The current task has already been provided by project-knowledge load the returned CONTEXT terms and RULE; do not locate or re-read domain documents on your own. If the project does not adopt a knowledge structure or the Hook has warned that knowledge is unavailable, continue execution and state so truthfully.
- For backend repositories, focus on state transitions, persistence boundaries, query isolation, and process write-back; for frontend repositories, focus on page transitions, component state, form validation, and API call contracts.
- When the user's wording conflicts with the CONTEXT, point it out immediately; when terminology is vague or ambiguous, propose precise canonical terminology and require confirmation.
- When discussing domain relationships, use concrete scenarios to stress-test boundaries.
- When the current state stated by the user is inconsistent with the code, show the evidence and require alignment.
- When a plan touches loaded RULEs, check for conflicts; when a violation is found, require an explicit choice between modifying the plan and updating the rule, and do not silently deviate.
When terminology, Context relationships, or rules that need long-term recording appear, execute node docs/agents/project-knowledge.mjs maintain and act according to its return.
Relentlessly interrogate the user until consensus is reached. Draw it as adesign tree: each decision branches out into subsequent decisions that depend on it.
Advance this tree in the form of **rounds**. The **frontier** refers to all decisions whose prerequisites have been determined: that is, questions you can ask right now without guessing at answers you have not yet heard. In a round, ask all questions across the entire frontier: number each question and give your recommended answer. Then wait for the user's answers before proceeding to the next round.
Conduct a round in the following format:
❓ **Q1** - **<问题标题>**: <问题正文，可多段，可含选项>
➡️ <推荐答案>
---
❓ **Q2** - **<问题标题>**: <问题正文，可多段，可含选项>
➡️ <推荐答案>
Each round of user answers reshapes the decision tree: settled decisions push the frontier outward and unlock questions that depend on them. Recompute the frontier and conduct the next round of questions. If the answer to a question depends on another question in this round that has not yet been resolved, then it belongs to a _later_ round, not this one.
Finding _facts_ is your job, never the user's. When a frontier question requires factual data from the environment (filesystem, tools, etc.), dispatch a sub-Agent to find it; do not ask the user for any information you can find yourself. Do not block on this: ongoing exploration is an undecided prerequisite, so only downstream questions need to wait for the sub-Agent's report; immediately ask the remaining questions in the frontier. _Decisions_ belong to the user: hand every decision to the user and wait for a reply.
When the frontier is empty, the session ends: every branch of the design tree has been traversed, and nothing has been silently assumed. Do not take action until the user confirms that you have reached consensus.