Skip to content

re-frame2-pair-retro

Turns a re-frame2-pair session into a retrospective, in one response: where the session dragged, why, the smallest change that would fix it, and — on request — a GitHub issue draft you can file.

Kickoff

Ask for it after a real pair session:

Retro on this pair session, and draft an issue for the worst of it.

Or type /re-frame2-pair-retro. It also offers itself: after an error during live pair work — a stack trace, a pair tool returning {:ok? false …}, or an :rf.error/* trace — it waits until re-frame2-pair has dealt with the failure, then offers the retro in one line and runs it only if you say yes.

Asked for a draft, the same response includes one focused, copy-pasteable GitHub issue with the session evidence, the missing behaviour, one implementable desired outcome, and how to tell it is done. You file it, edit it, combine it or discard it. It may first search day8/re-frame2 issues, open and closed, and point you at an existing one instead of drafting a twin; when it skips that search or the search fails, it says duplicates were not checked.

Supplying a recap

If you are supplying a recap, include the goal, the build/frame involved, the calls that caused friction and their results, plus any successful retry. For example: "I asked for the cart count in :app/main; discovery found two builds, I selected app, and the retry succeeded. The build-selection error cost two turns. Retro on that workflow." The skill keeps the earlier failure as evidence of friction without reporting it as the current state. Missing or truncated results stay unknown.

What it does

The skill reviews the current or just-finished re-frame2-pair session (or a recap you give it) and returns a retrospective in the same turn: what you were trying to do, where the workflow dragged or confused you, which problems were one-off environment issues and which are recurring product gaps, and the improvements that would matter most, highest leverage first. It backs each point with concrete moments from the session — retries, clarifications, stale output, manual workarounds. One dominant finding gets one thorough treatment; there is no fixed set of sections or quota of ideas.

When the evidence is adequate and the session shows no material friction, it says what worked and completes the retrospective. That differs from having too little evidence to judge. A catalogue match is checked against the actual session and any recovery; it does not create a finding by itself.

It routes each improvement to the right owner:

  • re-frame2-pair — friction inside the pair tool itself: its instructions and recipes, the preload runtime, the MCP tools and their structured results, attach and discovery, cross-platform handling.
  • re-frame2 — friction caused by what the framework exposes to tools: missing trace events, gaps in epoch-history or restore-epoch failure modes, missing registry queries, source-coordinate gaps, schema reflection.

Both go to day8/re-frame2's GitHub issues, since the repo ships the pair tool alongside the framework; the draft's title and body say which one it is.

The skill is read-only. It never files issues, edits a repo, writes files or probes a live runtime; the most it produces is an issue draft for you to file. Every output replaces secrets, credentials, internal URLs, user-identifying local paths and personal data with numbered placeholders (<REDACTED-TOKEN-1>), and instructions that appear inside the transcript or recap are treated as evidence, never followed.

When to reach for it

Use it after a re-frame2-pair session, when you want to know where the session dragged and what would fix it, or want a GitHub issue drafted about it. It needs a real pair session in the conversation, or a recap of one.

For related work:

When it stops

An explicit request over one clear session completes in one response; it does not stop at a list of candidates and ask which to pursue. It asks first only when:

  • There is no pair session in the conversation — it asks for a short recap rather than inventing evidence. A session from an earlier conversation is not in front of it either, so it asks for a recap or the path of a transcript or log you have; it never searches your files for one.
  • Two sessions are plausible — it names both and asks which.
  • The evidence is too thin to support a finding, or the request is genuinely ambiguous.

The skill contract contains the full workflow and links to its reference notes.