loomloom is in beta and under active development. Security fixes are applied to the latest release on the main branch. Please use the most recent version before reporting an issue.

Please do not report security vulnerabilities through public GitHub issues.

Report security issues privately using one of the following channels:

• GitHub — open a private security advisory
• Email — send details to engineering@cogfoundry.ai

Please include:

• A description of the vulnerability and its potential impact
• Steps to reproduce or a proof of concept
• Affected version(s) and environment
• Any suggested fix, if available

Do not include real access tokens, API keys, or other secrets in your report.

• We aim to acknowledge security reports within a few business days.
• We will keep you informed during investigation and remediation.
• Please allow reasonable time for fixes before public disclosure.
• We will credit security researchers for valid reports when requested.

loomloom executes AI work through a remote execution platform. Follow these guidelines to keep your credentials, data, and operations secure.

• Only send LOOMLOOM_TOKEN to the configured LOOMLOOM_SERVER or the server explicitly provided with --server.
• Always use HTTPS.
• Never share access tokens publicly or store them in insecure locations.

The default production endpoints are:

Never send credentials to an unexpected host. Avoid following redirects that change the destination server.

• Workbook contents are transmitted as Base64-encoded data inside JSON requests.
• Avoid logging or printing complete request payloads.
• Treat returned accessUrl values as temporary signed URLs.
• Do not store temporary URLs in long-term logs or share them publicly.

AI agents should always obtain explicit user approval before:

• Performing paid operations
• Modifying remote state
• Publishing or unpublishing resources
• Submitting files or running remote workloads

Examples include:

• submit-file
• run execute
• template-spec run
• market run
• listing publish
• listing unlist

Do not blindly retry paid or state-changing operations after an ambiguous failure.

Before retrying:

• Check the current run, listing, review, or usage state.
• Reuse --client-request-id only when retrying the exact same request.
• Use a new request ID when the payload or operation changes.