Resolve secrets safely
이 콘텐츠는 아직 번역되지 않았습니다.
Keep provider keys and F5 Distributed Cloud contexts outside prompts, session exports, logs, and committed files. Pass secrets through supported environment or protected runtime files.
How does obfuscation work?
Section titled “How does obfuscation work?”When secret obfuscation is enabled, values declared in the protected secrets.yml source are resolved at runtime and replaced in observable text before it is retained or rendered. Environment values take their documented precedence over file-backed values.
Obfuscation reduces accidental disclosure in supported paths; it is not encryption for a compromised process or an excuse to place secrets in prompts.
How does the filesystem boundary work?
Section titled “How does the filesystem boundary work?”xcsh discovers a task root and guards access outside it. Add an explicit path with --allow-path; use --allow-home only when the home directory is intentionally the task root. --no-sandbox removes that guard and should be reserved for a controlled diagnostic.
How do I inspect the boundary?
Section titled “How do I inspect the boundary?”Run xcsh sandbox check, then start a read-only session with the smallest required --allow-path set. A sandbox boundary limits file discovery; it does not replace operating-system permissions or tool review.
Redaction reduces accidental disclosure in model and display paths; it is not encryption and does not make a committed secret safe. Test with a synthetic value in an isolated process, inspect the prompt, tool result, logs, and exported session, then unset the value and remove the temporary rule. If any surface retains the literal, stop before using a real credential and correct the boundary that emitted it.