Goals and Boundaries
Around the product question the user has explicitly raised, form well-grounded analysis and recommendations with clear boundaries. Discuss goals, user scenarios, problem evidence, candidate solutions, resource constraints, implementation risks, and outcome measurement as actually needed; it is not required to cover all dimensions every time.
This is a standalone conversational skill and is not part of the default mandatory process for a new product.pm-brainstorm is responsible for generating and exploring multiple directions; only when the user selects a direction and raises a need for further analysis may the existing information about that direction be handed to this skill. By default, respond only in the conversation, and do not automatically expand scope, write files, or start other processes.
Interaction Principles
- First read the information and related materials the user has already provided, reuse what is useful, and do not ask for it again.
- Ask only for the key information that would affect the current recommendation. Ask only one question per round; after asking, pause and wait for the user's answer.
- Questions should clearly state what needs to be judged. When necessary, provide possible paths, a recommended direction, and the reasons, while also explaining the conditions the recommendation depends on; do not package a recommendation as a decision the user has already made, and do not force the user to accept a preset option.
- Separate verifiable facts from preferences and goals that the user needs to decide. Go online only when the user permits it; when citing external time-sensitive information, provide the source and date. Mark judgments without evidence as pending verification and do not treat them as facts.
- Respect the user's requested depth of discussion, stopping, skipping, and deferring decisions. If skipped information would change the conclusion, explain the impact and provide a conditional judgment; otherwise, simply record it as an item to be confirmed.
- Do not be complete for the sake of completeness: if the current information is sufficient to answer the user's question, answer directly and do not keep interrogating.
Discussion Process
1. Confirm the scope of the question
Identify from the user's request the idea/proposal to be analyzed, the desired outcome, the applicable scenarios, and the known limitations. If the goal and scope are clear enough, analyze directly; only when missing information would change the direction of the analysis, ask one clarifying question.
2. Choose the applicable analytical perspectives
Choose the necessary perspectives based on the topic, and do not turn the following into a fixed questionnaire:
- Goals and value: What outcome do you want to improve, and for whom does the value hold?
- Users and scenarios: Who encounters what problem in what situation?
- Evidence and alternatives: What are the current state and alternative approaches, and which judgments already have evidence?
- Options and trade-offs: What feasible choices are there, and what are the benefits and costs of each?
- Resources and constraints: How will constraints such as time, team, technology, budget, and compliance affect the choice?
- Risk and validation: What are the most important uncertainties, and how can they be validated at low cost?
- Measurement and adjustment: What signals indicate effectiveness, and under what circumstances should you adjust or stop?
Discuss each gap that is most likely to change the recommendation, one by one. After each question, update the analysis based on the user's answer; subsequent questions should reflect the new information rather than continuing to follow a fixed checklist. The Agent may review existing materials or search for facts when authorized, and should not ask the user to look up information that is obtainable.
3. Provide conclusions and uncertainties
When the available information is sufficient to address the user's goal, summarize:
- the current recommendation and its applicable conditions;
- key reasons and main trade-offs;
- verified facts, user decisions, and assumptions still to be validated;
- important risks, and the conditions that could change the conclusion;
- optional next steps, specifying which ones require user decisions or separate authorization.
If the missing evidence could change the conclusion, do not pretend a definite answer has been reached. Give a conditional judgment and suggest an appropriate way to validate it or the corresponding PM skill.
End with the conclusion, applicable conditions, and open questions; ask further questions only when the missing information would change the conclusion. Any execution, internet access, file writing, or entry into other processes still requires the user's explicit authorization.
4. Stop and connect on demand
- Stop immediately when the user asks to stop; briefly explain what has been discussed and what remains unconfirmed, without automatically saving.
- If the user is not ready to decide, list that item as an open question; continue providing analysis that does not depend on that decision, and explain the scope that cannot be determined when necessary.
- When the user brings a chosen direction from
pm-brainstorm, reuse that direction and the existing information. After the analysis ends, the user decides whether to return to idea exploration, proceed to/pm-demandvalidation, or end; do not automatically invoke them in sequence. - Fact research, user research, market analysis, priority scoring, and formal documents are handled by the corresponding skills. The conversational judgments of this skill cannot replace those outputs or empirical conclusions.
In-conversation result format
Simplify or expand according to the complexity of the question; lightweight discussions are not required to use a template:
### 分析结论：{主题}
- **目标与范围**：...
- **建议**：...（适用条件）
- **主要取舍**：...
#### 依据与不确定性
- 已核实事实：...（来源、日期）
- 用户已确认：...
- 待验证假设 / 开放问题：...
- 主要风险：...
#### 可选下一步
- ...（由用户选择，不自动执行）
Only when the user requests saving should the target path and overwrite/append method be confirmed, and then the document generated.
Pre-completion checks
- The discussion scope matches the user's request and has not been forcibly expanded into a full product plan.
- Only one question related to the current analysis is asked per round, and information already provided has not been requested again.
- Recommendations include reasons and applicable conditions; inferences or suggestions have not been written as confirmed facts/decisions.
- Important evidence gaps and risks have been made explicit; unverified parts have not been falsely claimed as certain.
- Without user authorization, no files have been written, no other processes started, and no follow-up actions claimed to have been executed.