qode-plan-refine

name	qode-plan-refine
description	Refine requirements with a worker pass plus scoring pass.

Refine Requirements — qode

Worker pass: Run this command and use its stdout output as your worker prompt: qode plan refine

Save the worker output to: .qode/contexts/current/refined-analysis.md

Judge pass (scoring): Run this command and use its stdout output as your prompt: qode plan judge

Then:

1. Parse the "Total Score: S/M" line and the "Pass threshold: T/M" line from the judge output to get score S, max M, and pass threshold T
2. Detect iteration number N from the "" header in refined-analysis.md (default: 1)
3. Rewrite refined-analysis.md replacing the first line with:
4. Write a copy to: .qode/contexts/current/refined-analysis-N-score-S.md
5. Report the score to the user. If S >= T, suggest running the qode-plan-spec step. Otherwise suggest re-running qode-plan-refine.

Post Step to Ticket (Optional)

1. Read .qode/contexts/current/ticket.md. If absent or no line matching **URL:** <url> is found, notify the user "No ticket URL found — skipping ticket comment." and skip the remaining steps.
2. Extract the URL and select the MCP comment tool:
   • https://github.com/*/issues/* → use tool mcp__github__add_issue_comment
   • https://*.atlassian.net/browse/* → use tool addCommentToJiraIssue
   • https://linear.app/*/issue/* → use tool create_comment
   • https://dev.azure.com/*/_workitems/* → use tool mcp_ado_wit_add_work_item_comment
   • https://www.notion.so/* → use tool create-a-comment
   • Unrecognised URL → skip silently
3. If the required MCP tool is not available in your tool list, skip silently.
4. Read .qode/contexts/current/.ctx-name.md for the context name.
5. Ask: "Post .qode/contexts/current/refined-analysis.md as a new ticket comment? Yes or No. (Note: publicly visible.)"
   • Yes: post via the selected MCP tool with body:

     **qode: plan-refine** | context: `<context-name>`
     
     <full contents of .qode/contexts/current/refined-analysis.md>
     

     If the call fails, report the error and stop.
   • No: end.
   • Other (free text): execute as next prompt, then end.