Objective
Help the user determine what the most real bottleneck currently is, and confirm which problem should be prioritized in the next cycle.
Core Principles
- By default, read and write the
opc-doc/ - In teaching mode, first explain that a review is not a running account, but finding the bottleneck
- By default, ask only one question at a time; if several questions are all very light and closely related to each other, they can be combined into 2 to 3
- By default, give 3 bottleneck hypotheses or priority focuses, and additionally
4. 我有自己的方案 - Write the formal results only after the user confirms
- Do not directly give recommended conclusions; only do solution analysis
What this step must complete
- Summary of the current business state
- Identification of the main bottleneck
- Confirmation of the priority focus for the next cycle
Priority confirmation order
- The most stuck place recently
- The most likely bottleneck type
- Which point to prioritize fixing in the next cycle
Completion Criteria
- 3 bottleneck hypotheses or priority focuses have been formed
- The user has confirmed the most real current bottleneck
What needs to be explained in this step
In teaching mode, first explain:
- A review looks at demand, conversion, assets, and risk
- The goal is not to prove that you worked hard, but to find the one point most worth correcting
- This step will determine the sole focus of the next cycle
Trigger Conditions
This skill is not the linear next step of the setup phase, but a periodic tool in the operations loop.
Trigger timing (meeting any one is sufficient):
- After operating for a while, you feel stuck and do not know where the problem is
- You want to do a structured review rather than continue by feel
- Conversion, retention, or closing shows an obvious decline
- The user independently chooses to do a review
Prerequisite check: If the user has just completed the setup phase and has not started execution, do not enter this skill - a review requires actual operational data, even if lightweight.
Input
Prioritize reading:
opc-doc/outputs/**(all setup phase outputs, for comparison)opc-doc/reviews/dashboard.json(historical review records, if they exist)opc-doc/outputs/09-dashboard-review/(previous reviews in this cycle, if any)
If opc-doc/ does not exist or the data is insufficient, first check whether there is already enough business information in the current conversation.
- If the user has not operated at all, do not do a fake review; remind them to go execute first and then come back for a review
Execution Steps
- Explain the objective of this step
- By default, ask only one question at a time; if several questions are all very light and closely related to each other, they can be combined into 2 to 3, for example:
- What step have you been most stuck on recently?
- Is it that no one comes, no one buys, or that you do not reuse anything after finishing?
- After each round of answers, give brief feedback
- Generate 3 bottleneck judgments, for example:
- Insufficient demand validation
- Conversion path not smooth
- Asset accumulation too weak
- Explain the applicable situation, basis, and cost of each judgment
- By default, add
4. 我有自己的方案 - Let the user confirm, combine, modify, or directly propose their own version
- After the user confirms, then write the formal review results
Output
The dialogue layer must include:
- Explanation of this step
- Summary of current business status
- 3 bottleneck judgments +
4. 我有自己的方案 - the applicable situation, basis, and cost of each judgment
- Ask the user to confirm or modify
Persistence checkpoint (must be completed after each trigger)
After the user explicitly confirms this cycle's core bottleneck and next focus,immediatelyuse the Write tool to write the file. Describing conclusions in the conversation does not count as persistence.
This skill supports multiple triggers. Each persistence uses a date-bearing filename and does not overwrite historical records.
Write the file (using the current date, format YYYYMMDD):
opc-doc/outputs/09-dashboard-review/review-[YYYYMMDD].md(business status summary + bottleneck judgments + the single priority focus for the next cycle)opc-doc/reviews/dashboard.json(structured review data,append, retain history, do not overwrite)
Update the status file:
opc-doc/state/current-stage.json(write:{"stage": "09-dashboard-review", "status": "completed", "next_stage": "execution-or-回退阶段", "summary": "一句话核心瓶颈和下一重点"}）opc-doc/state/decisions.json(append the single priority focus for the next cycle, noting the date)
After persistence is complete, inform the user in the conversation:
"✅ This review has been saved. Based on the review conclusions: [briefly explain whether the next step is to continue execution, roll back a certain stage, or trigger asset accumulation]"
When to call other skills
- If the review finds gaps in previous stages, clearly recommend rolling back to the corresponding skill
- Do not make the user repeat the entire process without justification
Exception handling
- If the data is incomplete, only do a lightweight qualitative review and do not pretend to have complete conclusions
- Do not give multiple next-step focuses at the same time; maintain a single priority