narrative-quality-auditor
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute
git rev-parse --show-toplevelduring initialization to resolve theAARON_SKILLS_ROOTvariable. This is a routine operation for tools designed to work within a Git repository structure. - [DYNAMIC_EXECUTION]: The workflow mentions the use of a local script,
validate-audit-artifact.py, to verify the integrity of audit results before they are persisted to the filesystem. This represents a functional security control for ensuring output validity. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it processes external brand narratives, claims, and "surface" sets which could contain adversarial instructions.
- Ingestion points: Untrusted brand canon, positioning content, and message experiments provided by the user for auditing (SKILL.md).
- Boundary markers: The instructions mandate the use of the "auditor-runbook's exact typed conversation header" to delineate the auditor's output from other content.
- Capability inventory: The agent has capabilities to execute shell commands (git) and perform file-write operations to the
memory/audits/path (SKILL.md). - Sanitization: The skill incorporates a validation step using
validate-audit-artifact.pyto check the generated audit artifact against a predefined schema before finalizing the write operation.
Audit Metadata