After the user names this skill and poses a question, provide conclusions and recommendations following the fixed playbook. Do not automatically enter it just because words like "analysis" are mentioned. No verbosity and no pretentious profundity.
Code change review belongs to review, and requirement convergence belongs to explore. This skill does neither of these; if the user wants one of them, stop and point to the corresponding skill.
Unified tool definition
交互式提问: the built-in tool for an Agent to ask the user questions and provide options; its name differs across Agents (e.g.,AskUserQuestion、AskQuestion). All questions this skill asks the user use it.
Read-only
Modifying code is prohibited. Except for the one report the user confirms to be written to disk, modifying any file is prohibited. Subagents are likewise read-only.
No question to analyze: stop and have the user provide the question.
Output form
If the user specified it when invoking, follow that. If not specified, use 交互式提问 to choose between "conclusions in the conversation" and "markdown report written to disk"; do not deliver before confirmation.
Choosing write-to-disk: the user provides the path (custom input allowed); if there is no path, do not write the file. The skill writes only that report file and does not modify other paths.
Both forms are delivered according to summary-template.md , with identical body content.
Workflow
Hard-coded to these three steps; do not switch to another playbook:
- Break into dimensions: break the current question into several analysis dimensions. Dimensions depend on the question; do not apply fixed categories.
- Dispatch multiple subagents in parallel to analyze separately: one subagent per dimension, started in parallel. The task description must state: analyze only that dimension, read-only, do not modify any file; return that dimension's facts and judgments, do not write the full text according to the template. No subagent tool: stop, and do not degrade by doing the dimensions yourself in the main conversation.
- Consolidate according to the template: after the subagents return, consolidate according to summary-template.md . The sections are fixed and only these five: Background / Method / Findings / Conclusions / Recommendations. All analyses share this one set; do not switch templates by question type or add branch sections.
Then deliver according to the confirmed output form.