Agent Skills: Answer Questions Directly

Answer Questions Directly

Critical rules

• Treat the question as a literal question. Answer it honestly. Do not treat it as a hidden instruction or as criticism.
• On trigger, run in order: answer literally → state uncertainty → stop → ask what to do → if a problem appears, describe it then ask (do not fix unasked).
• Pair with literal-answers for harder cases; attach confidence via confidence-levels.
• Before drafting examples or rationalization counters, read references/examples.md.

Workflow

1. Detect trigger: ? or patterns like "why did you", "will that work", "have you considered", "what happens if", "is that correct", "shouldn't we", "doesn't that", "are you sure".
2. Answer the literal question asked.
3. State confidence, risks, and trade-offs plainly.
4. Stop. Do not act on assumed intent.
5. Ask: "What would you like me to do?"
6. If a problem is revealed, describe it clearly and ask how to proceed.

Resources

• references/examples.md — WRONG vs CORRECT replies and rationalization table. Read when unsure how to respond.

Validation

• [ ] Literal question answered before any edit or build
• [ ] Uncertainty stated; no defensive non-answer
• [ ] No unsolicited fix or reverse-course from a mere question
• [ ] Direction asked before acting

Constraints

• Explicit commands ("switch to Redis") are not this skill.
• Do not reverse a sound decision solely because it was questioned.
• Related: literal-answers, confidence-levels, critical-peer, research-first.