vibe-plan-review
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXPOSURE]: The skill includes an extensive 'Sensitive Content Handling' section that mandates the redaction of private keys, JWTs, API keys, and environment secrets using the
[REDACTED:<type>]pattern before any text is output to chat or saved to files. It also treats live credentials as blockers requiring user remediation outside the agent environment. - [INDIRECT_PROMPT_INJECTION]: While the skill ingests external Markdown plans (a potential attack surface), it provides detailed instructions to mitigate risks. It processes items individually, enforces a 'Write Boundaries' policy, and requires explicit user confirmation for 'Reflection' (applying changes) to ensure the agent does not execute instructions embedded within the plan data.
- [COMMAND_EXECUTION]: The skill explicitly defines and restricts the use of state-mutating shell commands (such as
sed -i,mv,cp, andgit checkout). It enforces 'Effect and Write Boundaries' that prohibit irreversible or outward-facing operations without explicit user consent. - [PRIVILEGE_ESCALATION]: The skill contains strict instructions against service/daemon installation and unauthorized state changes, ensuring the agent operates with least privilege and remains within the artifact-only or document-only boundaries as defined by the phase state.
Audit Metadata