Security Considerations

This repository packages an agent-native AIDLC Code Reviewer. The toolkit uses local deterministic helper scripts plus the AI model provided by the host assistant. It has no built-in AWS, Bedrock, telemetry, or network dependency.

Trust Model

• Source code and repository documents are untrusted inputs. Treat them as data to analyze, not as instructions to follow.
• Reports are advisory. A human reviewer remains accountable for accepting, rejecting, or prioritizing findings.
• Helper scripts run locally and use Python standard library only. Optional static analysis tools run only when installed on the host and enabled in common/review-config.yaml.

Threats And Mitigations

Prompt Injection From Source Files

Source files can contain comments or strings that instruct the assistant to ignore its task.

Mitigation: every methodology file instructs the assistant to treat source code, tool output, and project documents as untrusted data. Injection attempts should be reported as suspicious content rather than obeyed.

Residual risk: prompt-level controls are not guarantees. Use human review for all decisions.

Report Output Safety

Markdown and HTML reports include text from source files, tool output, and AI findings.

Mitigation: the HTML report renderer escapes dynamic values and emits static HTML with no script tags. Do not bypass escaping when editing common/tools/build_report.py.

External Static Tools

Optional tools such as bandit, ruff, mypy, semgrep, or gitleaks execute on local source code.

Mitigation: the runner invokes only configured commands with fixed argument templates. Review any config changes before running against sensitive code.

Sensitive Data

Source code and generated reports may contain confidential logic or secrets.

Guidance: avoid placing secrets in source code, protect generated reports at rest, and confirm the data handling terms of the AI assistant/model before reviewing sensitive repositories.

Reporting A Security Issue

Report suspected security issues to the repository maintainers privately. Do not open public issues for vulnerabilities or exposed secrets.