Derived from the uploaded record
InstaShare scans the structured events to count meaningful user and assistant turns and tool-use blocks. The earliest and latest relevant timestamps produce a session duration when both are available. These values describe the captured transcript, not CPU time, billing, or uninterrupted active work.
The title comes from session metadata when present and falls back to useful transcript content. A slug-safe version is appended to the short random identifier, creating a URL that is easier to recognize while preserving the identifier used for lookup.
Some context exists only at runtime
The persisted Claude Code record may contain a base model identifier, but it does not always preserve the full display label, context-window marker, or reasoning effort shown in the live interface. The running agent can pass those optional values in request headers. The plan label may be read from local Claude account configuration.
Missing values degrade gracefully. InstaShare does not invent an effort level or plan, and the viewer can still show transcript-derived counts without those headers.
Metadata refreshes with a stable-link update
When the same session is shared again, counts, timestamps, duration, title, and available session metadata are recomputed. Header-less updates do not intentionally wipe earlier effort or plan values that cannot be reconstructed from the JSONL.