Communitygithub.com

robcsaszar/socratic

A seven-question post-ship retrospective skill for Claude Code, built around confidence, blind spots, and decay

Was ist socratic?

socratic is a Claude Code agent skill that a seven-question post-ship retrospective skill for Claude Code, built around confidence, blind spots, and decay.

Funktioniert mitClaude Code~Codex CLI~Cursor
npx skills add robcsaszar/socratic

In Ihrer bevorzugten KI fragen

Öffnet einen neuen Chat, in dem dieser Agent-Skill bereits geladen ist.

Dokumentation

Was macht socratic?

Zētein — to seek. Not a code review: an inquiry held after the work has shipped. Answer as both builder and skeptic; an answer that flatters is worthless.

Method

  1. Name the target. The argument names it — a feature ("emotes"), a change set ("PR #754", a branch, a commit range), or the whole repository. No argument — take the most recent merged PR. If the target can't be resolved (no match, no merged PRs), stop and ask. State it in one line before anything else.
  2. Send the seekers. Launch parallel Explore subagents — always plural, one per territory — and answer nothing until all return. Feature or change set: implementation, tests, docs/flows — anchored on the diff (gh pr diff <n> for PRs, git diff <range> for ranges). Repository: architecture, seams between modules, drift between code and docs. Each seeker returns findings with file:line citations; answers draw only on what the seekers return plus the session itself.
  3. Ask the seven, in order. One at a time, under the rules below. No hedging; "it depends" is banned unless followed by on what.
  4. Close with follow-ups. At most three concrete follow-ups the answers demand (issue-ready, one line each), then one sentence naming the answer that stung most.

The seven questions

  1. Quis — who? Who is this truly for — and who was forgotten? Name the user who hits it first and the one who never appeared in the design.
  2. Quid — what? What are you least confident about right now? The weakest certainty in what shipped, not the ugliest code.
  3. Quando — when? If this breaks in three months, what is the most likely reason? Time is the one test that always runs.
  4. Ubi — where? What is the biggest thing not yet realized about the situation — where does the blind spot live?
  5. Cur — why? What assumptions were made and never stated? Surface the unspoken reasoning behind each choice while it is still cheap to admit.
  6. Quomodo — in what way? What could the asker have done differently to make this work smoother? Answer honestly; politeness that hides a real answer helps no one.
  7. Quibus auxiliis — by what means? Freed from the request entirely — one unrequested, industry-leading addition. What would make this the reason people mention the product?

Rules

  • Evidence first — every answer cites a file, a line, a commit, or a moment in the session; without one it is speculation, and must be marked as such.
  • The questions run in order; never read ahead — the dream feature in question seven, read early, leaks optimism into the honesty question two demands.
  • No answer is a valid answer. Having nothing is not failure; never invent one because the question seems to demand it.
  • Nothing gets fixed during the inquiry — observations here, prescriptions only in the follow-ups.
  • Seekers scout; you judge. Never delegate a question's answer to a subagent.

Verwandte Skills