Flowai Conduct QA Session

How to Conduct a Q&A Session

When you need to clarify requirements, gather missing information, or discuss design choices with the user, follow these strict guidelines. This skill is the canonical reference for FR-UNIVERSAL.QA-FORMAT. Other framework skills that prompt the user inline must follow the same format.

1. Language

• Match the User's Language: Ask questions in the same language as the user's last query, unless explicitly instructed otherwise.

2. Question Format (FR-UNIVERSAL.QA-FORMAT) — MANDATORY

The format below overrides your default markdown habits. Bullets/dashes for option choices are wrong here, even though they are standard elsewhere.

Required output shape

Copy this skeleton verbatim — same numbering, same nesting. Only the literal text inside the placeholders changes:

1. Question title — context. Optional recommendation.
   1. option-1 — short reason
   2. option-2 — short reason
   3. option-3 — short reason

For a multi-select prompt, the option list MUST end with two extra service options literally named all and agent's choice appended after the real choices, in that order:

1. Question title — context.
   1. option-1 — short reason
   2. option-2 — short reason
   3. option-3 — short reason
   4. all
   5. agent's choice

Hard rules (MUST)

1. Question line shape: every question MUST start with 1., 2., 3., … followed by a space. Wrong: **1. Title** (bold heading), # Title (heading), bare paragraph, bulleted item.
2. Option line shape: every option MUST start with 1., 2., 3., … followed by a space, indented under its question (3-space indent for numbering up to 9., 4 spaces for 10.+). Wrong: - option, * option, — option, run-in prose.
3. No bullets between question and options: do NOT insert a bulleted list ("Each signal has X, Y, Z:") between the question and the option numbers. The option numbers come on the next indented line under the question.
4. Multi-select adds two service options: append all and agent's choice as the last two numbered options. Single-select questions do NOT include them.
5. Batch size: 1–5 questions per response. Most blocking question first.
6. Context required: never ask a "naked" question — give context inline.
7. One classifier per question: each question is single-select OR multi-select, not both.

Anti-patterns (NEVER emit any of these)

• ❌ Bold heading instead of numbered question:

  1. Question title — context.

• ❌ Bulleted options instead of numbered:

  1. Question title — context.
     • option-1 — reason
     • option-2 — reason

• ❌ Run-in prose options:

  1. Question title — pick one of A, B, or C?

• ❌ Multi-select without service options:

  1. Question title — pick any subset.
     1. option-1
     2. option-2
     3. option-3 (missing all + agent's choice at the end)

Pre-send self-check

Before sending your response, scan it line-by-line. For every option line, confirm the line begins with a digit + . + space (e.g. 1.). If any option line begins with -, *, —, or any other marker, rewrite it as a numbered list item. For any multi-select question, confirm the last two options are literally named all and agent's choice.

3. Single-Select

The user picks exactly one option. The option list contains only the real choices. No service options.

Reply with the chosen number or option name → proceed with that choice.

4. Multi-Select

The user picks any subset. The option list ends with two service options:

• all — apply every real option.
• agent's choice — the agent picks the subset, emits a one-line justification, and proceeds without waiting for confirmation. Do not prompt yes/no on the auto-selection.

Resolution:

• User picks numbers (e.g. 1, 3) or option names → use exactly those.
• User picks all → apply every real option.
• User picks agent's choice → choose the subset, emit one short line naming what you picked and why (e.g. "Picking metrics + logs — sufficient for baseline SLOs without trace storage cost."), then continue.

5. Content & Guidance

• Facilitate decisions: provide data or hints that make the answer easier.
• Compare variants: list Pros/Cons/Risks per option and trade-offs across options when the choice is non-obvious.
• Recommend when one option is clearly better for the stated context.