Cloud workflows
Turn findings into a work plan.
The optional AI assistant uses Claude to turn a completed Cloud report into up to five ordered actions. Each action explains its priority, suggests next steps, describes verification, and links to the underlying findings.
AI work plans require an active Cloud project licence, service availability, and an enabled team preference (on by default). They supplement scanner evidence; they do not execute changes or certify a release.
Create your first plan
- Open Team → AI work plans and review the data-sharing description. AI work plans are on by default; an owner or administrator can disable them.
- Open a completed report with supported open findings and select Create AI work plan.
- Wait for queued processing. The report page checks for completion automatically; you can leave and return later.
- Review the proposed actions against the actual code, configuration, and full report. Use the finding links to inspect scanner evidence.
A plan stays attached to the report it was created from. It is not automatically regenerated after another scan. Saved plans remain readable even when AI is disabled or the daily allowance has been used, until their report is removed.
One new plan per team per day
The allowance is shared across every project and member of a team, including Solo workspaces. It resets at 00:00 UTC. A request belongs to the UTC date on which it was submitted. Adding members or deleting a report does not restore a used allowance.
Only one request can be in progress per team. Successful plans use the daily allowance; failed requests do not. Reading a saved plan is free of this limit. A report with a saved plan cannot be regenerated.
Requests are also subject to rate limits and a shared service budget. A failed request can still use provider resources, so repeated failures may exhaust the service budget without consuming the team’s daily allowance.
What is sent to Anthropic?
Only a bounded list of up to 20 supported open findings is sent, prioritized by the fixed check severity and whether the finding is new:
- internal finding IDs;
- fixed check names, descriptions, category, severity, estimates, and guidance from LaraBeacon’s own check catalogue;
- the technical status (failed or error), the new-finding flag, and counts of open and omitted findings.
Project names, repository identifiers, branches, commits, URLs, user details, file paths, source snippets, exception messages, stack traces, diagnostic details, and report-supplied guidance are excluded. Enabling source snippets for Cloud reports does not enable sending them to Claude. Custom analyzers outside the catalogue are omitted.
Disabling the team preference prevents new requests; queued requests are checked again before processing. A request already sent to Anthropic cannot be recalled. Anthropic processes submitted data under its applicable API terms and retention settings. See the privacy information.
Understand the limits
Claude receives no source code or runtime values. It can propose a check-level remediation sequence, but cannot prove exploitability, diagnose the exact cause, or verify a fix. An errored analyzer is an incomplete check, not a confirmed vulnerability.
Large reports and unsupported checks may be omitted; the saved plan shows that count. Always inspect the full report. AI suggestions never alter scanner results, severities, finding statuses, baselines, or CI exit codes.