History, receipts, and repeat runs

Filter by status or mode, then open a run to see its terminal timeline. The per-run amount is an output-based model estimate; Settings shows the settled managed allowance, prepaid balance, and active reservations.

Retry continues an interrupted run from its checkpoint. Inspect exact run version opens the recorded workflow and receipt without changing the current Flow. Neither silently reuses an expired approval for a write.

Open a Run Receipt and choose Improve this run. When normalized proof supports one option, Codelit preselects it and explains the exact typed evidence behind it. Otherwise the receipt makes no recommendation. Review the side-by-side checks, replay evidence, cost, latency, permissions, and evidence quality, then choose Use this version only when the bounded comparison passes.

After that candidate completes a real run, Keep as my default pins the exact successful workflow snapshot while preserving the verified one-step rollback.

• A suggestion can come only from an ungated external action, an ungated medium- or high-risk tool used by the run, or a receipt with no declared outcome check. Output text and model inference are never used.
• The candidate stays unproven until you run it.
• Supported older receipts receive only safe additive defaults. Unsupported, malformed, incomplete, or execution-truncated proof cannot create a suggestion, comparison, or promotion.
• Codelit rebuilds the baseline from the selected receipt and rejects a baseline from any other run.
• A missing or ineligible receipt explains what could not be verified and offers Browse proven runs, Open Agent Teams, or Remix and review without recording a misleading improvement event.
• Undo improvement restores the exact sanitized receipt workflow only when no newer edits would be lost.
• The current session can link back to the public receipt, but the persisted improvement draft keeps only the sanitized rollback workflow and version hashes, not the receipt ID or URL.
• Suggestion telemetry keeps only its fixed reason and improvement type. Useful or Needs work keeps only its fixed rating, run mode, and improvement type. Neither sends the receipt ID, result, private input, raw evidence, or secrets.

A completed Run with my key, local, or managed run offers Do this job again. Codelit rechecks the current runtime, tools, delivery target, approval gates, and saved spend ceiling, asks for fresh run inputs, then starts the original mode and creates a new linked receipt.

• The save-state stays in this browser.
• It stores the input shape, workflow reference, caps, approvals, and bounded receipt reference, not credentials, prior input values, raw evidence, or outputs.
• If the safety boundary changed, Codelit stops and tells you what to review instead of starting the job.

Clearing finished tasks removes them from the notification feed permanently while preserving durable run history. Delete a receipt separately when you no longer need its stored output.

A public Run Receipt is a reviewed, revocable proof link. Completed receipts also offer a social proof image made from the same bounded public title and summary plus status, mode, step count, duration, and canonical path. Halted receipts and private step content cannot create that asset. Adding a receipt to Outcome Gallery remains a separate opt-in with a stricter outcome-only projection; removing a Gallery entry keeps the receipt active, while revoking the receipt removes both discovery and Remix access.