sepia-recreate
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [NO_CODE]: The skill consists entirely of markdown instructions without any associated scripts, binaries, or configuration files that would allow for code execution.
- [INDIRECT_PROMPT_INJECTION]: The skill processes target content provided by the user for rewriting tasks. It mitigates injection risks by explicitly instructing the agent to treat this target as untrusted data rather than authoritative instructions. This follows the recommended evidence chain by establishing ingestion points while defining strict boundary logic.
- [COMMAND_EXECUTION]: While the skill references a local file path (
../sepia/SKILL.md), it explicitly restricts the search scope to a specific sibling directory and prohibits searching the home directory, current working directory, or global paths, which prevents arbitrary file access or path traversal attacks.
Audit Metadata