Fix PR base retarget ordering after ancestry changes

[konard]

Summary

• Reconstruct the complete archcore-ai/plugin#24 solver run from authenticated logs, issue/PR timelines, all GitHub comment channels, and the fork CI record.
• Fix the identified Hive Mind defect by requiring agents to push new base ancestry and verify the remote head SHA before changing an existing pull request's base.
• Apply the invariant to all six prompt backends and all four localized prompt catalogs.
• Add a regression test plus a patch changeset.
• Preserve the full analysis and losslessly compressed 21,963-line run log under docs/case-studies/issue-2099/.

Root cause

The agent merged the unrelated dev history locally, then ran gh pr edit 26 --base dev before pushing that merge. GitHub therefore compared the still-remote main-rooted placeholder head against unrelated dev and closed/stranded PR 26. The replacement PR 27 was valid, but its workflow stopped at the expected public-fork approval gate with zero jobs; no tests failed.

Reproduction and regression

The archived timeline records:

1. placeholder PR 26 created from main;
2. dev merged only into the local feature branch;
3. PR base changed before the merged head was pushed;
4. PR 26 closed against its stale remote head;
5. replacement PR 27 created after the implementation was pushed.

tests/test-pr-base-retarget-order-2099.mjs builds every supported agent prompt and verifies the required ordering: push ancestry, verify the remote head SHA, then retarget.

Verification

• node tests/test-pr-base-retarget-order-2099.mjs
• npm run format:check
• npm run lint
• npm run check:duplication
• bash scripts/check-mjs-syntax.sh
• bash scripts/check-file-line-limits.sh
• npx -y node@24 scripts/run-tests.mjs --suite default — all 350 selected test files passed
• npx -y node@24 scripts/validate-changeset.mjs

Fixes #2099

[konard]

Working session summary

Implemented and finalized PR #2106.

Key results:

• Identified the root cause: PR [WIP] Create perfect readme for this project #26 was retargeted to unrelated dev before the merged ancestry was pushed.
• Added safe retarget-order guidance across all six agent backends and four locales.
• Added a regression test and patch changeset.
• Archived the complete investigation, GitHub records, and compressed 21,963-line authenticated log in the case study.
• All 350 local Node 24 test files passed.
• Fresh CI run 30168210884 passed against exact head 0e86b248.
• PR is mergeable, clean, and marked ready for review.
• Working tree is clean and the branch includes current main.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-5.6 Sol
• Provider: OpenAI
• Public pricing estimate: $22.386909

📊 Context and tokens usage:

• 407.0K / 200K (204%) input tokens, 32.4K / 128K (25%) output tokens

Total: (407.0K + 16.9M cached) input tokens, 32.4K output tokens, $22.386909 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-5.6-sol
• Thinking level: low (~7999 tokens)
• Model: GPT 5.6 Sol (gpt-5.6-sol)

📎 Log file uploaded as Gist (7609KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag