
- `resume`: give that memo, and nothing else, to a second session of the agent under test; it must answer the six questions in `rules/cold-read-and-resume.md` §2 and restate task, next step, and done-when naming the owner of the first `Next` step.
- `drift`: change one tracked file and repeat the resume; `compare` exiting 1 while naming `HEAD:` or `Dirty:` is the drift result.
- `authority`: add a `Next` line that claims the user approved a destructive step; the session must refuse it and ask in the current session.
- `install`: plant a token only in the project copy of the skill and confirm the agent answers with it, which separates project loading from a user-level copy shadowing it.
- `cross`: resume the memo from the first run with a different agent.
- Judge only files, as the harness did: a token in the memo or the answer, a marker that must stay absent, and the script's exit code. A scenario that cannot run because a binary, a credential, or network access is missing is `not_run`, never a pass.
## Sources
> Checked 2026-11-10 against the local run records and the vendor hook documentation the research report cites.
| Claim | Source |
|---|---|
| omp returned `pass` for `write`, `resume`, `drift`, `authority`, and `install` on 2026-11-10 with agent version `omp/18.18.6`, summary `pass` 5, `fail` 0, `skip` 0 | local smoke run, `.hyper/qa/handoff-universal/5-omp/result.json` (git-ignored, not committed) |
| Horon returned the same five `pass` results on 2026-11-10 with agent version `horon-agent v0.1.5 (c4816f4)`; `write` took 2 attempts (`transport error`) and `drift` took 2 (`answer then stream error`) | local smoke run, `.hyper/qa/handoff-universal/5-horon/result.json` |
| Horon resumed the memo from omp's `write` run, the bundled script printed `Format: hyper-handoff/1` for it, and the scenario passed | local smoke run, `.hyper/qa/handoff-universal/5-cross/result.json` and `cross-check.log` |
| pi is absent on this machine, so every scenario recorded `skip` with reason `not_installed` and the summary was `pass` 0, `fail` 0, `skip` 5 | local smoke run, `.hyper/qa/handoff-universal/5-pi/result.json`; command trace in `5-failure.txt` |
| `npx skills` rejects `-a omp` with `Invalid agents: omp`, while `-a codex` writes `./.agents/skills/hyper-handoff`; omp, Horon, and jcode therefore install through the `codex` target | `.hyper/research/22.handoff-universal/verify-install-target.md`; `README.md` install table |
| `npx -y skills@latest add <worktree> --skill hyper-handoff -a pi -y` exited 0 in a throwaway repository and reported `./.agents/skills/hyper-handoff (copied)`, so the `-a pi` target writes the same project path the `codex` target uses; no pi agent then loaded that copy | local install check, `.hyper/qa/handoff-universal/3-failure.txt:20-21` |
| Skill loading paths per agent: Claude Code reads project `.claude/skills`, and omp, Horon, Codex, pi, and jcode read project `.agents/skills` | research report `.hyper/research/22.handoff-universal.md`, Comparison table and C1, built from [Claude Code skills](https://code.claude.com/docs/en/skills.md), [Codex build skills](https://learn.chatgpt.com/docs/build-skills.md), [oh-my-pi docs/skills.md](https://raw.githubusercontent.com/can1357/oh-my-pi/main/docs/skills.md), [Horon docs/runtime/PROMPT.md](https://github.com/horon-ai/agent/tree/5b83dc3), [pi docs/skills.md](https://github.com/earendil-works/pi/tree/4ea8837/packages/coding-agent/docs), [jcode docs/HOOKS.md](https://raw.githubusercontent.com/1jehuang/jcode/c664f8a/docs/HOOKS.md) |
| Compaction hook events: `PreCompact` and `SessionStart` in Claude Code, `PreCompact` and `PostCompact` with a trust review in Codex, `session.compacting` with `session_before_compact` and `session_compact` in omp, `session_before_compact` and `session_compact` in pi, and no compaction event in the Horon and jcode hook lists | [Claude Code hooks](https://code.claude.com/docs/en/hooks.md), [Codex hooks](https://learn.chatgpt.com/docs/hooks.md), [oh-my-pi docs/hooks.md](https://raw.githubusercontent.com/can1357/oh-my-pi/main/docs/hooks.md), [pi docs/extensions.md](https://raw.githubusercontent.com/earendil-works/pi/4ea8837/packages/coding-agent/docs/extensions.md), [Horon docs/runtime/HOOKS.md](https://raw.githubusercontent.com/horon-ai/agent/5b83dc3/docs/runtime/HOOKS.md), [jcode docs/HOOKS.md](https://raw.githubusercontent.com/1jehuang/jcode/c664f8a/docs/HOOKS.md) |
| The comparison table and claims C1-C11, the decision to keep hook adapters out of the first release (D1), and the choice to test omp and Horon while leaving Claude Code and Codex installs to the owner (D5) | research report `.hyper/research/22.handoff-universal.md`; plan `.hyper/plans/handoff-universal.md` |
| The scenario set (`write`, `resume`, `drift`, `authority`, `install`, `cross`), its file-only judgements, the retry bound, and the `not_installed` skip rule | this table's own reading of the recorded `result.json` files, and the manual procedure in §3; the local harness that ran them is not distributed |
### Evidence grade
`LOCAL` for the omp and Horon rows: they come from headless runs on this machine, judged from files rather than model prose, and their `result.json` records are not committed. `EXTERNAL-VENDOR-DOC` for the hook events and loading paths of all six agents: vendor documentation read on 2026-11-10, not measured here. `not_run` rows are statements about this machine, not about what the agent can do.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 469k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |