This repository is an educational workshop, not a production agent service. The exercises can create billable cloud resources and execute code in a managed sandbox. Use only synthetic fixture data and a dedicated development API key with appropriate account-level controls.
Please do not open a public issue containing credentials, exploit details, or sensitive data. Use GitHub's Report a vulnerability / private vulnerability reporting feature if it is enabled for this repository. Otherwise, contact the repository maintainer privately through the contact information on their GitHub profile. Do not assume a guaranteed response time.
Include a minimal reproduction, affected files or versions, impact, and recommended mitigation if known. Never include live API keys or private user data.
- Keep
ANTHROPIC_API_KEYin.envor your environment; never commit.env, logs containing secrets, or session state. - Confirm
.env.examplecontains placeholders only. - Use the smallest tool permissions and network egress needed for the exercise.
- Use a session spend cap; check account-level limits independently.
- Review generated commands and files before executing or applying them locally.
- Delete unused sessions and uploaded files, and archive agents when appropriate.
- Treat agent-generated reports as untrusted data; verify against the provided fixture.
check.pyfollows this rule: it reproduces evidence only by running samples it parses from the localfixture/folder, and it never runs thecommandfields or any other code from the agent's report.check.py stretchreads the agent's corrected docs without running them, andreplay.pyonly prints recorded events.
Only the current default branch is maintained. This workshop is a learning artifact and does not promise security fixes or long-term support for earlier revisions.