Add rejected recommendations tracking to prevent re-flagging

GPT reviewer now skips anything listed in a cumulative
"Rejected Recommendations" section at the bottom of plan files.
Claude appends rejections there during integration, building a
negative cache across iterations.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

plan-refine

@@ -8,7 +8,9 @@ set -euo pipefail
 # Prompts (inlined)
 # ──────────────────────────────────────────────
 
-EVAL_PROMPT="Carefully review this entire plan for me and come up with your best revisions in terms of better architecture, new features, changed features, etc. to make it better, more robust/reliable, more performant, more compelling/useful, etc. For each proposed change, give me your detailed analysis and rationale/justification for why it would make the project better along with the git-diff style change versus the original plan shown below."
+EVAL_PROMPT="Carefully review this entire plan for me and come up with your best revisions in terms of better architecture, new features, changed features, etc. to make it better, more robust/reliable, more performant, more compelling/useful, etc. For each proposed change, give me your detailed analysis and rationale/justification for why it would make the project better along with the git-diff style change versus the original plan shown below.
+
+IMPORTANT: If the plan contains a '## Rejected Recommendations' section at the bottom, these are suggestions that were already considered and deliberately rejected in previous iterations. Do NOT re-propose anything listed there. Focus your energy on genuinely new improvements instead."
 
 INTEGRATE_PROMPT="I asked ChatGPT to review your plan. I want you to REALLY carefully analyze their plan with an open mind and be intellectually honest about what they did that's better than your plan. Then I want you to come up with the best possible revisions to your plan (you should simply update your existing document for your original plan with the revisions) that artfully and skillfully blends the \"best of all worlds\" to create a true, ultimate, superior hybrid version of the plan that best achieves our stated goals and will work the best in real-world practice to solve the problems we are facing and our overarching goals while ensuring the extreme success of the enterprise as best as possible; you should provide me with a complete series of git-diff style changes to your original plan to turn it into the new, enhanced, much longer and detailed plan that integrates the best of all the plans with every good idea included."
 
@@ -422,7 +424,8 @@ ChatGPT's feedback is at: ${FEEDBACK_FILE}
 You have Read, Edit, and the Morph edit_file MCP tool. Prefer the Morph edit_file tool (mcp__morph-fast-tools__edit_file) for edits — it's faster and handles large changes better than legacy Edit. Edit the plan file directly — do NOT output the plan content to stdout.
 - Preserve the YAML frontmatter block (between the --- delimiters) at the top unchanged.
 - Only modify the content below the frontmatter.
-- Print a brief summary to stdout as you work: which changes you're accepting, rejecting, or modifying, and why. Keep it concise — bullet points, not essays."
+- Print a brief summary to stdout as you work: which changes you're accepting, rejecting, or modifying, and why. Keep it concise — bullet points, not essays.
+- CRITICAL: For any recommendations you REJECT, append them to a '## Rejected Recommendations' section at the very bottom of the plan file (create it if it doesn't exist, or append to it if it does). Each entry should be a bullet with a brief reason for rejection. This prevents future reviewers from re-proposing the same changes. Format: '- **<recommendation summary>** — rejected because <reason>'. Keep the section cumulative across iterations — never remove previous entries."
 
         local CLAUDE_STREAM_ARGS=(claude -p "$CLAUDE_PROMPT"
             --allowedTools "Read,Edit,Write,mcp__morph-fast-tools__edit_file,mcp__morph-fast-tools__warpgrep_codebase_search"