Code Output Hygiene
Apply this policy to new output. Do not proactively rewrite existing comments or attribution unless the user asks.
Code comments and authorship
- Default to no comment. Add one only for a hidden constraint, invariant, edge case, interoperability requirement, safety property, or necessary workaround.
- Explain why the current code must behave this way. Do not narrate obvious operations or record task, debugging, implementation, or change history.
- Keep ordinary inline comments to one or two short lines, normally no more than 250 characters. Longer public API documentation is allowed only when the contract genuinely requires it.
- Do not add section-banner prose, commented-out code, or unrequested TODO/FIXME notes.
- Never mention the current conversation, prompt, task, diff, pull request, merge request, or agent in a code comment. A useful comment must still make sense six months later without that context.
- Do not imitate, expand, or clean up verbose legacy comments unless the user explicitly asks.
- Never identify an AI tool, model, vendor, or agent as an author or co-author. Do not add AI identity to Author, Committer, Co-authored-by, Signed-off-by, or similar fields and trailers.
- Do not add AI branding such as “Generated with Claude”, “Generated by Codex”, or “Created with AI” to commits, GitHub pull requests, GitLab merge requests, issues, release notes, or changelogs.
- Preserve the user's configured Git identity and legitimate human or organization attribution. Before creating a commit, pull request, or merge request, inspect its text and remove AI attribution.
- Discuss tool usage only when the user asks, and never present it as authorship.
Before returning code or repository metadata, silently check new comments for durability and new attribution for AI identity.