image-editing
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill consists of instructional markdown and a structured YAML tracking file. It defines a workflow for image editing and revision management without using executable scripts, remote downloads, or dangerous shell commands.
- [CREDENTIALS_UNSAFE]: The skill includes explicit instructions for the agent to avoid placing credentials in prompts, source files, or output logs, which adheres to secure configuration management practices.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from a local file (references/edit-record.md) which contains user-provided instructions for image changes.
- Ingestion points: The agent reads the 'requested_change' and 'exact_text' fields from the YAML block in references/edit-record.md.
- Boundary markers: The data is structured within a YAML block, though no explicit 'ignore instructions' delimiters are provided for the text fields.
- Capability inventory: The skill requires access to an external image-editing tool or API and involves reading/writing files in the references and outputs directories.
- Sanitization: There is no explicit text sanitization mentioned, but the data is used to inform image editing tasks rather than code execution.
Audit Metadata