feature/Unsafe-fix-looks-safe · L-260926-8a069f

[lchoquel]

The human validation panel and the shared Markdown prose (the agent CLI, the remaining errors of fix, and the /validate rendered Markdown) printed an unsafe suggested fix exactly like a safe one, so nobody could tell that rename-model needs checking before it is applied. One shared label now reads 💡 Suggested fix (unsafe, confirm before applying): for an unsafe fix while a safe fix keeps 💡 Suggested fix:, and the validate, fix, agent CLI and error-model docs, with the FixSafety docstring, say how an unsafe fix is shown and that pipelex fix bundle never applies it.

Closes L-260926-8a069f

🤖 Generated with Claude Code

---

Summary by cubic

Unsafe suggested fixes now print a distinct label everywhere they appear, so a rename-model fix no longer looks identical to a safe one that pipelex fix bundle would apply on its own.

• The shared label reads 💡 Suggested fix (unsafe, confirm before applying): for unsafe fixes across the human validation panel, the agent CLI Markdown, the /validate rendered Markdown, and fix's remaining errors.
• The automatic-fix footer and count now include only safe fixes, since the command never applies unsafe ones.
• The validate, fix, agent CLI, and error-model docs plus the FixSafety docstring now describe how unsafe fixes are labeled and that pipelex fix bundle never applies them.

^{Written for commit 97bf0b8. Summary will update on new commits.}

[lchoquel]

Review, round 3 at profile 4, bar necessity (cubic, codex review, codex adversarial, code-review medium): clean, nothing fixed or deferred. Earlier rounds: round 1 at profile 2 was clean; round 2 at profile 4 fixed the FixSafety docstring and the validate.md footer wording in 97bf0b8.

[lchoquel]

Superseded by #1276, which carries this branch's commits and landed on dev as a7924f6.