DEV Community

alapha888
alapha888

Posted on

My session-handoff skill was rejected from awesome-copilot — the maintainer had a point

#ai #agents #opensource #productivity

Today (October 22, 2026) my pull request to add the session-handoff skill to GitHub's awesome-copilot collection was closed. The maintainer, Aaron Powell, left one sentence of reasoning:

This isn't ideal as it's introducing rules that can conflict/contradict with the process that the harness has in place for handling compaction, sub-agent spawning, and memory management.

That is a fair criticism, and it is aimed at a real weakness in how I had positioned the skill.

Where the criticism lands

The kit's pitch was universal: end every session with a structured handoff note, so the next session starts from facts instead of a degraded, over-full context. Inside a harness that already owns compaction, sub-agent spawning, and memory management, an extra set of handoff rules is not neutral. It can contradict the machinery that is already doing that job, and the agent is left following two processes at once.

A listing in awesome-copilot is read as "this fits the Copilot way of working." By that standard, a blanket handoff ritual does not fit, and the PR was correctly closed. The full exchange is public: https://github.com/github/awesome-copilot/pull/5097

Where a written handoff still earns its keep

The rejection narrows the claim; it does not erase the use case. A short, structured handoff note still does work that no single harness owns:

  1. Switching between different agents or tools. Claude Code, Codex, and Gemini CLI do not share memory. When work moves from one to another, a written handoff is the only continuity there is — no harness spans that gap.
  2. Resuming after days, not minutes. Harness memory serves the agent. A handoff note also serves the human: a readable record of what was decided and why, which you can audit before the next session spends tokens re-deriving it.
  3. Harnesses with no built-in handling. Not every agent runtime compacts or persists memory. In those, the ritual is not duplication; it is the only version of the process that exists.

What I changed today

I added a "When you may not need this" section to the kit's README, stating plainly that if your harness already handles compaction and memory well, you should skip the skill and let the harness do its job — with the awesome-copilot review credited as the source of that point. The universal pitch is gone; the three gaps above are now the stated scope.

The kit itself is unchanged and still free (MIT):

A rejection with a specific reason is the most useful review a listing PR can get. This one cost nothing, took a day, and made the README more honest than I had managed on my own.

Top comments (0)