| name | prompt-summary |
| description | Summarize the prompts used in the current conversation as a numbered list. Use when the user wants to see the prompt chain, review their prompts, or share the sequence of prompts that produced a result. |
Prompt Summary
Review the conversation history and produce a concise numbered list of the user prompts that drove this session.
Instructions
Step 1: Extract Prompts
Scan the full conversation history. Collect every user-submitted prompt in order. Exclude:
- System prompts and injected context
- Slash command invocations (e.g., /test, /pr) unless the user typed additional instructions with them
- Tool approval responses (yes/no clicks)
- Empty or whitespace-only messages
Step 2: Classify Each Prompt
Tag each prompt as one of:
- investigation -- directed research, exploration, or analysis
- implementation -- requested code changes, file creation, or builds
- refinement -- adjusted output format, wording, scope, or style
Step 3: Format the Output
Produce a numbered list. Each entry has the prompt text (quoted or paraphrased to stay concise) followed by a brief annotation of what it accomplished.
Format:
1. "the prompt text" -- what it accomplished
2. "the prompt text" -- what it accomplished
Rules:
- One line per prompt, no sub-bullets or headers
- Paraphrase long prompts to keep each line scannable (aim for under 120 characters for the prompt portion)
- Keep annotations to half a sentence
Step 4: Add a Summary Line
End with a single line summarizing prompt count by phase. Group adjacent prompts of the same type.
Example: "Three prompts for investigation, two for refinement."
Only count phases that actually appeared. Do not force all three categories.
Output
Print the numbered list and summary line directly -- no code fences, no headers, no preamble. The output should be immediately copy-pasteable into Slack or a document.
