Keep the goal narrow
Share one task-specific artifact: Preserve stack traces, timestamps, and long diagnostic lines.
A diagnostic paste should capture the smallest useful time window around a failure. The goal is to preserve evidence - Timestamps, error types, correlation, and stack structure - While removing credentials and personal data.
Use these three checks before you turn share logs into a link for another person.
Share one task-specific artifact: Preserve stack traces, timestamps, and long diagnostic lines.
Recommended first step: Choose plain text or the closest log syntax
Before upload: Review the share logs checklist below to replace live values while preserving the context a recipient needs.
Share logs tip: How to share error logs safely adds this format-specific context: Select the causal window and redact diagnostic data systematically.
A well-scoped share logs artifact preserves the task-specific context below and can expire when that task ends.
Include the complete error and a short lead-up window; give recurring hosts and users consistent redaction labels so relationships remain visible.
Keep the failing command, toolchain version, first causal error, and nearby output instead of uploading an entire repetitive build log.
Use the paste as a temporary conversation artifact while the approved logging and incident platforms remain the system of record.
Search for authorization headers, cookies, tokens, connection strings, email addresses, customer identifiers, internal hostnames, and signed URLs before uploading.
For share logs, a made-up but representative example is safer than a production export. Expiration can end future access to the source link, but it cannot recall a recipient’s copy.
Compare privacy controls →For a private share logs artifact, encrypted mode applies AES-GCM in the browser before upload. Its key stays after # in the share URL and outside the normal server request.
Understand client-side encryption →Practical answers about the access, retention, and context choices for share logs.
Include enough events before and after the failure to establish cause, but remove unrelated hours of repetitive output.
No. Pattern detection can miss contextual data and unknown credential formats, so a human redaction pass is still required.