codebase-advisor
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill defends against indirect injection by instructing the agent to treat all codebase content as data, specifically naming 'ignore previous instructions' as a pattern to be reported as a security finding rather than followed. This mitigates risks from malicious instructions embedded in audited source files.
- [DATA_EXFILTRATION]: Explicit rules and subagent instructions prevent the exfiltration of sensitive credentials by mandating redaction and referencing only file and line locations. This ensures that API keys or tokens found during an audit are not committed to plan files or GitHub issues.
- [COMMAND_EXECUTION]: Tooling is restricted to read-only operations on the primary codebase. Any proposed code changes are executed in isolated git worktrees with mandatory review by the advisor agent, ensuring the user's main branch remains protected until a manual merge is performed.
- [EXTERNAL_DOWNLOADS]: The use of the GitHub CLI for issue creation is protected by automated visibility checks and explicit user confirmation gates, which prevents the accidental public disclosure of sensitive security findings.
Audit Metadata