Youish is a local Codex skill, but writing samples are personal. Treat drafts, voice samples, and live-eval transcripts as sensitive.

• API keys, access tokens, passwords, cookies, or credentials.
• Customer data, private drafts, legal or medical details, HR material, or internal company facts.
• Live-eval transcripts, local profile cards, ledger files, or voice samples that include private text.

Use redacted examples when opening issues.

Prefer file-based inputs for private drafts when using local lab tools:

python3 scripts/failure_fixture.py \
  --case-id redacted_example_01 \
  --source-file source.local.md \
  --failed-rewrite-file failed.local.md \
  --desired-rewrite-file desired.local.md

Inline --source, --failed-rewrite, --rewrite, and --desired-rewrite values can land in shell history. Use *.local.md / *.local.json files for private material, and only paste tool-redacted output into public issues.

scripts/live_eval.py sends selected fixture text to the configured API endpoint only when OPENAI_API_KEY is set. It refuses custom API URLs unless --allow-custom-api-url is passed, and saved transcripts must end in .local.jsonl. Transcript saves are hash-only by default; use --save-raw-source or --save-raw-output only for public-safe text.

--replay-jsonl does not call the API, but it requires raw output records for deterministic replay. Raw outputs may contain private source text, so only save replayable transcripts locally.

Keep *.local.jsonl, *.local.md, and *.local.json files local. They are debugging artifacts, not release artifacts.

For public-safe bugs, open an issue. For anything involving leaked secrets, private drafts, unsafe transcript handling, or token exposure, contact the maintainer privately before posting details publicly.

If you accidentally commit sensitive text to a fork or branch, rotate affected credentials and remove the content from public history before linking it in an issue.