Skip to content

hani-q/qstack / skills/qstack-ask-as-questions

qstack-ask-as-questions

Verified

Take every open question the previous assistant answer left for the user and ask each one through the host's structured question tool, rephrased for a named reader (product manager by default), with an ELI10 version, lettered options, and one recommendation. Use when invoked as `/qstack-ask-as-questions [role]`, when the previous answer ends with decisions the user must make, or when the user says they do not understand what is being asked of them. Not for a plan's open-questions section, which `/qstack-ask-plan-open-questions` owns.

/qstack-ask-as-questions

Turn the open questions in the immediately previous assistant answer into questions a reader outside the work can answer. Ask them one round at a time with the host's structured question tool, and report the answers.

This skill reads the conversation, not the plan. A plan's open-questions section belongs to /qstack-ask-plan-open-questions, which also writes the answers back into the plan. This skill writes nothing.

Read the invocation

The argument is free text naming the reader. Take the role from it and, when present, the product that reader owns.

  • /qstack-ask-as-questions (product manager)
  • /qstack-ask-as-questions ceo
  • /qstack-ask-as-questions support lead on the habit tracker

Choose the lens from the role table in skills/qstack-explain-for/SKILL.md and follow that table where it sits rather than copying it here. The default reader is a product manager.

Collect the questions

Read the previous substantive assistant answer. A question is anything that answer leaves for the user to decide: a sentence that asks, a numbered "your call" list, a choice between two named options, or a statement that work is waiting on the user's word. Collect every one. Do not add a question the answer did not raise, and do not drop one because it seems minor.

If the previous answer holds no such question, say so in one short line and stop.

Each question carries what the answer already knows about it: what the choice is, why it came up, what each way costs, and what the answer recommended if it recommended anything. That context is the raw material for the rephrasing; nothing is invented beyond it.

Shape each question

Every question is asked in this shape, in this order:

  1. The header, when the tool has one: a short label naming the subject, not the question.
  2. For a <role>: the question at the reader's altitude, written under the role's row of the explain-for table. Lead with the reader's silent question from that row. Assume the reader has not seen the work, the files, or the conversation, so carry the context an outsider needs: what was being done, why it stopped on this, and what waits on the answer. Drop file paths and internal names unless the reader has to use them.
  3. ELI10: the same question in two or three sentences a ten-year-old follows. A concrete comparison works; jargon does not.
  4. Options: lettered A, B, C, and so on, each in the reader's language, each saying what happens if it is picked and what it costs. The recommended option is listed first as A with (Recommended) in its label and a one-sentence reason in its description. If the previous answer recommended nothing, recommend the option that keeps the most existing decisions intact and say that is why.

The role and ELI10 versions carry the same facts and options the original question carried and add none of their own. They change the altitude, not the content.

Ask

Use the host's structured question tool when it exists. One call carries every question from the answer, up to the tool's limit; a larger set is asked in rounds of that size. Each question's text is items 2 and 3 above, separated by a blank line. The options are item 4, letter first in each label, recommended first.

Without a structured question tool, print the same shape as plain text, one block per question, and end with one line asking the reader to answer by letter, such as 1A 2B.

Report

After the answers arrive, print one line per question: the header, the letter chosen, and what it means in the reader's words. If an answer was free text rather than a letter, quote it. Then state in one line what happens next as a result, and stop. This skill does not carry out the decisions; the answers go back to whatever work was waiting on them.

slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall